DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 1
Developmental Changes in Semantic Knowledge Organization
Anna V. Fisher¹
Samuel L. Ventura²
Christopher J. MacLellan³
¹ Department of Psychology, Carnegie Mellon University, Pittsburgh PA
² Department of Statistics, Carnegie Mellon University, Pittsburgh PA
³ Human Computer Interaction Institute, Carnegie Mellon University, Pittsburgh PA
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 2
Semantic knowledge is a crucial aspect of higher cognition. Theoretical accounts of semantic
knowledge posit that relations between concepts provide organizational structure that converts
information known about individual entities into an interconnected network, in which concepts
can be linked by many types of relations (e.g., taxonomic and thematic relations). The goal of the
present research is to address several methodological shortcomings of prior studies on the
development of semantic organization, by using a variant of the Spatial Arrangement Method
(SpAM) to collect graded judgments of relatedness for a set of entities that can be cross-
classified into either taxonomic or thematic groups. In Experiment 1 we used the cross-classify
SpAM (CC-SpAM) to obtain graded relatedness judgments and derive a representation of
developmental changes in the organization of semantic knowledge. In Experiment 2 we validated
the findings of Experiment 1 by using a more traditional pairwise similarity judgment paradigm.
Across both studies, we found that an early recognition of links between entities that are both
taxonomically- and thematically-related preceded an increasing recognition of links based on a
single type of relation. The utility of CC-SpAM for evaluating theoretical accounts of semantic
development is discussed.
Keywords: Semantic Knowledge; Cognitive Development; Conceptual Change
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 3
Developmental Changes in Semantic Knowledge Organization
Semantic knowledge refers to information that is learned about entities (e.g., objects,
animals, and people) and relations between them. Many accounts of semantic knowledge posit
that information known about individual entities is organized into an interconnected network
according to relations between them (Bjorklund, 1987; Collins & Loftus, 1975; 2003; Crowe &
Prescott, 2003; McClelland & Rogers, 2003), and that this organization plays a critical role in a
wide array of cognitive tasks (Belke, Humphreys, Watson, Meyer, & Telling, 2008; Bjorklund &
Jacobs, 1985; Fisher, Godwin, Matlen, & Unger, 2014; Moores, Laiti, & Chelazzi, 2003;
Roediger & McDermott, 1995). Therefore, understanding how the organizational structure of
semantic knowledge is acquired and how it changes over the course of development and learning
is a core aspect of understanding higher-order cognition.
Research on semantic organization development has primarily focused on the influence
of different types of relations on organizational structure. For instance, entities may be linked by
taxonomic relations when they share features that indicate that they are of the same “kind” (e.g.,
both chickens and eagles have wings and beaks), or by primarily thematic relations when they
are associated with the same environment (e.g., both chickens and cows can be found on a farm).
The influence of these relations may change over the course of development. For instance,
computational models of semantic development suggest that learning features that entities share
increasingly organizes knowledge according to taxonomic relations (Hills, Maouene, Maouene,
Sheya, & Smith, 2009; Kemp & Tenenbaum, 2008; McClelland & Rogers, 2003). However,
characteristics of paradigms used to study developmental changes in the influence of different
relations on semantic organization may have limited the degree to which the data they yield can
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 4
inform this question. Below, we first review prior research on semantic development and the
paradigms they have used, then describe a novel paradigm designed to provide new insights into
the development of semantic organization.
Insights into Semantic Development from Match-To-Sample Tasks
Much of semantic development research has used match-to-sample tasks, in which
participants are asked to match a target item with one of at least two other items. Commonly, the
items that participants can match to the target includes a taxonomic match and a thematic match
(e.g., Smiley & Brown, 1979; Walsh, Richardson, & Faulkner, 1993; Waxman & Namy, 1997).
For instance, given a target dog, participants may be asked to choose between a thematic match
(e.g., bone) and a taxonomic match (e.g., cat). Researchers can then assess children’s preferences
for matching on the basis of different types of relations. Early research using this paradigm
indicated that children up to approximately age six prefer thematic matches, after which children
increasingly prefer taxonomic matches (Smiley & Brown, 1979). Subsequent research suggested
that children’s responses can be influenced by multiple factors. For instance, Waxman and Namy
(1997) found that instructions to “choose another one” yield a greater proportion of taxonomic
choices, whereas instructions to “choose the one that goes best with [the target]” yield a greater
proportion of thematic choices.
In an alternate version of this paradigm, the items that participants can match to the target
include one item that is related to the target on some dimension, and one or more unrelated
items. This approach allows researchers to investigate whether children can differentiate between
related versus unrelated items, rather than children’s preference for one type of one relation over
another. In one such study, items belonged exclusively to the domain of foods, such that
taxonomically-related items were foods of the same kind (e.g., meats, dairy, etc.), and
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 5
thematically-related items were foods associated with the same meal (e.g., breakfast, lunch, etc.)
(Nguyen & Murphy, 2003). In a related study, items belonged to multiple domains (Nguyen,
2007). Across both studies, children selected both taxonomically- and thematically-related items
more frequently than unrelated items from an early age (age two in Nguyen, 2007; age four in
Nguyen & Murphy, 2003)
Despite yielding insights into developmental changes in children’s recognition of and
preference for different types of relations, the degree to which findings from studies using the
match-to-sample paradigm can shed light on the developmental trajectory of semantic
organization may be limited in several respects. First, participants can only judge a small set of
items per trial, and the number of trials given to young children is typically low due to children’s
difficulty with long tasks. Consequently, data are collected for only a small set of items in a
given study. Therefore, no match-to-sample studies have assessed whether participants can
match a target item with more than one item that bears a given type of relationship to it. For
instance, a study may reveal that participants can match a target dog to a taxonomic match cat,
but not whether dog is integrated with a group of taxonomically-related entities in a semantic
knowledge network. This limitation particularly hinders the evaluation of predictions made by
computational models that simulate learning-driven changes in semantic knowledge
organization, such as the prediction that learning increasingly organizes knowledge of entities
into taxonomic groups (Hills et al., 2009; Kemp & Tenenbaum, 2008; McClelland & Rogers,
This limitation may be compounded by the possibility that children’s ability to match a
target to a taxonomically-related item is influenced by the presence of overlapping thematic
relations between them. For instance, many items designated as taxonomic matches for targets
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 6
across studies may in fact be both taxonomically- and thematically-related to their respective
targets, such as table-chair in Waxman and Namy’s (1997) study: Tables and chairs often co-
occur, and are each other’s strongest thematic associates according to free association norms (D.
Nelson, McEvoy, & Schrieber, 1998). Similar issues characterize taxonomic matches in other
studies (e.g., carrot-corn in Nguyen & Murphy, 2003; dog-cat in Smiley & Brown, 1979). The
observation that children often justify taxonomic matches thematically (Hashimoto, McGregor,
& Graham, 2007; Tversky, 1985; Walsh et al., 1993) provides evidence for the contribution of
thematic relations to the recognition of taxonomic relations. This potential role for taxonomic
and thematic overlap highlights the need for assessing children’s ability to recognize taxonomic
relations in both the presence and absence of thematic relations, which would further strain the
limits of match-to-sample paradigms.
Second, judgments in match-to-sample paradigms are binary: A given item is either
chosen as a match for a target, or it is not. In contrast, entities may be more or less related to each
other to varying extents. Therefore, it is preferable for an assessment of semantic organization to
collect a graded measure of the degree to which participants perceive entities as related.
Insights into Semantic Development from Word Association Tasks
Free Word Association
Free word association has been used in psychology to study properties of the mind for
well over a century. Most relevant to the present discussion are the studies that used this
technique to discover the syntagmatic-to-paradigmatic shift phenomenon (for review, see K.
