Lexical organization in typically developing children and children with Down syndrome: a structural network analysis

Lexical network analysis provides a powerful framework for examining how words are organized in the mind, as well as the semantic, categorical, and associative links that structure the mental lexicon. This study investigates the organization, structural properties, and types of lexical relations present in individual lexical networks of typically developing children and children with Down syndrome. Lexical relations were examined using network-based structural analyses derived from co-occurrence patterns, and measures of density, centrality, modularity, hubs, semantic clustering, and word-type distributions were computed, together with the analysis of taxonomic and thematic relations. The results reveal significant differences between the two populations, despite comparable vocabulary sizes. These differences are discussed in terms of variation in the mechanisms underlying lexical acquisition and organization, as well as the role of experience in psycholinguistic development. Overall, this study provides empirical evidence on early lexical organization in typical development and Down syndrome, with implications for theories of language development and for educational and therapeutic approaches aimed at strengthening vocabulary.

Early vocabulary development constitutes a core process in language acquisition and in the formation of cognitive development. Understanding how the lexicon is organized in children's minds makes it possible to examine the interaction between fundamental cognitive processes—such as attention, memory, categorization, and abstraction—and the linguistic experience that shapes early learning. From this perspective, lexical organization is not limited to the accumulation of words, but rather involves the construction of a network of conceptual relations that evolves as children acquire increasing semantic and associative knowledge.

Research on semantic memory suggests that lexical knowledge is organized through relational structures connecting concepts according to similarity, association, and experience. From this perspective, semantic organization reflects both conceptual similarity and contextual regularities in language exposure. Developmental approaches to categorization have also emphasized the interaction between language and conceptual structure. Accordingly, development involves domain-specific representational changes that reorganize children's understanding of categories over time (Carey, 1985). These perspectives suggest that semantic development cannot be reduced to vocabulary growth alone, but instead reflects broader conceptual and experiential structure.

Distributional theories propose that semantic relations emerge from regularities in word co-occurrence across communicative contexts, such that semantic similarity can be inferred from patterns of linguistic distribution and usage (Landauer and Dumais, 1997; Contreras Kallens and Christiansen, 2025). From this perspective, children progressively extract semantic structure from the statistical properties of linguistic input, such that words that systematically appear together become increasingly integrated within the developing semantic system (Landauer and Dumais, 1997; Hills et al., 2010). Accordingly, co-occurrence-based networks are not understood merely as representations of superficial statistical patterns, but as approximations of the experiential and distributional structure through which lexical-semantic knowledge develops. Consistent with this view, previous research has shown that distributional semantic models capture meaningful aspects of early lexical organization and successfully predict patterns of vocabulary acquisition and semantic growth in children (Li et al., 2004; Hills et al., 2010). In parallel, network science approaches have shown that developing lexicons exhibit non-random structural properties such as clustering, hubs, and small-world architecture, reflecting self-organizing processes in semantic growth (Hills et al., 2009; Carlson et al., 2014). These models provide a framework for understanding how differences in cognitive processing and experience may lead to distinct trajectories of lexical-semantic organization across developmental populations.

Lexical networks have become a central conceptual and methodological tool for representing the mental lexicon as a dynamic system composed of interconnected nodes. Such networks reveal fundamental properties of lexical structure, including connectivity, clustering, hierarchy, and modularity (Steyvers and Tenenbaum, 2005). Examining these properties provides a structural framework that has been linked to how words may be activated and retrieved during language processing (Hills et al., 2010; Siew, 2019) and how specific network configurations may facilitate the acquisition of new vocabulary. Accordingly, lexical development is not viewed as a linear process, but as an organized expansion that follows the pathways of the existing semantic network, where highly connected words facilitate the integration of new lexical items, giving rise to small-world–like properties in semantic structure (Hills et al., 2010; Steyvers and Tenenbaum, 2005; Watts and Strogatz, 1998).

Network approaches are also compatible with spreading activation models of semantic memory, which propose that lexical access emerges through the propagation of activation across interconnected semantic representations. Within this framework, structural properties such as connectivity, clustering, and hub organization may influence the efficiency and direction of lexical activation and retrieval processes (Collins and Loftus, 1975; Siew, 2019).

