ArticlePDF Available

Abstract and Figures

The organization of our knowledge about the world into an interconnected network of concepts linked by relations profoundly impacts many facets of cognition, including attention, memory retrieval, reasoning, and learning. It is therefore crucial to understand how organized semantic representations are acquired. The present experiment investigated the contributions of readily observable environmental statistical regularities to semantic organization in childhood. Specifically, we investigated whether co-occurrence regularities with which entities or their labels more reliably occur together than with others: (1) Contribute to relations between concepts independently, and (2) Contribute to relations between concepts belonging to the same taxonomic category. Using child-directed speech corpora to estimate reliable co-occurrences between labels for familiar items, we constructed triads consisting of a target, a related distractor, and an unrelated distractor in which targets and related distractors consistently co-occurred (e.g., sock-foot), belonged to the same taxonomic category (e.g., sock-coat), or both (e.g., sock-shoe). We used an implicit, eye-gaze measure of relations between concepts based on the degree to which children (N=72, age 4-7 years) looked at related versus unrelated distractors when asked to look for a target. The results indicated that co-occurrence both independently contributes to relations between concepts, and contributes to relations between concepts belonging to the same taxonomic category. These findings suggest that sensitivity to the regularity with which different entities co-occur in children’s environments shapes the organization of semantic knowledge during development. Implications for theoretical accounts and empirical investigations of semantic organization are discussed.
Content may be subject to copyright.
1
This is the accepted version of the following article: Unger, L., Vales, C., & Fisher, A. V. (2020). The Role of
Co-Occurrence Statistics in Developing Semantic Knowledge. Cognitive Science, 44(9), e12894, which has
been published in final form at https://onlinelibrary.wiley.com/doi/full/10.1111/cogs.12894. This article may
be used for non-commercial purposes in accordance with the Wiley Self-Archiving Policy
[http://olabout.wiley.com/WileyCDA/Section/id-828039.html].
The Role of Co-Occurrence Statistics in Developing Semantic Knowledge
Layla Unger¹
Catarina Vales2
Anna V. Fisher2
¹ Department of Psychology, Ohio State University, Columbus OH
2Department of Psychology, Carnegie Mellon University, Pittsburgh PA
Acknowledgements
This work was supported by a Graduate Training Grant awarded to Carnegie Mellon University by the
Department of Education, Institute of Education Sciences (R305B040063), by NSF award # BCS-
1918259 to the second and third authors, and by the James S. McDonnell Foundation 21st Century
Science Initiative in Understanding Human Cognition Scholar Award (220020401) to the third author.
We thank Anna Vande Velde, Kristen Boyle, participating children, parents, and schools for their vital
contributions to this research.
Data Availability Statement
Raw data, processed data, and scripts for processing and analyzing data have been made available
through the Open Science Framework at https://osf.io/9de7p/.
2
Abstract
The organization of our knowledge about the world into an interconnected network of concepts linked
by relations profoundly impacts many facets of cognition, including attention, memory retrieval,
reasoning, and learning. It is therefore crucial to understand how organized semantic representations
are acquired. The present experiment investigated the contributions of readily observable
environmental statistical regularities to semantic organization in childhood. Specifically, we
investigated whether co-occurrence regularities with which entities or their labels more reliably occur
together than with others: (1) Contribute to relations between concepts independently, and (2)
Contribute to relations between concepts belonging to the same taxonomic category. Using child-
directed speech corpora to estimate reliable co-occurrences between labels for familiar items, we
constructed triads consisting of a target, a related distractor, and an unrelated distractor in which
targets and related distractors consistently co-occurred (e.g., sock-foot), belonged to the same
taxonomic category (e.g., sock-coat), or both (e.g., sock-shoe). We used an implicit, eye-gaze measure
of relations between concepts based on the degree to which children (N=72, age 4-7 years) looked at
related versus unrelated distractors when asked to look for a target. The results indicated that co-
occurrence both independently contributes to relations between concepts, and contributes to
relations between concepts belonging to the same taxonomic category. These findings suggest that
sensitivity to the regularity with which different entities co-occur in children’s environments shapes the
organization of semantic knowledge during development. Implications for theoretical accounts and
empirical investigations of semantic organization are discussed.
Keywords: semantic development; conceptual development; co-occurrence; taxonomic; semantic
organization; visual search
ROLE OF CO-OCCURRENCE IN SEMANTIC DEVELOPMENT 3
1. Introduction
Our knowledge about the world around us is not merely a repository for all the information we
have acquired. Instead, our knowledge forms a semantic network of concepts linked by
meaningful relations, such as taxonomic links between members of the same category (e.g.,
dog and elephant), and links between concepts that are associated thematically by virtue of
participating in the same events (e.g., dog and leash). Moreover, this organization according to
meaningful relations profoundly influences many higher cognitive functions, including memory
encoding and retrieval (Bower, Clark, Lesgold, & Winzenz, 1969), reasoning (Heit, 2000),
learning (Jimura, Hirose, Wada et al., 2016; Tse, Langston, Kakeyama et al., 2007), and visual
attention (Moores, Laiti, & Chelazzi, 2003). However, we are not born with organized networks
of semantic knowledge. How does how semantic knowledge become organized in the course of
development?