Nelson, 1977). This phenomenon describes a developmental transition in free word association
tasks, such that young children’s responses tend to be syntagmatic – i.e., temporally contiguous
with the stimulus word in a syntactic sequence (e.g., soft-pillow) – whereas older children and
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 7
adults’ responses tend to be syntagmatic – i.e., come from the same grammatical class as the
stimulus word and have a related meaning (e.g., synonyms such as soft-gentle or opposites such
. The syntagmatic-paradigmatic distinction is therefore similar to the distinction
between thematic and taxonomic relations, respectively.
The syntagmatic-to-paradigmatic shift occurs between five and ten years of age, although
several deviations from this general pattern have been documented (e.g., paradigmatic responses
are higher across ages for common word classes such as nouns, Entwisle, Forsyth, and Muuss
(1964), and frequent words, K. Nelson (1977)). Overall, paradigmatic responses increase
between ages five and 10 as follows: 61.2%-78.1% for nouns, 16.8%-78.5% for adjectives, and
16.6%-59.6% for verbs (Entwisle, 1966).
A number of theories have been proposed to account for the syntagmatic-to-paradigmatic
shift (for review, see K. Nelson, 1977). Although an in-depth discussion of these theoretical
accounts is beyond the scope of this paper, the class of theories that assume that responses on
free word association tasks reflect perceived semantic relatedness between the words’ referents
(e.g., Anderson & Gordon, 1973) is relevant to the present discussion. Under this interpretation,
the syntagmatic-to-paradigmatic shift phenomenon is consistent with the idea that knowledge
organization undergoes change with development, and aligns well with match-to-sample task
findings that show a shift from thematic to taxonomic preferences
Similar to match-to-sample tasks, traditional free word association tasks suggest that
development and learning may lead to changes in knowledge organization, but only yield a
binary measure (i.e., associated vs. not associated) of perceived relatedness between individual
Additional types of paradigmatic response include coordinates (e.g., ‘table-chair’), superordinates (e.g., ‘table-
furniture), and contrasts (e.g., sell-buy).
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 8
pairs of words. However, a variant of the free word association task – the semantic fluency
paradigm – provides richer data that can potentially illuminate knowledge organization and its
The Semantic Fluency Paradigm
In the semantic fluency paradigm, data that can be used to derive graded measures of
relatedness between groups of entities is collected by asking participants to list as many items as
they can think of from a target category (e.g., “Animals”), in a short period of time (Crowe &
Prescott, 2003; Storm, 1980; Winkler-Rhoades, Medin, Waxman, Woodring, & Ross, 2010). The
proximity of items to each other across lists is taken as a distance measure that captures the
degree to which items are perceived as related. These distance data are then submitted to
analyses designed to derive representations of the way in which mental representations of these
items are organized according to relationships between them. For example, hierarchical cluster
analysis identifies local clusters of items linked by short distances that are in turn embedded
within higher-order clusters of clusters. An examination of the relationships that apparently
determine whether concepts are closely linked is taken to reveal the relationships that organize
semantic knowledge. For example, Crowe and Prescott (2003) determined that hierarchical
clustering solutions for children aged 5 to 10 were predominantly organized according to
thematic relationships, such that children tended to list animals associated with the same
environment together (see also Lucariello, Kyratzis, & Nelson, 1992). Using an alternate analysis
approach, Winkler-Rhoades et al. (2010) found organizational structure governed by thematic
relations for children aged 4 to 9 and adults across both urban and rural North American
communities. Similarly, Storm (1980) found no significant differences between hierarchical
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 9
clustering solutions for age groups ranging from Kindergarten to Adults, although the
relationships that governed these solutions were not clear.
However, this apparent static developmental trajectory may have been due to a number of
characteristics of the semantic fluency task that may bias the outcome data it yields. First, over
90% of the animals that participants listed across studies were mammals. Data collected from
within a taxonomic category have limited potential to reveal distinctions between concepts based
on taxonomic relations. Second, the analysis procedures used in semantic fluency studies further
constrain the composition of the datasets. Because the proximity measure cannot be derived for
any item that does not appear at least once in the same list with each of the other items
participants produce, items that do not meet this criterion are excluded. Items included in
analysis are further constrained by the fact that researches typically exclude infrequent items,
such as those produced by fewer than 20% of participants.
Finally, the order in which participants list items in the semantic fluency task may be
influenced both by the organization of semantic knowledge, and the way in which semantic
knowledge is searched during memory retrieval. Theoretical accounts of this process suggest that
it consists of switches between two stages: Global search, in which memory is probed for a new
item to list based on global cues such as overall frequency, and local search, in which memory is
probed based on local cues such as relatedness with previously listed items (e.g., Hills, Todd, &
Jones, 2015). The use of global cues such as overall frequency may account for the bias towards
listing mammals in the semantic fluency task, as their labels may be more common. Importantly,
recent evidence suggests that local search involves a sequential transition between the last listed
item and a new item to which it is semantically related, rather than an entire cluster of related
items (Hills et al., 2015). Although the nature of the semantic relationships that such sequential
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 10
transitions follow requires further study, Hills, Jones, and Todd (2012) demonstrated that search
through a model semantic knowledge network in which relationships are derived from patterns
of lexical co-occurrence successfully captured semantic fluency task performance. Taken
together, these issues indicate that the semantic fluency task may provide a meaningful measure
of search through semantic space that focuses on familiar animals such as mammals, and follows
certain relational links such as word-to-word co-occurrence. By the same token, the semantic
fluency task may be a biased measure of the overall structure of semantic organization.
The purpose of the present research is to investigate the influence of taxonomic relations,
thematic relations, and their overlap on the development of semantic organization. To collect a
graded measure of semantic organization while avoiding the limitations of previous paradigms,
we used an approach that yields a full set of pairwise relatedness judgments without relying on
spontaneous production data. Specifically, we pre-selected a set of organisms drawn from cross-
cutting taxonomic and thematic groups. To examine the influence of these relations on semantic
organization, we conducted two studies in which we asked participants to make relatedness
judgments of these stimuli by arranging representations of the organisms on a grid in a task
modeled on the Spatial Arrangement Method introduced by Goldstone (1994). Because the task
includes a set of items that are cross-classifiable into either taxonomic or thematic groups, we
refer to it as the Cross-Classify Spatial Arrangement Method (CC-SpAM).
The CC-SpAM task yields distances between the full set of cross-classifiable items,
which is taken as an index of the degree to which they were perceived as related. Prior research
demonstrated that this approach provides an efficient means of collecting pairwise relatedness
judgments for a set of items that are consistent with participants’ responses in other, well-
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 11
established similarity judgment tasks (Hout, Goldinger, & Ferguson, 2012; Montez, Thompson,
& Kello, 2015). Although this task requires participants to judge relatedness within two-
dimensional space, it yields representations of structure that are comparable to those derived
from tasks that do not have this constraint, such as rating the similarity of individual pairs of
items (Hout et al., 2012). Moreover, Lowe (2001) demonstrated that semantic space can be well-
characterized in two dimensions, suggesting that relatedness judgments made along two-
dimensions can capture the structure of semantic organization.
Prior research suggests that children can use a spatial layout to make graded relatedness
judgments, and that these judgments are correlated with other forms of reasoning on the basis of
relationships. For example, in Howard and Howard’s (1977) study, children in 1st to 6th grade
made relatedness judgements of animals pairs by placing each animal in one of six adjacent
“cages”. The youngest children made judgments on the basis of a continuous size dimension,
suggesting that children can convert their perceptions of relatedness into distance judgments.
However, it is worth noting that the constraint of judgments to one dimension (a row) and
stimulus animals to land mammals limits the degree to which this study’s results can illuminate
the influence of multiple types of relationships on semantic organization.