Moreover, lexical expansion is shaped not only by the density or centrality of particular nodes, but also by the nature of the semantic relations that connect them. From a developmental perspective, taxonomic and thematic relations also have been consistently identified as foundational organizing principles of the lexicon. Taxonomic relations are based on shared features and category membership, supporting hierarchical organization, whereas thematic relations are grounded in functional, contextual, or event-based co-occurrence between concepts. This distinction is not merely descriptive but reflects different mechanisms of semantic organization and learning. For example, it has been shown that, during early stages of development, children tend to rely primarily on thematic relations, which are closely linked to concrete experience and the functional use of objects (Lucariello et al., 1992). Over time, taxonomic relations become increasingly prominent, reflecting a shift toward more abstract and hierarchically organized conceptual structures (Waxman and Markow, 1995; Arias-Trejo and Plunkett, 2013).

In neurodevelopmental conditions such as Down syndrome (DS), this linguistic structuring may differ substantially. The linguistic profile associated with Down syndrome is characterized by relatively preserved comprehension alongside marked difficulties in expressive language and morphosyntax (Chapman, 1997; Abbeduto et al., 2007). These differences have been linked to variation in memory, processing speed, and attentional capacities, all of which directly influence the consolidation and structuring of vocabulary (Vicari et al., 2000; Chapman and Hesketh, 2000).

From a network-based perspective, the study of the lexicon in Down syndrome provides a valuable opportunity to examine alternative trajectories of language acquisition. Previous research has demonstrated that children with Down syndrome are sensitive to basic semantic associations between words (e.g., Barrón-Martínez et al., 2022), suggesting that their lexical networks are organized, at least in part, around experiential and contextual links. More broadly, network-based approaches to early lexical development have shown that the structure of children's vocabularies exhibits systematic properties —such as clustering, small-world structure, and the presence of highly connected nodes— that emerge as a function of lexical growth (Hills et al., 2010; Beckage et al., 2011). Importantly, these studies have also demonstrated that differences in vocabulary composition can produce substantial differences in network structure, even when vocabulary size is comparable, highlighting the relevance of examining qualitative variation across developmental populations.

In particular, the work by Beckage et al. (2011) provides important evidence regarding the relationship between lexical structure and atypical language development in late talkers, revealing potential cognitive constraints. Using semantic network analysis derived from early productive vocabularies, the study compared typically developing children and late talkers matched on vocabulary size. This distinction is theoretically relevant because network-based models predict that lexical growth depends not only on vocabulary quantity, but also on how semantic relations become internally structured during development. Findings revealed that, despite producing a similar number of words, late talkers exhibited lexical networks with lower connectivity and reduced small-world structure, suggesting reduced network integration. The study showed that delayed lexical growth may not simply reflect slower vocabulary accumulation, but also differences in the structural organization of semantic relations within the developing lexicon. Furthermore, they argued that lexical acquisition is strongly influenced by the existence of highly connected words within the semantic network, with highly connected words facilitating subsequent vocabulary expansion. These findings support the idea that qualitative properties of lexical organization may play a critical role in developmental language outcomes.

Related evidence from network-based studies has also emerged in other neurodevelopmental and atypical populations. Research using semantic network approaches in autism spectrum disorder has shown that individuals may exhibit smaller or differently organized semantic clusters despite relatively preserved associative strength, supporting the view that atypical development may involve qualitative alterations in lexical organization (Ehlen et al., 2020). Although this literature remains limited, these findings collectively suggest that neurodevelopmental conditions may involve distinct trajectories of lexical-semantic self-organization, affecting both the global architecture of lexical networks and the balance between thematic and taxonomic relations.

Despite these advances, most previous work has focused on typical development or on comparisons between typical and late talkers, leaving open the question of how lexical networks are structured in neurodevelopmental conditions such as Down syndrome. In particular, empirical evidence directly comparing the internal structure of individual lexical networks in typically developing children and children with Down syndrome remains limited, especially studies that integrate global structural metrics with a systematic analysis of the semantic nature of the links connecting words.

The present study addresses this gap by conducting a comparative analysis of individual lexical networks matched for vocabulary size, combining network-based structural measures with a detailed classification of lexical relation types. By holding vocabulary size constant, this approach makes it possible to isolate qualitative differences in lexical organization that are not driven by the number of words produced, but by the structural and semantic principles that shape the mental lexicon in each population. In doing so, the study extends prior network-based research by incorporating both topological properties and the semantic content of lexical connections, providing a more comprehensive account of early lexical organization across developmental trajectories.

Based on this framework, the present study addresses the following research questions:

How do children with typical development (TD) and children with Down syndrome (DS) differ in the structural properties of their lexical networks when matched for vocabulary size?

What differences exist in the types of lexical relations (taxonomic vs. thematic) that organize the lexicon in each group?

What differences exist in the types of lexical relations (living vs. non-living items) that organize the lexicon in each group?