Recent spatial arrangement studies provide evidence that children can use a two-
dimensional layout to make relatedness judgments of entities that are correlated with the way in
which children reason about these entities. In a longitudinal study, Fisher, Godwin, and Matlen
(2015) found that the degree to which preschool-age children’s spatial relatedness judgments
differentiated between both taxonomically- and thematically-related entities versus unrelated
entities was correlated with the rate at which children made inductive inference judgments on the
basis of these relationships. For instance, children who placed alligator and crocodile closer
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 12
together than alligator and fish also showed higher rates of inductive inferences from alligator to
crocodile versus an unrelated lure. Moreover, developmental increases in differentiation on the
spatial arrangement task were linked with higher rates of inductive inferences for related versus
unrelated animals. Therefore, young children’s two-dimensional spatial relatedness judgments
provide a meaningful measure of semantic organization that is linked with other forms of
The spatial arrangement paradigm was implemented in two different forms in the present
research. In Experiment 1, we generated a cross-classifiable set of 15 organisms that equally
afford grouping into three taxonomic groups (mammals, birds, and plants) and three thematic
groups (farm, aquatic, and wild/zoo). For example, the set of stimuli includes organisms such as
‘seaweed’ (thematically-related to other aquatic organisms, and taxonomically-related to other
plants), and ‘chicken’ (thematically-related to other farm organisms, and taxonomically-related
to other birds). The specific thematic groups were selected for two reasons. First, these thematic
groups frequently influenced the organization of children’s animal lists in semantic fluency
studies, suggesting that information about these groups is represented in children’s knowledge of
animals. Second, these thematic groups have been used in prior developmental research (Blaye,
Bernard-Peyron, Paour, & Bonthoux, 2006; Smiley & Brown, 1979; Waxman & Namy, 1997).
In this experiment, participants were asked to arrange cards depicting the same
predetermined set of 15 organisms on a game board three times. The use of multiple arrangement
trials was designed to give participants the opportunity to arrange the organisms in different
ways. Each of these three arrangement trials was preceded by a different prompt, in which
participants were asked to put together things that “go together,” “are the same kind of thing,” or
“match”. The use of multiple prompts was based on past findings that prompt wording may
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 13
influence the relationships on which participants focus (Lin & Murphy, 2001; Waxman & Namy,
The implementation of the spatial arrangement paradigm in Experiment 2 was designed
to address two possible factors that may influence children’s judgment in Experiment 1. First, to
address the possibility that children may have difficulty applying what knowledge they have of
cross-cutting relationships while arranging a full set of items on a board, children in Experiment
2 made relatedness judgments of individual pairs of items. Second, although the images of
organisms used in Experiment 1 were designed to control for differences in immediately visible
perceptual similarity between related versus unrelated organisms (see Methodology section), the
mere presence of images may render perceptual features more accessible. Therefore, organisms
in Experiment 2 were represented using wooden blocks. Note that organisms were also
represented using blocks in both Fisher et al. (2014) and Howard and Howard (1977), suggesting
that children can maintain which organisms blocks represent in memory during a spatial
arrangement task. Finally, many of the organisms used in Experiment 1 were replaced with new
organisms in Experiment 2 to test whether the patterns of developmental changes observed in
Experiment 1 generalize to new items.
To study the development of semantic organization across an age range that has been
characterized as a period of dramatic organizational change (e.g., Carey, 1985), we administered
these tasks to children from four age groups, ranging from pre-school to second grade. To derive
a representation of the end point of this developmental trajectory, we also collected data from
adults. We then used these data to assess the degree to which concepts linked by taxonomic,
thematic, or both relations were perceived as more closely related to each other than unrelated
concepts in each age group. Additionally, we used the data collected from Experiment 1 in
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 14
analyses designed to derive representations of the latent structure underlying semantic
organization in each age group.
The sample included 18 participants in each of five age groups: Preschool (Mage=4.5
years, SD=0.83; 11 female), Kindergarten (Mage=5.6 years, SD=0.52; 8 female), First Grade
(Mage=6.5 years, SD=0.44; 9 female), Second Grade (Mage=7.6 years, SD=0.72; 9 female), and
Adults (Mage=19 years, SD=1.84; 9 female). Children were recruited from pre-schools and
schools in a middle-class area in a Northeastern US city. Adults were recruited from the
undergraduate population at a private university in the same city and participated in exchange for
partial course credit. Additionally, 10 adults were recruited via Amazon Mechanical Turk to
complete a calibration questionnaire, for which they were compensated at a rate of $5/hour.
Calibration Questionnaire. To identify a set of plants and animals that were each
strongly associated with one of three thematic groups (farm, aquatic, and wild/zoo), 45 plant and
animal names were presented in a questionnaire administered to adults recruited via Amazon
Mechanical Turk. Participants were asked to rate each item according to the degree to which it
was associated with each of the three thematic groups. Ratings were made on a four-point Likert
scale, such that participants could indicate whether a given plant or animal was “Not
Associated”, “Weakly Associated”, “Moderately Associated” or “Strongly Associated” with
either a Farm, Aquatic, or Wild/Zoo environment. Responses on the Likert scale were converted
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 15
to numerical values, such that “Not Associated” corresponded to a value of 1, and “Strongly
Associated” corresponded to a value of 4.
Picture Sorting Task. The materials included a game board, a stimulus sheet, 15
stimulus cards, and a camera. The game board was a 10x10 grid of 2.5” squares. The stimuli
were a set of 15 plants and animals that were cross-classifiable into one of three taxonomic
groups (mammal, bird, or plant) and one of three thematic groups (farm, water, and wild/zoo).
Assignment of stimuli to taxonomic groups was based on whether the organism was a mammal,
bird, or plant. Assignment of stimuli to thematic groups was based on the results from the
calibration questionnaire. Each of the five stimuli assigned to a given thematic group were those
that achieved an average rating of ≥ 3 out of 4 on the calibration questionnaire for that group (see
Table 1). For a given item, ratings for its assigned thematic group (M=3.59) were significantly
higher than its ratings for the other two thematic groups (M=1.76, t(38.32)=19.16, p<.0001).
Association ratings for items to their respective groups were comparable for each of the three
thematic groups (Aquatic, Farm, and Wild/Zoo Ms=3.68, 3.64, and 3.44).
Plants and animals used in Experiment 1.
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 16
The selected items were presented as black and white line drawings intended to minimize
shared visual features of items to each other. To test for potential differences in visual similarity
between items from the same versus different groups, these stimuli were submitted to a Scale
Invariant Feature Transform (SIFT) analysis, a computer vision algorithm which computed the
pairwise featural similarity between the line drawings (D. G. Lowe, 1999). Line drawings of
organisms within the same taxonomic or thematic group were no more similar to each other than
drawings of organisms in different groups, F(3,101)=.62, p=.607. These line drawings were both
arranged in a random order on the stimulus sheet (see Figure 1) and depicted individually on
Participants were tested individually in a quiet space. First, participants were told that
they were going to play a game in which their job was to help a fictional character, Zibbo,
organize his favorite things in “a few different ways”. The experimenter then showed
participants the stimulus sheet (depicted in Figure 1), named each organism, and removed the
Figure 1. Images used in Experiment 1, presented in the configuration used for the stimulus
sheet shown to participants at the beginning of the experiment.
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 17
sheet from view. Each participant then completed three arrangement trials, in which they were
asked to arrange the same set of 15 stimulus cards on the game board. Each trial began with a
prompt to put together plants and animals that: “Go together”, “Are the same kind of thing”, or
“Match”. These prompts were selected based on prior evidence that different prompts promote
either thematic or taxonomic grouping (Deák & Bauer, 1995; Waxman & Namy, 1997). Each
participant completed three arrangement trials, one for each prompt. The order of the prompts
was counterbalanced across participants.
Following the prompt, the experimenter placed one of the 15 stimulus cards on the lower
right central square of the game board, and then named and presented each of the remaining
cards for the participant one-by-one to place on the board. Each arrangement trial presented
stimulus cards in a different, pre-determined, pseudo-random order in which no more than two
organisms from the same taxonomic or thematic group were presented consecutively. During a
given trial, participants were allowed to move cards that they had placed earlier in the trial. At
the conclusion of each trial, the experimenter photographed the board to record the locations of
the cards, then removed the cards before starting the next trial (see Figure 2 for examples of
participant-produced boards). The average duration of each trial was approximately 4 minutes.
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 18
Figure 2. Example of a participant-produced board for each age group.
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 19
The photographs taken following each arrangement trial were scored by treating the
10x10 board as a coordinate plane, identifying the coordinates of each card, and calculating the
Euclidean distance between the points specified by the coordinates. The range of possible
distances was 1 (cards adjacent along a horizontal row or vertical column) to 12.73 (cards on
diagonally opposite corners of the board).
Perception of Relatedness Analyses
To determine the extent to which participants were sensitive to different types of
relationships, each of the pairwise distances between the 15 stimuli (121 in total) were assigned
to one of three within-subjects Prompt conditions (i.e., Same Kind, Go Together, and Match),
and one of four within-subjects Relationship conditions (i.e., Both-taxonomic-and-thematic,
Taxonomic, Thematic, and Unrelated). The Relationship condition assignment was determined
on the basis of the allocation of stimuli to groups given in Table 1. These pairwise distance data
were submitted to ANOVAs to assess the effect of Prompt and Relationship on the whole sample
and on each age group. Bonferroni-corrected pairwise comparisons were used to further explore
To determine whether Prompt wording influenced participants’ relatedness judgments,
the pairwise distance data for each age group were submitted to a 3x4 Repeated Measures
ANOVA, with within-subjects factors of Prompt and Relationship. These analyses revealed no
main effect of Prompt nor significant interaction between Prompt and Relationship for any of the
age groups (all Fs<2.7, all ps>.1). A subset of the data consisting of only the pairwise distances
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 20
for each prompt when it was used for the first arrangement trial were submitted to the same 3x4
Repeated Measures ANOVA, which yielded comparable results (all Fs<1.3, all ps>.3).
Therefore, all subsequent analyses were collapsed across the Prompt types.
The pairwise distance data were then submitted to a 4x5 Mixed ANOVA, with one
within-subjects factor of Relationship, and one between-subjects factor of age group. Mauchly’s
test indicated that the assumption of sphericity had been violated for the Relationship factor
(χ2(5) = 65.30, p < .001), therefore degrees of freedom were corrected using Greenhouse-Geisser
estimates of sphericity (ε = 0.753). This analysis revealed an overall main effect of Relationship,
F(2.27,188.41)=120.21, p<.001, ηp2= .59 (MBoth=2.66, MTax=3.26, MThem=3.49, MUnrelated=3.81),
no main effect of Age, F(4,83)=1.35, p=.26 (MPre-K=3.37, MK=3.71, M1st=3.65, M2nd=3.32,
MAdult=3.68), and a significant interaction between Relationship and Age, F(9.08,188.41)=10.71,
p<.001, ηp2=.34. Pairwise comparisons between Relationship conditions revealed that all
conditions were significantly different from each other (all ps<.05).
To determine the influence of relationship in each age group, pairwise distance data for
each age group were then submitted to separate Repeated Measures ANOVAs, with Relationship
as the within-subjects factor (see Figure 3). The effect of Relationship was significant in all age
groups, all Fs>7.52, ps<.01 (ηp2 for Preschool, Kindergarten, 1st Grade, 2nd Grade, and
Adults=.27, .56, .59, .63, and .82, respectively). Pairwise comparisons were used to further
explore these differences. Adults differentiated between pairs of items from all four Relationship
conditions, with Both-taxonomically-and-thematically-related pairs placed closest
(Mdistance=2.16), followed by Taxonomically-related pairs (Mdistance=2.85), then Thematically-
related pairs (Mdistance=3.76), with Unrelated pairs placed farthest from each other (Mdistance=
4.27), all ps<.01.
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 21
In contrast, Preschoolers placed Both-taxonomically-and-thematically-related pairs
(Mdistance=2.96) closer together than all other pairs of items (Mdistances= 3.40, 3.33, 3.42 for
Thematic, Taxonomic, and Unrelated pairs, respectively), all ps<.05, but did not differentiate
between pairs from any other conditions, all ps>.9. Kindergarteners only placed Both-
taxonomically-and-thematically-related pairs (Mdistance= 3.15) closer than Unrelated items
(Mdistance= 3.85), p<.05. First and Second Grade students showed an intermediate pattern in
which they placed Both taxonomically- and thematically-related pairs (Mdistances= 2.72 and 2.38)
closer than all other pairs, all ps<.001, and placed Taxonomically-related pairs (Mdistances= 3.47
and 2.91) closer than Unrelated items (Mdistances= 3.90 and 3.61), all ps<.05. For Second Grade
students, the differences between Taxonomic and Thematic pairs (Mdistance= 3.36) and between
Thematic and Unrelated pairs were marginally significant, ps=.07 and .06, respectively.
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 22
Overall, these analyses suggest that early in development, semantic organization captures
links between entities that are related along both taxonomic and thematic dimensions, but does
not differentiate between concepts that are linked by a single relationship type versus unrelated
concepts. However, with development, semantic organization becomes further differentiated
according to links between concepts that are related along a single dimension, and according to
the specific type of relation linking concepts.
Taxonomic & Thematic Thematic
Figure 3. Mean distances between pairs of items in each Relationship condition for each age
group in Experiment 1. Error bars represent standard error of the mean.
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 23
Structure Derivation Analyses
To explore the latent structure underpinning organization of semantic knowledge for the
organisms participants judged, we applied average linkage hierarchical clustering (Hartigan,
1975) to the pairwise distances between the 15 items for each age group. This form of analysis
assigns items that are linked by short distances to local clusters that are in turn embedded within
higher-order clusters of clusters. Similar to the sensitivity analyses, we observed no systematic
differences between the hierarchical clustering solutions for each type of Prompt for any age
group. Therefore, we averaged the pairwise distances between the 15 items across all three
Prompts and all participants from a given age group to produce a single distance matrix for each
age group, which we then submitted to hierarchical cluster analysis.
The resultant hierarchical clustering solutions are presented as a dendrogram for each group
in Figure 4. Dendrograms are tree-like structures, composed of nodes (represented by vertical
lines), and arcs linking these nodes (represented by horizontal lines). The 15 items appear as the
terminal nodes of each dendrogram. These terminal nodes are connected by arcs to form internal
nodes that group the 15 items into hierarchical clusters. The height of the nodes captures both the
similarity of items within a cluster and the distinctiveness of items in different clusters. Internal
nodes at lower heights reflect greater similarity.
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 24
Taxonomic & Thematic Thematic
Figure 4. Dendrograms for each age group. Colors indicate the type of relationship linking
organisms within a cluster.
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 25
Qualitative assessment of these dendrograms reveals dramatic age-related changes. The
dendrogram for adult participants is highly organized at all hierarchical levels. At the highest
level, the items are split according to a plants-versus-animals distinction, and within the animals
cluster, items are further split according to a mammals-versus-birds distinction (although Whale
violates this distinction, and is clustered thematically with Penguin). Within these three
taxonomic groups, items are grouped into local clusters of both taxonomically- and thematically-
related items. In contrast, dendrograms for the Preschool participants primarily contain local
clusters of items that are both taxonomically- and thematically-related and are in turn embedded
within idiosyncratic higher-order clusters that do not reflect any readily identifiable relationships.