According to network-based models of lexical development, which propose that semantic organization emerges through the interaction between linguistic input, cognitive constraints, and the progressive structuring of lexical relations (Steyvers and Tenenbaum, 2005; Hills et al., 2010; Siew, 2019), the present study hypothesized that children with Down syndrome would exhibit patterns of lexical organization that differ from those observed in typically developing children matched for expressive vocabulary size.

Given previous findings showing that atypical language development may be associated with less integrated semantic networks and differences in connectivity, clustering and semantic relation patterns (Beckage et al., 2011; Ehlen et al., 2020), it was expected that children with Down syndrome would exhibit qualitatively distinct lexical network configurations. Additionally, based on developmental evidence suggesting a stronger reliance on contextual and experience-based learning in Down syndrome, it was expected that thematic relations would play a more prominent role than taxonomic relations in the organization of the lexicon in this group.

Finally, the study explored whether differences in lexical content and semantic network structure could reflect distinct developmental trajectories in the emergence and organization of the mental lexicon.

Participants

Given the substantial variability in language development characteristic of children with Down syndrome, participants were selected based on the comparability of expressive vocabulary size rather than chronological age. Previous research has shown that children with Down syndrome display marked heterogeneity in lexical acquisition and expressive language abilities, even among individuals of similar chronological age, making vocabulary size a more reliable indicator of lexical development for comparative analyses (Fenson et al., 2007; Martin et al., 2009). Consistent with network-based studies comparing typical and atypical lexical development (e.g., Beckage et al., 2011), the present study aimed to examine differences in the structural organization of lexical networks while controlling for vocabulary size, thereby minimizing the confounding effect of differences in lexical quantity.

The initial sample consisted of 83 children [41 with Down syndrome, M = 47.1 months, and 42 with typical development (TD) M = 29.2 months], whose data were obtained through parental reports and previously collected language records. In the DS group, the clinical diagnosis was reported and confirmed by parents or legal guardians. All participants were monolingual Mexican Spanish-speaking children. Participation was voluntary, and written informed consent was obtained from the children's legal guardians.

From this initial sample, a first sub-sample of children with Down syndrome (n = 41) was used for comparisons between real and randomized lexical networks. Inclusion in this sub-sample required complete lexical data and sufficient overlap with the corpus-based co-occurrence matrix, ensuring that reliable individual lexical networks could be constructed from the available vocabulary data.

The final sub-sample used for direct group comparisons consisted of 20 children: 10 with typical development (aged between 19 and 58 months; M = 41.5, SD = 12.76) and 10 children with Down syndrome (mental age ranging from 19 to 32 months; M = 22.1, SD = 4.46). Given the high interindividual variability observed in the DS population, a stricter matching procedure was implemented to ensure comparability between individual lexical networks. Specifically, each child with Down syndrome was paired with a typically developing child with a closely comparable expressive vocabulary size, allowing a maximum difference of ±8 words between pairs. This procedure reduced the overall sample size but increased the internal validity of the comparisons by enabling one-to-one analyses of lexical network structure under equivalent lexical conditions. Matching participants on vocabulary size is particularly important in network analyses because network properties such as connectivity, clustering, and hub formation are strongly influenced by the number and composition of lexical items included in the network. Thus, controlling for vocabulary size allowed the analyses to focus on qualitative differences in lexical organization rather than differences attributable merely to vocabulary quantity.

By implementing pairwise matching procedures, the present study sought to examine whether children with Down syndrome and typically developing children organize lexical-semantic information differently despite having comparable productive vocabularies.

Inclusion and exclusion criteria focused on the absence of additional neurodevelopmental conditions, hearing impairments, or neurological conditions other than Down syndrome. This procedure ensured comparable levels of lexical development between groups for the analysis of lexical networks.

Ethics statement

The study was conducted in accordance with the ethical standards of the institutional research committee and with the 1964 Helsinki declaration and its later amendments. Informed consent was obtained from all caregivers.

Materials

Children's expressive vocabularies were assessed using the MacArthur–Bates Communicative Development Inventories (CDI). Typically developing children completed the standard Spanish version of the inventory, whereas children with Down syndrome completed an adaptation designed for their cognitive and linguistic characteristics (Galeote et al., 2006). The CDI is one of the most widely used instruments for evaluating early language development and consists of a parent-report checklist in which caregivers indicate the words the child spontaneously understands and/or produces. The inventory includes hundreds of lexical items distributed across semantic and grammatical categories such as animals, food, toys, people, actions, descriptive terms, and routine expressions. Because it captures productive vocabulary based on children's everyday language use, the CDI has shown strong validity and reliability across multiple languages and developmental populations (Fenson et al., 2007).