Higher-order structure begins to emerge in the dendrogram for Kindergarten participants, in
which both local and higher-order clusters are governed by both thematic and idiosyncratic
relationships. The organization of the dendrogram for First Grade students is more principled:
The influence of links between taxonomically- and thematically-related items is evident in local
clusters, and the influences of purely taxonomic and purely thematic links are evident in both
local and higher-order clusters. This trend towards principled organization at multiple levels
becomes more apparent in the dendrogram for Second Grade students, which is first split into
two higher-order clusters that capture a plants-versus-animals distinction. Within the plants
cluster, items are organized into local clusters of both taxonomically- and thematically-related
items. The animals cluster includes a thematic water cluster (Whale, Penguin, and Beaver), a
taxonomic bird cluster (Turkey, Ostrich, Eagle), and clusters of both taxonomically- and
To quantitatively assess developmental changes apparent in the dendrograms, the
hierarchical clustering solution for each age group was compared to the solution for each other
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 26
age group following the Fowlkes and Mallows (1983) method. In this method, each solution to
be compared is cut at multiple clustering levels to produce a set of flat partitions that contain 2 to
n-1 clusters, where n is the number of items. The two solutions are then compared in terms of
their similarity at each partition level k, where k is the number of clusters in the partitioning, and
similarity is the degree to which the items are partitioned into the same clusters. The result of this
comparison is the B(k) statistic, which ranges from 0 (minimally similar) to 1 (maximally
similar). Values for B(k) can be computed both for the comparison itself, and to determine the
range of B(k) that would be expected if the similarities between partitions were produced by
chance (see formulae 2.10 and 2.11 in Fowlkes & Mallows, 1983). When the observed value of
B(k) falls outside this range for a given partition level, one can conclude that the hierarchical
clustering solutions are significantly similar at that level. The outcomes of this analysis are
summarized in Table 2.
The findings summarized in Table 2 reveal that the number of levels on which clustering
solutions were similar decreased with increasing distance in age between the age groups
compared. These results present a pattern of gradual developmental change in the structure of
semantic knowledge from Pre-Kindergarten to Adulthood. Overall, these results corroborate the
Proportion of levels on which dendrograms are similar.
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 27
age-related changes observed in the qualitative interpretation of the dendrograms and the
outcome of the perception of relatedness analyses.
Taken together, the results of the sensitivity and structure derivation analyses reveal
several developmental changes. Both analyses provide converging evidence that: 1) Whereas
preschool-age children appear to only recognize links between organisms that are related along
multiple dimensions, older children increasingly recognize links between organisms that are
related along one dimension, and 2) Starting at approximately 2nd grade-age, taxonomic relations
are prioritized over thematic relations. The structure derivation analysis additionally
demonstrated that higher-order clusters become increasingly governed by conventional relations.
Specifically, whereas the clusters identified for the youngest age group lacked higher-order
structure, clusters identified for older age groups were increasingly organized at higher-order
levels according to taxonomic and/or thematic relations.
Due to the novelty of CC-SpAM, possible effects of characteristics of this paradigm on
children’s judgments must be considered. For instance, although we selected images with the
intention of reducing correspondence between visual similarity and relatedness, and validated
our selections using SIFT analysis (D. G. Lowe, 1999), it is possible that presence of images
rendered perceptual features more accessible and affected participants responses.
Another possible concern is that developmental patterns observed in this study stemmed
from young children failing to understand the instructions or apply their knowledge to the task
when asked to arrange a whole set of cross-classifiable organisms, rather than from changes in
the organization of semantic knowledge. This possibility seems unlikely because young
children’s responses were not random, as one might expect if children did not understand the
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 28
instructions. Instead, the analyses showed interpretable patterns in the responses of 4-year-old
children. Still, task demands remain a potential concern for Experiment 1.
To address these issues, we conducted Experiment 2 to assess semantic organization in
the same age groups using an alternate version of the spatial arrangement paradigm. Specifically,
participants made relatedness judgments of individual pairs of organisms that were represented
using unpainted wooden blocks. The use of blocks to represent organisms eliminates any
influence of immediately perceptible characteristics on relatedness judgments. Because
participants only judged a single pair of organisms at a time, Experiment 2 eliminated cognitive
demands related to arranging a full set of items. We additionally selected new organisms for
Experiment 2, to verify that the patterns observed in Experiment 1 can be generalized within the
domain of living things, and are not stimulus-specific.
The sample included 12 participants in each of five age groups: Preschool (Mage=4.50
years, SD=0.20; 7 female), Kindergarten (Mage=5.93 years, SD=0.38; 5 female), First Grade
(Mage=6.96 years, SD=0.61; 6 female), Second Grade (Mage=8.00 years, SD=0.41; 5 female), and
undergraduate Adults. Participants were recruited from the same populations as in Experiment 1.
Materials for this experiment included the same 10x10 grid as in Experiment 1, and two
wooden blocks. The blocks were used to represent the set of organisms listed in Table 3. This set
consisted of six Target organisms, and, for each Target organism, four organisms with which it
was paired during the experiment: A Taxonomically-related organism, a Thematically-related
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 29
organism, a Both-taxonomically-and-thematically-related organism, and an Unrelated organism.
These four types of pairs constituted four Relationship conditions that were analogous to the
Relationship conditions in Experiment 1, but included some new stimuli to test the
generalizability of the patterns observed in Experiment 1 to new items
. The resultant 24 pairs
were supplemented with four Filler pairs consisting of different organisms that bore no clear
taxonomic or thematic relationship with each other and were used to encourage children to use
the full grid (see Procedure section).
The assignment of organisms to these pairs was based on the results of the calibration
questionnaire administered in Experiment 1 in which adult participants rated the association
strength of a set of plants and animals with three thematic groups on a 4-point Likert scale.
Specifically, each Target and its Thematically-related and Both-taxonomically-and-thematically-
related items received a rating of ≥ 3 for the same thematic group (M=3.68). At the same time,
the Taxonomically-related and Unrelated items for a given Target received a rating of<2.5 for
the relevant thematic group (M=1.67).
Because it was difficult to find a sufficient number of new Wild/Zoo organisms familiar to children to include as items in this
set of stimuli, this thematic group was eliminated from Experiment 2.
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 30
Participants were tested individually in a quiet space. The task and procedures were
similar to those used in Fisher et al. (2015; 2014). To begin, participants were told that they were
going to play a game in which their job was to help a fictional character, Zibbo, organize his
favorite things, and the instruction that Zibbo wanted to, “Put things that are the same kinds of
things or that go together close together, and things that don’t go together or are not the same
kind of thing far apart”. These instructions were repeated again after the participant had
completed one half of the arrangement trials, with the order of the terms “go together” and “same
kinds of things” reversed. These terms were used to avoid biasing participants towards making
judgments based on a specific relationship type.
Following these instructions, participants were presented with trials in which they made
relatedness judgments of the Relationship pairs (Table 3) and Filler pairs. During each trial, the
experimenter first labeled one block as one of the organisms in the pair and placed it on one of
the four central squares of the grid, then labeled the second block as the other item in the pair and
gave it to the participant to place. Taking the grid as a 10x10 Cartesian plane, the central squares
corresponded to coordinates (5,5), (5,6), (6,5), and (6,6). The central square used for each trial
Organisms represented in Experiment 2. Each target organism was paired with a taxonomic
match, a thematic match, a both taxonomic and thematic match, and an unrelated organism.
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 31
was randomized across trials such that each square was used with equal frequency. The order in
which the Relationship pairs were presented was pseudo-randomized such that pairs in the same
Relationship condition did not appear more than twice consecutively. Within the Relationship
pairs, the order in which the Target (e.g., chicken) and the organism with which it was paired
(e.g., penguin, goat, turkey or tulip) were used to label the blocks was counterbalanced such that
the Target was used to label the first and second block an equal number of times. For instance,
the Target organism chicken was used twice to label the block that the experimenter placed, and
twice to label the block that the participant placed.
The procedure for Filler pair trials was identical, with the exception that the experimenter
placed the first block on one of the squares on or near the periphery of the grid. This placement
was designed to encourage participants to use the full grid. Filler pair trials were approximately
evenly interspersed among Relationship pair trials. For all trials, once the participant had placed
the second block on the grid, the experimenter recorded its location and removed both blocks
from the grid. The order of trials was counterbalanced, such that half the participants received
the trials in reverse order.