In the present study, only expressive vocabulary data were used. For each participant, the set of words reported as produced in the CDI constituted the basis for constructing an individual lexical network. Specifically, each word produced by the child was represented as a node within the network. Semantic links between nodes were not derived from direct associative tasks, but from normative co-occurrence relations extracted from child-directed speech corpora in Spanish. This approach assumes that words frequently appearing together in natural linguistic environments are more likely to develop meaningful associative or semantic connections within the child's lexicon.

To estimate these relations, co-occurrence matrices were generated using corpora from CHILDES and complementary Spanish-language child-directed speech datasets. Co-occurrence analyses examined the frequency with which words appeared within a predefined moving window across utterances in child-directed speech. Words with statistically significant co-occurrence patterns were considered linked within the lexical network. Thus, individual networks were constructed by mapping each child's productive vocabulary onto the normative co-occurrence matrix, preserving only the relations between words actually produced by the child.

This procedure has been widely used in developmental network research to model early lexical organization. Previous studies have shown that co-occurrence-based semantic networks capture important structural properties of children's developing vocabularies, including clustering, hub formation, and small-world organization (Hills et al., 2009; Beckage et al., 2011). Rather than relying exclusively on adult semantic judgments, co-occurrence-based approaches provide an ecologically grounded representation of the linguistic environment to which children are exposed. Research has further demonstrated that these distributional patterns are closely associated with lexical acquisition and semantic growth during development (Hills et al., 2010; Carlson et al., 2014).

Network analyses were conducted using LEXNET-UAEM 1.0.6 (Rubí and Falcón, 2018) and Gephi (Bastian et al., 2009), which were employed to compute network metrics including density, degree, centrality, modularity, clustering, and semantic community detection.

Procedure

Co-occurrence matrix derived from a linguistic corpus

To estimate lexical relations among the words included in the CDI, a co-occurrence matrix was generated following procedures derived from the HAL model (Lund and Burgess, 1996) and the co-occurrence detection method proposed by Li et al. (2004). All analyses were conducted using LEXNET-UAEM 1.0.6 (Rubí and Falcón, 2018).

The set of co-occurrences was extracted from a corpus of 462,326 words drawn from the CHILDES database, which includes both child-directed and child-produced speech in interactions involving children aged 10 months to 12 years and their caregivers. LEXNET-UAEM computed the frequency with which two words appeared within a five-word window. Thus, a word A was connected to a word B if A appeared within the five words following B, resulting in directed networks in which the relation A → B does not imply reciprocity. The window size was fixed at five words in accordance with prior evidence from models of early linguistic input (Hills et al., 2010).

The co-occurrence processing yielded a complete co-occurrence matrix for the 445 nouns included in Inventory II of the CDI. This matrix was used to construct both global lexical networks and individual lexical networks.

Prior to network construction, the raw data were preprocessed by standardizing word forms, removing duplicates, and ensuring consistency between CDI entries and corpus-based representations. Only nouns present in both the CDI and the corpus were included in the analysis.

Generation of individual lexical networks

For each child, an individual lexical network was derived from the words reported as produced by caregivers. In these networks, nodes corresponded to words in the child's expressive vocabulary, and directed edges represent directed co-occurrence-based associations between lexical items in the normative co-occurrence matrix. The networks were constructed by importing the individual vocabularies and the co-occurrence matrix into LEXNET-UAEM (Rubí and Falcón, 2018), and subsequently into Gephi (Bastian et al., 2009) for analysis and visualization.

Generation of randomized lexical networks

Because the structure observed in individual lexical networks cannot be adequately interpreted without a reference indicating how much organization might arise by chance, randomized networks were generated based on the size of each child's vocabulary. For each of the 41 children with Down syndrome, a set of 100 randomized networks was constructed, preserving vocabulary size (n nodes) while selecting words randomly from the full set of items identified in the inventories. Connections were then generated using the same co-occurrence matrix employed to build the real networks.

To obtain an expected reference profile under random conditions, structural values derived from these 100 randomized networks (e.g., mean degree, density, average path length, and clustering coefficient) were averaged. These randomized networks preserve the general statistical properties of the linguistic environment while eliminating the structure derived from the specific selection of words that characterizes each child's vocabulary.

Structural analysis

For each network, structural metrics characterizing lexical organization were computed, including density, defined as the proportion of existing connections relative to the total number of possible connections; mean degree, reflecting the average number of links per node and indexing overall vocabulary connectivity; and several centrality measures, which identify words with a structurally prominent role in the network, either due to high connectivity or their function as bridging nodes between semantic subgroups. In addition, modularity was estimated to quantify the extent to which the network is organized into densely connected semantic communities, and hubs were identified as highly connected nodes that concentrate a disproportionate number of links and facilitate the integration and expansion of the lexical system.

Classification of lexical relations

Beyond global structural metrics describing the topology of the lexical network, it is essential to consider the nature of the links connecting words, as different types of semantic relations may differentially contribute to lexical organization and growth. Analysis of these links complements the structural characterization of the network by providing qualitative information about the semantic principles guiding its formation and development.

In this framework, the types of relations supporting connectivity among lexical nodes were examined. Links between words were classified using a semi-automated procedure that combined semantic resources (Spanish WordNet) with manual coding in ambiguous cases. Classification criteria followed those proposed by Borghi and Caramelli (2003) and Wu and Barsalou (2009).

Taxonomic relations were organized into three levels: superordinate (e.g., animal–cat), coordinate (e.g., chair–armchair), and subordinate (e.g., cat–Siamese). Thematic relations were classified based on typical locations, associated temporality, common situations, or instrumental functions. Finally, each noun was classified as referring to either a living being or an object. This procedure provided a systematic and reproducible classification of semantic relations, as shown in Table 1.

Comparative analyses between groups: typical development and down syndrome

Individual and global lexical networks were analyzed and visualized using Gephi (Bastian et al., 2009). For each network, standard centrality metrics (degree) were computed, highly connected nodes (hubs) were identified, and modularity algorithms were applied to detect semantic communities.

Comparisons between the typically developing (TD) and Down syndrome (DS) groups focused on four main dimensions:

Global network structure, assessing differences in cohesion, connectivity, and overall organizational patterns.

Distribution and organization of hubs and clusters, considering their structural role within each network.

Nature of semantic links, distinguishing between thematic and taxonomic relations.

Conceptual distribution of words, comparing categories such as living beings and non-living items.

This analytical approach enabled a fine-grained characterization of the internal organization of children's lexicons in both populations and facilitated the evaluation of differences in their conceptual and semantic architecture.

The results are presented at two complementary analytical levels. First, we examine whether the structural properties of lexical networks in children with Down syndrome differ from those expected by chance through comparisons with randomized networks. Second, we compare lexical network structure and content between children with typical development and children with Down syndrome matched by expressive vocabulary size.

DS networks vs. random networks

First, we examine whether the structural properties observed in the lexical networks of the Down syndrome group reflected a specific conceptual organization or could be explained by random effects, 41 networks of children with Down syndrome were analyzed. For each participant, a set of 100 randomized networks was constructed, preserving vocabulary size (n nodes) while selecting words randomly from the full set of items identified in the inventories. Connections were then generated using the same co-occurrence matrix employed to build the real networks. Structural values derived from these 100 randomized networks (e.g., mean degree, density, average path length, and clustering coefficient) were averaged and used as a baseline for each participant.

Comparisons between real networks and their corresponding randomized counterparts indicated that the lexical organization of the Down syndrome group cannot be attributed to chance. Overall, real networks exhibited more robust structural patterns than randomized networks (Table 2).

These results indicate that lexical networks in the Down syndrome group are more highly connected, more clustered, suggesting that the conceptual organization of their vocabularies reflects meaningful regularities of the linguistic environment rather than stochastic properties.

Consistent with this interpretation, networks from children with DS showed greater lexical cohesion and a more compact global structure than would be expected under random conditions. In particular, the presence of hubs and semantic clusters reflects a systematic and non-random pattern of lexical organization, comparable to structural principles previously reported in typically developing populations.

Group comparisons: typical development vs. down syndrome

Second, we compared differences in lexical structure and content, paired vocabularies from both groups were analyzed. The final analyses included 20 children: 10 with typical development and 10 with Down syndrome. The groups were matched by expressive vocabulary size, such that each child with DS was paired with a child with TD with a maximum difference of ±8 words. This methodological decision, replicated from previous studies comparing distinct language trajectories by matching lexical level rather than chronological age (Yoder and Warren, 2004), allowed for the examination of differences in lexical organization independently of overall vocabulary size. Prior research has shown that the organization and structural properties of the lexicon depend primarily on accumulated vocabulary rather than age (Hills et al., 2009; Steyvers and Tenenbaum, 2005; Beckage et al., 2011). Accordingly, vocabulary matching provides a more valid basis for comparing qualitative differences in lexical structure between groups.

Structural metrics

For each participant, global lexical network metrics were computed, including number of nodes, number of edges, mean degree, and clustering coefficient. Comparisons of these metrics revealed differences in the structural organization of the lexicon between the two groups (Table 3). The DS group showed a higher number of edges compared to the TD group, despite comparable vocabulary size.