The outcome measure for this experiment was the Manhattan distance between the two
wooden blocks in each Relationship trial (i.e., the number of grid squares between two blocks
measured along horizontal rows and vertical columns). As in Experiment 1, these distances were
assigned to one of four Relationship conditions: Taxonomically-related, Thematically-related,
Both-taxonomically-and-thematically-related, and Unrelated. These data were then submitted to
ANOVAs to assess the effect Relationship on each age group. Bonferroni-corrected pairwise
comparisons were used to further explore condition differences.
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 32
The data for each age group were submitted to a Repeated Measures ANOVA, with one
within-subjects factor of Relationship. The effect of Relationship was significant in all age
groups (Fs>10.4, ps<.0001), with the exception of the Pre-Kindergarten age group (F=.992,
p=.409). Distances between pairs of organisms in each Relationship condition in each Age
Group are depicted in Figure 4. Pairwise comparisons were used to further explore the effect of
Relationship in the age groups in which it was significant.
Adults differentiated between pairs of items from all four Relationship conditions, with
Both-taxonomically-and-thematically-related pairs placed closest (Mdistance= 3.14), followed by
Taxonomically-related pairs (Mdistance=5.00), then Thematically-related pairs (Mdistance=6.32),
with Unrelated pairs placed farthest from each other (Mdistance= 8.35), all ps<.05. In contrast,
Kindergarteners only placed Both-taxonomically-and-thematically-related pairs (Mdistance=3.99)
closer than Taxonomically-related (Mdistance=6.40) and Unrelated pairs (Mdistance=5.90), ps <.05.
First-graders differentiated between Both-taxonomically-and-thematically-related pairs
(Mdistance=4.53) and all other pairs (Mdistance for Taxonomic, Thematic, and Unrelated=6.93, 6.13,
and 7.43, respectively), and between Thematically-related pairs and Unrelated pairs, ps<.01.
Finally, Second-graders differentiated between Both-taxonomically-and-thematically-related
pairs (Mdistance=3.49) from all other pairs (Mdistance for Taxonomic, Thematic, and Unrelated
=5.39, 5.33, and 7.06, respectively), and Taxonomically-related pairs and Thematically-related
pairs from Unrelated pairs, ps<.05.
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 33
The results of Experiment 2 corroborate the main findings of Experiment 1: organisms
that are related along multiple dimensions are differentiated from other organisms from an earlier
age than organisms related along either taxonomic or thematic dimensions alone. The
comparability between the patterns observed in both experiments suggests that this observed
trajectory is not due to task-specific characteristics, such as the way in which organisms are
represented (via pictures vs. labels), the way in which relatedness judgments are made (by
arranging a full set of cross-classifiable organisms together vs. by judging individual pairs), or
the specific organisms that participants judged.
Taxonomic & Thematic Thematic
Figure 4. Mean distances between pairs of items in each Relationship condition for each age
group. Error bars represent standard error of the mean.
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 34
The purpose of the present experiments was to use CC-SpAM, a novel variant of the
Spatial Arrangement Methodology (Goldstone, 1994; Hout et al., 2012) to shed new light on
developmental changes in the semantic organization of related concepts. By collecting spatial
relatedness judgments of a full set of cross-classifiable items, this methodology has multiple
advantages over prior approaches. First, this methodology can reveal the extent to which a given
concept is integrated into a network of other concepts. Second, it allows participants to make
graded judgments of the degree to which items are related, rather than a binary judgment that
forces participants to decide that items are either related or unrelated. Third, the pre-selection of
a set of cross-classifiable items overcomes any biases that may be introduced by paradigms in
which participants spontaneously produce items themselves, such as the tendency for participants
in semantic fluency studies to primarily produce mammals. Finally, the selection of cross-
classifiable items also affords an investigation of the role of different relations and their overlap
in semantic organization.
In Experiment 1, this paradigm was implemented by asking children from four age
groups (from Pre-Kindergarten to Second Grade) and adults to judge the relatedness of a set of
cross-classifiable organisms. This set included organisms that were either both- taxonomically-
and thematically-related, taxonomically- or thematically-related only, or unrelated. Experiment 2
was designed to validate the results of Experiment 1 by eliminating any cognitive demands
placed by asking participants to judge a full set of items together, eliminating any influence of
immediately perceptible visual similarity by representing items using wooden blocks, and
collecting relatedness judgments for new organisms. These manipulations introduced several
procedural differences between Experiments 1 and 2, which allowed us to test whether the
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 35
patterns observed using CC-SpAM in Experiment 1 would also manifest under different testing
Across both studies, the earliest evidence for semantic knowledge organization took the
form of a distinction between organisms linked by overlapping taxonomic and thematic relations
versus unrelated organisms. In contrast, adults exhibited a highly principled pattern of
relatedness judgments based on both the number and type of relations between organisms.
Specifically, adults judged organisms linked by both taxonomic and thematic relations to be
more closely related than unrelated organisms and those linked by either type of relation alone,
organisms linked by either type of relation to be more closely related than unrelated organisms,
and organisms linked by taxonomic relations to be more closely related than thematically-related
organisms (see Wisniewski & Bassok, 1999, for similar patterns in adults). Organization based
on either taxonomic or thematic relations alone appeared to emerge over the course of early
school years. These results suggest that young children can capitalize on the fact that things that
are of the same “kind” may also be associated with the same environment as they are learning to
differentiate between related versus unrelated entities. With age, children demonstrate an
increasing ability to recognize links between entities that are based on a single type of relation.
The comparability of trajectories observed across Experiments 1-2 suggests that the CC-
SpAM paradigm provides a valid assessment of developmental changes in semantic organization
that is not skewed by task demands, the way in which items are represented, or specific items.
Moreover, these findings corroborate a small body of prior evidence for a role of overlapping
taxonomic and thematic associations in the development of semantic organization, which are
discussed in the following section.
Role of Overlapping Relations in the Development of Semantic Organization
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 36
The results of the present experiments are consistent with a handful of prior findings that
recognition of links between entities that are both taxonomically- and thematically-related
precedes recognition of taxonomic relations alone. First, match-to-sample studies that have asked
children to justify their choices found that children often provide thematic justifications for
taxonomic matches (Hashimoto et al., 2007; Tversky, 1985; Walsh et al., 1993). For instance, in
Tversky’s (1985) study, children who successfully identified a taxonomic match (e.g., saw–
screwdriver) nonetheless often provided thematic justifications for their choices (e.g., “The saw
saws the screwdriver”). These justifications suggest that thematic association may contribute to
young children’s recognition of taxonomic links.
More direct evidence for the role of taxonomic and thematic overlap comes from a recent
study conducted by Blaye et al. (2006), in which children were given items sorted into
taxonomic groups, and asked whether new candidate items that either were or were not
taxonomically-related to items in these groups could be incorporated into them. Both sets of
taxonomically-related and unrelated candidate items included a subset of items that were
perceptually similar or thematically associated with items within the groups. At age five,
children tended to incorporate items that were perceptually similar or thematically associated
with items in a group, and reject those that were not, regardless of taxonomic relatedness. Use of
taxonomic relatedness as the criterion for incorporation into taxonomic groups gradually
increased from age six to age ten. This developmental pattern suggests that the recognition of
taxonomic relations is only gradually differentiated from the recognition of other, overlapping
forms of relatedness over the course of development.