Networks from children with Down syndrome also showed higher mean degree and clustering coefficient than networks from typically developing children. To statistically assess these group differences, independent-samples t tests were conducted comparing network metrics across the 10 vocabularies in the TD group and the 10 vocabularies in the DS group. Results revealed a significant difference in mean clustering coefficient, with DS networks showing higher values (M = 0.19, SD = 0.04) than TD networks (M = 0.13, SD = 0.066), t (18) = 2.37, p < 0.05. DS networks also showed a marginally higher number of edges (M = 701.0, SD = 659.6) compared with TD networks (M = 263.7, SD = 155.7), t (18) = 2.37, p = 0.056.

These network metrics should be interpreted in light of the method used to derive lexical relations. Because all connections were generated from a corpus-based co-occurrence matrix, they reflect distributional patterns rather than direct semantic relations. Importantly, the clustering coefficient was computed using a normalized measure implemented in Gephi (Bastian et al., 2009), which accounts for the proportion of existing connections among a node's neighbors relative to all possible connections. Thus, clustering captures local interconnectedness beyond overall network density.

Graph visualizations revealed that, although semantic communities or clusters emerged in both groups, their nature and distribution differed markedly. In the TD group, an overall balance was observed between clusters organized around conceptual categories (e.g., animals, foods) and clusters based on functional associations. In contrast, networks in the DS group tended to exhibit more local and denser clusters, with predominantly functional or contextual connections.

Node size reflected word degree, allowing for the identification of lexical hubs. In networks from children with Down syndrome, hubs typically corresponded to words that were highly frequent in everyday experience, whereas in typically developing children, hubs were more diverse and included words belonging to broader conceptual categories (see Figure 1).

Lexical relations (taxonomic vs thematic)

Each individual vocabulary was analyzed to classify both the types of lexical relations and the predominant noun types.

Two primary types of relations were considered:

Taxonomic relations, based on shared category membership or properties.

Thematic relations, based on functional or contextual associations.

Comparative analyses showed that children with Down syndrome exhibited a clear predominance of thematic (72%) over taxonomic (28%) relations (see Figure 2). Their vocabularies were structured primarily around functional associations, such as objects used together or elements of daily routines. In contrast, typically developing children displayed a more balanced distribution between thematic (59%) and taxonomic (41%) relations. In this group, categorical connections were proportionally more frequent, reflecting organization by semantic fields (Table 4).

A binomial mixed-effects model revealed a significant effect of group (β = 0.58, SE = 0.06, z = 9.44, p < 0.001), indicating that children with Down syndrome showed higher odds of producing thematic relative to taxonomic relations compared to typically developing children (OR = 1.78). This pattern suggests that children with typical development tend to organize their lexicon around conceptual categories, whereas children with Down syndrome rely predominantly on contextual and functional links.

Types of nouns (living vs. non-living items)

Vocabularies were further classified according to the semantic nature of the nouns produced, distinguishing between living beings (people, animals) and non-living items (objects, food, toys, clothing, furniture).

As shown in Table 5, children with Down syndrome produced proportionally fewer nouns referring to living beings (24%) and a higher proportion of non-living items (76%) compared to typically developing children who showed the inversed proportion (living items, 53%; non-living items 47%) (see Figure 3). This pattern suggests a more functional and context-dependent lexical organization, with a stronger focus on objects and elements of the immediate environment.

A binomial mixed-effects model revealed a significant effect of group (β = −0.59, SE = 0.18, z = −3.35, p = 0.001), indicating that the proportion of living vs. non-living words differed between children with typical development and those with Down syndrome (OR = 0.55, 95% CI [0.39, 0.78]).

Lexical networks in children with Down syndrome (DS) exhibit a systematic, non-random organization. Comparisons with randomized networks revealed higher clustering coefficients, greater mean degree, and the consistent presence of hubs in real networks, indicating that lexical structure in DS reflects genuine principles of conceptual organization rather than chance effects, consistent with foundational models of non-random network organization and efficiency (Watts and Strogatz, 1998; Newman, 2003).