Results consistent with this interpretation also come from list memory studies, in which
participants are asked to recall lists of words that are unrelated, linked by both taxonomic and
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 37
thematic relations, or linked by taxonomic relations alone. Whereas children recall lists of words
that are both taxonomically- and thematically-related better than lists of unrelated words from
early childhood onward, better recall for lists of words that are only taxonomically-related versus
unrelated words emerges only later in childhood (Bjorklund & Jacobs, 1985; Monnier &
The present studies expanded upon this evidence by directly assessing children’s graded
judgments of relatedness of different entities. Moreover, the present findings provide the first
indication that overlapping taxonomic and thematic associations may also play a role in the
development of thematic differentiation. The following sections discuss the implications of these
findings for models of semantic development, and questions that remain open following this
Implications for Models of Semantic Knowledge Organization
Existing computational models designed to predict learning-related changes in semantic
organization examine how the structure of a simulated semantic network consisting of
representations of entities linked by shared features changes either with increasing exposure to
entities, or as new features of entities are learned. In many such models, features are
characteristics identified by researchers or adult participants as associated with each entity.
Although such features can refer to a range of characteristics, researchers have typically included
perceptual features, such as is black and has four legs, and excluded “encyclopedic” features
referring to facts about entities, such as is expensive. However, other types of incorporated
features vary, and include functional (e.g., is used for transport, Hills et al., 2009), behavioral
(e.g., barks, Kemp & Tenenbaum, 2008), and intrinsic (e.g., is alive, Rumelhart & Todd, 1993)
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 38
Despite variations in the features included in different models, these models undergo
broadly similar changes across simulated developmental time. The earliest form of organization
to emerge during learning consists of global groups that are poorly internally differentiated,
followed by internal refinement of these groups. For example, given a small amount of exposure
to a set of organisms, Rumelhart’s (Rumelhart & Todd, 1993) connectionist model identified a
global division between plants and animals. With further exposure, these groups became
internally differentiated into groups of different kinds of plants and animals (see also,
McClelland & Rogers, 2003). In a related study, Hills et al. (2009) analyzed noun-feature
networks for a set of early-acquired nouns. When nouns in the network were linked on based on
only one common feature, semantic organization included a single, densely interconnected
cluster. As the number of shared features increased, semantic space progressed from divisions
into broad groups to divisions into increasingly fine-grained groups. Similar patterns have
emerged in Bayesian models (Kemp & Tenenbaum, 2008) and models given perceptual features
only (Quinn & Johnson, 2000).
The developmental trajectory predicted by these models and the trajectory observed in
the present studies converge in some respects, and diverge in others. These patterns are best
illuminated by inspection of the dendrograms depicted in Figure 4. Consistent with model
simulations, a global division between plants and animals did emerge prior to a finer-grained
division of the animals group into mammals and birds. However, the influence of both
taxonomic and thematic relations and their overlap resulted in a more complex developmental
progression. First, the plant-animal division only emerged after a transition from divisions
between organisms related along multiple dimensions to divisions between organisms related
along a single dimension. Second, thematic associations continued to influence semantic
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 39
knowledge differentiation during the progression from global to increasingly fine-grained
groups. For instance, the plants cluster evident in Second Grade was internally differentiated into
farm plants and water plants groups.
This divergence between predicted trajectories and observed developmental patterns may
have resulted from the fact that computational models created to date simulate developmental
change purely based on learning features. This type of input supports differentiation between
groups of entities that share features with each other and are therefore taxonomically-related, but
is not relevant to other types of relations, such as thematic relations. Given sources of
information relevant to different types of relations, children may recognize links between entities
that are related along multiple dimensions before discovering links between entities related along
a single dimension, as the present results suggest. Further research investigating changes in
organizational structure simulated by computational models given multiple types of relations
would be necessary to shed light on this possibility.
The novel approach to studying semantic development implemented in this study opens
up several avenues for future research. First, this study used hierarchical clustering to provide
continuity with prior semantic organization development studies (Crowe & Prescott, 2003;
Storm, 1980), such that any deviations between the findings of this and previous studies are
attributable to methodological rather than analysis differences. However, hierarchical clustering
constrains the relations that can be identified within the resultant structure because the use of a
specific distance threshold level necessarily partitions items into non-overlapping clusters. As the
distance level is increased, these non-overlapping clusters can be merged, so that individual
items can belong to a hierarchical set of clusters. For instance, chicken can be clustered with
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 40
birds, and birds in turn can be hierarchically clustered with other animals. However, chicken
cannot belong to both a birds cluster and a farm cluster, even if participants placed the chicken
card closer to both bird cards and farm cards than to unrelated cards. Therefore, structure
derivation analyses that allow items to belong to multiple overlapping groups may provide a less
constrained description of semantic knowledge organization development. Additive clustering,
an analysis approach that is currently in active development (e.g., Lee, 2002), has potential to
address this issue. In this approach, any concept may be assigned a number of cluster
memberships. For example, using additive clustering, we might find that chicken belongs to both
a cluster that includes other farm organisms, and a cluster that includes other birds. We can then
examine how these cluster memberships evolve with age, which could provide further insight
into the development of semantic organization. However, the fact that the refinement of additive
clustering algorithms is currently ongoing renders it an approach that should be explored in
Second, although we explored variability in relatedness judgments due to task demands
by investigating the effects of different arrangement prompts in Experiment 1, sources of
variability in relatedness judgments may be explored further in several ways. For instance,
although we did not find an effect of manipulating arrangement prompts, stronger manipulations
such as repeating prompts multiple times or demonstrating a given type of relationship at the
beginning of a trial may systematically affect relatedness judgments. If such effects are found, it
would suggest that young children can make relatedness judgments purely on the basis of either
taxonomic or thematic relationships when strongly prompted to do so, but not when making
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 41
Additionally, individual differences between participants’ judgments were not explored
in our experiments. For instance, it is possible that some younger participants were sensitive to
relationships to which their age group appeared insensitive, or that some older participants were
insensitive to relationships to which their age group appeared sensitive. Exploring the sources of
such potential individual differences, such as the types of prior experience that shape semantic
knowledge organization, would provide an illuminating avenue of future research.
Finally, although taxonomic and thematic relations studied here are the most extensively
discussed and investigated relations in cognitive development literature, they are only two of a
range of relation types. For instance, some items used in the present experiments could also be
linked according to ecological or predator-prey relations. The apparent lack of influence of these
relations on participants’ relatedness judgments does not indicate that these relations do not
influence semantic organization. Instead, the fact that items used in these experiments were
thoroughly cross-classifiable by taxonomic and thematic relations may have constrained the
range of relationships that influenced participant judgments. As Murphy (2001) demonstrated,
the presence of items within a set that can be grouped according to a specific relationship
prompts participants to sort all items according to that relationship. Moreover, participants may
only consider relationships other than taxonomic and thematic when specifically prompted by a
context that renders these alternate relationships salient. Accordingly, the course of semantic
organization development may entail a wider variety of changes than those revealed in the
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 42
The present study introduced a novel paradigm for investigating developmental changes
in the semantic organization of concepts that can be linked by multiple cross-cutting
relationships. In contrast with previous research that has investigated whether children prefer or
recognize individual types of relationships, the present study revealed a role for overlapping
relationships in the development of semantic organization. Specifically, the developmental
trajectory observed suggests that semantic knowledge organization progresses from an early
recognition of links between entities that are both taxonomically- and thematically-related, to an
increasing recognition of links based on a single type of relation. The influence of overlapping
relations indicates that current models of semantic organization development, which only
incorporate links between entities based on feature overlap, should be expanded to additionally
incorporate links based on thematic association.
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 43
Anderson, J. R., & Gordon, B. H. (1973). Human associative memory. Washington DC:
Belke, E., Humphreys, G. W., Watson, D. G., Meyer, A. S., & Telling, A. L. (2008). Top-down
effects of semantic knowledge in visual search are modulated by cognitive but not
perceptual load. Perception & Psychophysics, 70(8), 1444-1458.
Bjorklund, D. F. (1987). How age changes in knowledge base contribute to the development of
children's memory: An interpretive review. Developmental Review, 7(2), 93-130.
Bjorklund, D. F., & Jacobs, J. W. (1985). Associative and categorical processes in children's
memory: The role of automaticity in the development of organization in free recall.
Journal of Experimental Child Psychology, 39(3), 599-617.
Blaye, A., Bernard-Peyron, V., Paour, J.-L., & Bonthoux, F. (2006). Categorical flexibility in
children: Distinguishing response flexibility from conceptual flexibility. European
Journal of Developmental Psychology, 3, 163-188.
Carey, S. (1985). Conceptual change in childhood. Cambridge, Massachusetts: MIT Press.
Collins, A. M., & Loftus, E. F. (1975). A spreading-activation theory of semantic processing.
Psychological review, 82(6), 407.
Cree, G. S., & McRae, K. (2003). Analyzing the factors underlying the structure and
computation of the meaning of chipmunk, cherry, chisel, cheese, and cello (and many
other such concrete nouns). Journal of Experimental Psychology: General, 132(2), 163.
Crowe, S. J., & Prescott, T. J. (2003). Continuity and change in the development of category
structure: Insights from the semantic fluency task. International Journal of Behavioral
Development, 27(5), 467-479.
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 44
Deák, G., & Bauer, P. J. (1995). The Effects of Task Comprehension on Preschoolers′ and
Adults′ Categorization Choices. Journal of Experimental Child Psychology, 60(3), 393-
Entwisle, D. R. (1966). Form class and children's word associations. Journal of verbal learning
and verbal behavior, 5(6), 558-565.
Entwisle, D. R., Forsyth, D. F., & Muuss, R. (1964). The syntactic-paradigmatic shift in
children's word associations. Journal of Verbal Learning and Verbal Behavior, 3(1), 19-
Fisher, A. V., Godwin, K. E., & Matlen, B. J. (2015). Development of inductive generalization
with familiar categories. Psychonomic bulletin & review, 1-25.
Fisher, A. V., Godwin, K. E., Matlen, B. J., & Unger, L. (2014). Development of
Category‐Based Induction and Semantic Knowledge. Child Development.
Fowlkes, E. B., & Mallows, C. L. (1983). A method for comparing two hierarchical clusterings.
Journal of the American Statistical Association, 78(383), 553-569.
Goldstone, R. (1994). An efficient method for obtaining similarity data. Behavior Research
Methods, Instruments, & Computers, 26(4), 381-386.
Hartigan, J. A. (1975). Clustering Algorithms. New York, NY: John Wiley & Sons.
Hashimoto, N., McGregor, K. K., & Graham, A. (2007). Conceptual organization at 6 and 8
years of age: Evidence from the semantic priming of object decisions. Journal of Speech,
Language and Hearing Research, 50(1), 161-176.
Hills, T. T., Jones, M. N., & Todd, P. M. (2012). Optimal foraging in semantic memory.
Psychological review, 119(2), 431.
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 45
Hills, T. T., Maouene, M., Maouene, J., Sheya, A., & Smith, L. (2009). Categorical structure
among shared features in networks of early-learned nouns. Cognition, 112(3), 381-396.
Hills, T. T., Todd, P. M., & Jones, M. N. (2015). Foraging in semantic fields: how we search
through memory. Topics in cognitive science, 7(3), 513-534.
Hout, M. C., Goldinger, S. D., & Ferguson, R. W. (2012). The versatility of SpAM: A fast,
efficient, spatial method of data collection for multidimensional scaling. Journal of
Experimental Child Psychology: General, 142(1), 256-281.
Howard, D. V., & Howard, J. H. (1977). A multidimensional scaling analysis of the development
of animal names. Developmental Psychology, 13(2), 108.
Kemp, C., & Tenenbaum, J. B. (2008). The discovery of structural form. Proceedings of the
National Academy of Sciences, 105(31), 10687-10692.
Lee, M. D. (2002). A simple method for generating additive clustering models with limited
complexity. Machine Learning, 49(1), 39-58.
Lin, E. L., & Murphy, G. L. (2001). Thematic relations in adults' concepts. Journal of
Experimental Psychology: General, 130(1), 3-28.
Lowe, D. G. (1999). Object recognition from local scale-invariant features. Paper presented at
the Proceedings of the seventh IEEE international conference on computer vision, Corfu.
Lowe, W. (2001). What is the Dimensionality of Human Semantic Space? Connectionist Models
of Learning, Development and Evolution (pp. 303-311): Springer.
Lucariello, J., Kyratzis, A., & Nelson, K. (1992). Taxonomic knowledge: What kind and when?
Child Development, 63(4), 978-998.
McClelland, J. L., & Rogers, T. T. (2003). The parallel distributed processing approach to
semantic cognition. Nature Reviews Neuroscience, 4(4), 310-322.
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 46
Monnier, C., & Bonthoux, F. (2011). The semantic‐similarity effect in children: Influence of
long‐term knowledge on verbal short‐term memory. British Journal of Developmental
Psychology, 29(4), 929-941.
Montez, P., Thompson, G., & Kello, C. T. (2015). The Role of Semantic Clustering in Optimal
Memory Foraging. Cognitive science.
Moores, E., Laiti, L., & Chelazzi, L. (2003). Associative knowledge controls deployment of
visual selective attention. Nature neuroscience, 6(2), 182-189.
Murphy, G. L. (2001). Causes of taxonomic sorting by adults: A test of the thematic-to-
taxonomic shift. Psychonomic Bulletin & Review, 8(4), 834-839.
Nelson, D., McEvoy, C., & Schrieber, T. (1998). The University of South Florida word
association, rhyme, and word fragment norms. Retrieved from
Nelson, K. (1977). The syntagmatic-paradigmatic shift revisited: a review of research and theory.
Psychological Bulletin, 84(1), 93.
Nguyen, S. P. (2007). Cross-classification and category representation in children's concepts.
Developmental Psychology, 43(3), 719-731.
Nguyen, S. P., & Murphy, G. L. (2003). An Apple is More Than Just a Fruit:
Cross‐Classification in Children's Concepts. Child Development, 74(6), 1783-1806.
Quinn, P. C., & Johnson, M. H. (2000). Global‐Before‐Basic Object Categorization in
Connectionist Networks and 2‐Month‐Old Infants. Infancy, 1(1), 31-46.
Roediger, H. L., & McDermott, K. B. (1995). Creating false memories: Remembering words not
presented in lists. Journal of experimental psychology: Learning, Memory, and
Cognition, 21(4), 803.
DEVELOPMENTAL CHANGES IN SEMANTIC ORGANIZATION 47
Rumelhart, D. E., & Todd, P. M. (1993). Learning and connectionist representations. Attention
and performance XIV: Synergies in experimental psychology, artificial intelligence, and
cognitive neuroscience, 3-30.
Smiley, S. S., & Brown, A. L. (1979). Conceptual preference for thematic or taxonomic
relations: A nonmonotonic age trend from preschool to old age. Journal of Experimental
Child Psychology, 28(2), 249-257.
Storm, C. (1980). The semantic structure of animal terms: A developmental study. International
Journal of Behavioral Development, 3(4), 381-407.
Tversky, B. (1985). Development of taxonomic organization of named and pictured categories.
Developmental Psychology, 21(6), 1111-1119.
Walsh, M., Richardson, K., & Faulkner, D. (1993). Perceptual, thematic and taxonomic relations
in children’s mental representations: Responses to triads. European Journal of
Psychology of Education, 8(1), 85-102.
Waxman, S. R., & Namy, L. L. (1997). Challenging the notion of a thematic preference in young
children. Developmental Psychology, 33(3), 555-567.
Winkler-Rhoades, N., Medin, D., Waxman, S. R., Woodring, J., & Ross, N. O. (2010). Naming
the animals that come to mind: effects of culture and experience on category fluency.
Journal of Cognition and Culture, 10(1), 205-220.
Wisniewski, E. J., & Bassok, M. (1999). What makes a man similar to a tie? Stimulus
compatibility with comparison and integration. Cognitive Psychology, 39(3), 208-238.