Differences in lexical network structure

By matching children with typical development (TD) and children with DS on expressive vocabulary size, this study isolates qualitative differences in lexical organization from quantitative differences in vocabulary growth. Contrary to expectations derived from previous network-based studies in atypical developmental populations (Beckage et al., 2011; Ehlen et al., 2020), children with Down syndrome did not exhibit weaker or less connected lexical networks. Instead, their lexical networks showed denser clusters and stronger local connectivity, suggesting that atypical development may involve alternative forms of semantic structuring rather than reduced organization per se, whereas TD networks exhibited a more distributed structure. These differences are reflected in structural measures such as clustering coefficient and degree, which in network science have been associated with patterns of local interconnectedness and potential efficiency of information flow within the network (Watts and Strogatz, 1998; Steyvers and Tenenbaum, 2005; Hills et al., 2010). From this perspective, the observed patterns suggest distinct configurations in how lexical items are organized and potentially accessed across groups.

The observed structural differences may have important implications for lexical processing. Network structure constitutes a necessary first step for understanding processing, as the topology of connections constrains how activation can potentially spread across lexical items. However, the present study does not directly assess processing dynamics, and therefore these interpretations should be understood as structural inferences rather than direct evidence of differences in lexical processing. The present study is specifically designed to characterize the structural organization of the mental lexicon rather than its dynamic activation properties. Future research could build on the network structures identified here by implementing spreading activation models or other computational approaches to explicitly examine how differences in connectivity and clustering influence lexical access over time. Thus, the present findings provide a structural foundation that can inform subsequent investigations into the dynamic mechanisms underlying lexical processing in both populations.

An important caveat concerns the interpretation of the structural properties derived from the lexical networks. Because all connections were defined based on a co-occurrence matrix extracted from linguistic input, the resulting network structure reflects distributional regularities rather than direct semantic relations independent of the input. In this sense, the observed organization captures the extent to which individual vocabularies align with the statistical structure of the linguistic environment. This consideration is particularly relevant for interpreting differences in connectivity across groups. Alternative approaches based on association norms or other semantic similarity measures could yield different patterns of connectivity and, consequently, different structural properties. Therefore, the present findings should be understood within a distributional framework of lexical organization, in which network structure is constrained by the statistical properties of linguistic input.

Importantly, the predominance of thematic and context-based relations observed in the DS networks may reflect distinct trajectories of lexical-semantic self-organization across developmental populations. Whereas TD children showed a more balanced integration of thematic and taxonomic relations, children with Down syndrome appeared to rely more strongly on contextual and experiential associations among words. This pattern may suggest that lexical growth in DS is shaped by organizational principles that prioritize event-based and situational regularities over abstract categorical integration. From a developmental perspective, these findings support the possibility that differences between populations are not limited to delayed lexical acquisition, but may involve qualitatively distinct pathways in the emergence and restructuring of semantic networks. Consequently, the study of lexical networks in Down syndrome may provide important insights into how cognitive constraints, linguistic experience, and semantic organization interact during language development.

Thematic vs. taxonomic relations as organizing principles

Analyses of lexical relations revealed a predominance of thematic, context-based relations in the DS group, in contrast to a more balanced integration of thematic and taxonomic relations in TD children. In TD, the greater weight of taxonomic relations reflects the emergence of more abstract, hierarchically structured conceptual representations (Waxman and Markow, 1995; Arias-Trejo and Plunkett, 2013). In contrast, the predominance of thematic relations in the DS group suggests a lexical organization grounded in functional associations, routines, and situational context, consistent with experience-based accounts of conceptual development in early childhood (Miller, 1999; Nelson, 2007).

Importantly, this difference may reflect distinct underlying mechanisms of semantic organization, where TD development progressively integrates categorical knowledge, whereas DS development may prioritize functional and context-dependent relations.

The higher connectivity observed in the DS group reflects a greater number of co-occurrence-based links among the words included in their vocabularies. This pattern is consistent with the predominance of thematic relations identified in this group, whereas the TD group showed a more balanced distribution between taxonomic and thematic relations. The greater proportion of thematic connections in the DS networks may be related to the type of linguistic and experiential input associated with this population, which has been described as more context-dependent and grounded in routine-based interactions, potentially favoring associative and situational links over hierarchical, category-based organization.

Local cohesion and experiential grounding in down syndrome

The higher local clustering observed in DS lexical networks may be consistent with cognitive and learning profiles that have been widely described in this population. Individuals with Down syndrome typically show slower processing speed, limitations in verbal short-term memory, and a stronger reliance on contextual support during learning (Chapman, 1997; Vicari et al., 2000). From a developmental perspective, these characteristics have been interpreted within frameworks emphasizing atypical but structured developmental trajectories, in which cognitive organization emerges through experience-dependent processes rather than as a simple delay of typical development (Karmiloff-Smith, 1998).

Within this framework, the observed pattern of higher local cohesion may reflect a lexicon that is more strongly grounded in repeated, contextually rich experiences, where associations are reinforced through frequent co-occurrence in everyday routines. Such organization is consistent with learning profiles that privilege context-dependent and functionally embedded representations over more abstract, category-based structures. From a network perspective, this type of organization may facilitate lexical access within familiar contexts, while potentially constraining the integration of more distant or hierarchically structured semantic relations.

Differences in noun types and conceptual anchoring

Group differences were also observed in the semantic nature of produced nouns. Children with DS produced proportionally fewer nouns referring to living beings and a higher proportion of non-living, concrete objects. This pattern suggests that lexical development in this group may be more strongly anchored in manipulable objects and elements of everyday routines, which are directly accessible through perception and action. This distribution may reflect differences in the balance between perceptual, functional, and conceptual constraints on early word learning. Words referring to non-living objects are often associated with immediate interaction, repeated routines, and stable contextual cues, which may facilitate their acquisition and integration into the lexicon. In contrast, biological categories such as animals and people may require higher levels of abstraction and generalization across instances, placing greater demands on memory, categorization, and representational flexibility.

The observed pattern is consistent with experience-based accounts of early semantic development, in which lexical knowledge is initially organized around recurrent events, routines, and contextual interactions rather than abstract taxonomic categories (Nelson, 2007). In this sense, the predominance of non-living, manipulable objects in the DS group may reflect a lexicon that is more tightly grounded in event-based and context-dependent representations. This interpretation aligns with the predominance of thematic relations and local clustering observed in the present study, suggesting a coherent profile of lexical organization centered on concrete, functionally grounded experience.

Implications for psycholinguistic theory

Together, these findings contribute to psycholinguistic accounts of lexical development by showing that lexical organization is not determined solely by vocabulary size, but by the structure and nature of the relations linking words. Rather than reflecting a purely quantitative accumulation of lexical items, development involves changes in how words are interconnected within a structured semantic system (Hills et al., 2010; Beckage et al., 2011).

Our results support models of the mental lexicon that conceptualize lexical knowledge as a structured network shaped by both distributional properties of linguistic input and cognitive constraints on learning (Steyvers and Tenenbaum, 2005; Siew, 2019). While both groups exhibit general network properties such as hubs and clusters, the underlying architecture of their lexical networks diverges systematically, reflecting distinct trajectories of conceptual organization. In typically developing children, the presence of a more balanced integration of taxonomic and thematic relations is consistent with accounts proposing a gradual shift toward increasingly abstract, category-based representations (Waxman and Markow, 1995; Arias-Trejo and Plunkett, 2013). In contrast, the predominance of thematic relations and stronger local clustering observed in children with Down syndrome suggests a lexicon that is more tightly grounded in contextual and experiential associations (Nelson, 2007; Karmiloff-Smith, 1998). Rather than reflecting a simple developmental delay, the observed pattern suggests a distinct pathway of semantic organization in Down syndrome, characterized by a reduced role of biologically grounded categories and an increased reliance on context-driven associations. This pattern aligns with theories emphasizing the role of experience-dependent learning and the influence of cognitive factors such as memory and processing constraints on lexical organization.

Present results further suggest that variability in lexical organization may arise from differences in how learners sample and integrate linguistic input, leading to distinct network configurations even under comparable levels of vocabulary size. This highlights the need for psycholinguistic models to account for both the statistical structure of the input and the learner's interaction with that structure in shaping the mental lexicon. These findings may also have implications for educational and therapeutic interventions. Children with Down syndrome may benefit from leveraging thematic organization, routine-based learning, and existing lexical hubs as entry points for introducing broader conceptual categories. Overall, the network-based approach adopted here offers a powerful framework for understanding diverse developmental trajectories of the mental lexicon and their implications for early language acquisition.

Future directions

Future research could build on the network structures identified here by implementing spreading activation models (e.g., Siew, 2019; Citraro et al., 2026) to explicitly examine how differences in connectivity and clustering influence lexical access over time. In addition, extending this network-based approach to longitudinal designs would allow researchers to examine how lexical network structure evolves within individuals and whether early patterns of local cohesion or thematic dominance predict later vocabulary growth and categorical reorganization. Integrating online measures of lexical processing, such as eye-tracking or reaction-time paradigms, would further enable direct links between network structure and real-time lexical access to be established.

Further work should also explore how targeted interventions influence lexical network architecture, particularly whether instructional approaches designed to promote taxonomic relations can induce measurable changes in network organization in children with Down syndrome. Finally, expanding this framework to other neurodevelopmental profiles would help clarify which aspects of lexical network structure are shared across populations and which reflect condition-specific developmental trajectories.