Decades of extensive research have revealed that children’s semantic knowledge becomes
gradually organized according to multiple meaningful relations between concepts (Blaye,
Bernard-Peyron, Paour, & Bonthoux, 2006; Fenson, Vella, & Kennedy, 1989; Lucariello, Kyratzis,
& Nelson, 1992; Tversky, 1985; Unger, Fisher, Nugent, Ventura, & MacLellan, 2016; Walsh,
Richardson, & Faulkner, 1993). Moreover, a number of findings suggest that developmental
changes in semantic organization are likely shaped by experience. For example, reliable
differences in semantic organization have been documented between younger and older
children (Carey, 1985; Unger et al., 2016), children who grow up in urban versus rural settings
(Coley, 2012), children who show a particular interest in a domain versus those who do not
(e.g., dinosaurs; Gobbo & Chi, 1986), and adults with different patterns of expertise in a domain
4
(e.g., arborists and landscape workers; Medin, Lynch, Coley, & Atran, 1997). Although these
findings are informative about the development of semantic organization, they also highlight a
critical, unresolved question: What sources of input in our experiences with the world shape
the development of semantic organization? The purpose of the present research is to examine
the contributions of one potentially key source of environmental input: The regularity with
which entities (or their labels) reliably co-occur.
In what follows, we first elaborate upon why accounting for the sources of environmental input
that contribute to the formation of links between concepts is critical for understanding the
development of semantic organization. We then review extant evidence that supports the
possibility that co-occurrence contributes to the development of semantic organization.
Specifically, we review evidence that: (1) Co-occurrence regularities that may contribute to
semantic organization development are ubiquitous in both language and visual input, (2) A
sensitivity to co-occurrence regularities is present in a variety of domains beginning early in
development, and (3) Co-occurrence regularities may have contributed to the developmental
trajectories of semantic organization documented in prior research. Finally, we present an
experiment designed to investigate the contributions of co-occurrence to semantic organization
during development.
1.1 Importance of Understanding the Contributions of Environmental Input
To study the development of semantic organization, researchers typically pursue an approach
in which they first identify sets of concepts as related according to some criteria, then assess
whether children exhibit behavior indicating that these concepts are related in their semantic
networks. For example, researchers may first identify sets of items that the researchers
5
themselves or independent adult raters judge to be taxonomically related, then test whether
children sort the sets into groups of taxonomically related items (e.g., Blaye et al., 2006), recall
lists of words comprised of taxonomically related items better than lists of unrelated words
(e.g., Monnier & Bonthoux, 2011), or choose a taxonomically related item as a “match” to a
target from a set of options (e.g, Fenson et al., 1989; Lucariello et al., 1992; Waxman & Namy,
1997). Researchers then draw inferences about the types of links that organize children’s
semantic knowledge based on children’s behavior, such as inferring that taxonomic relations
contribute to organizing children’s semantic knowledge if children sort items that adults
identified as taxonomically related into the same group. This approach has been employed
extensively in prior research to study the emergence of taxonomic relations in semantic
organization in particular because these relations have traditionally been held (e.g., Inhelder &
Piaget, 1964) to organize adult semantic networks and support a key facet of adult intelligent
behavior: The generalization of acquired knowledge to new entities (Heit, 2000). For example,
upon learning that robins have hollow bones, knowledge of taxonomic relations between birds
can support generalizing this knowledge about robins to other birds.
However, observing that children in a study treat items that adults judge to be related as
related tells us little about the sources of this behavior. Instead, such observations raise two key
questions. First, if children treat items as related because they understand that the items are
related in the same way that adults do, what is the source of this understanding? For example,
what drives children’s ability to sort taxonomically or thematically related items together?
Second, is it valid to conclude from these observations that children truly share adults’
understanding of semantic relatedness, or may children’s behavior be driven by some other
6
source of input? For example, if taxonomically related items are also linked by some other
source of environmental input, children’s behavior may be driven -- at least in part -- by this
input, without necessarily or uniquely reflecting the contribution of taxonomic knowledge.
Consider a trivial example of this issue: If all items from a given taxonomic category are of the
same color, a tendency for children to group together taxonomically related items could be
based either on an understanding of taxonomic relations, or on color.
Addressing both questions requires investigating sources of input that drive children’s
treatment of items as related. To date, a number of studies have made substantial inroads into
understanding one source of input that may contribute to links between taxonomically related
concepts in particular: Similarity in perceptual features, such as color and shape. This research
has revealed that perceptual similarity may both: (1) Serve as a basis for children’s treatment of
taxonomically related items as related in the absence of a true understanding of taxonomic
relatedness (e.g., Imai, Gentner, & Uchida, 1994; Sloutsky, Kloos, & Fisher, 2007), and (2) Act as
a foundation upon which taxonomic relations are built (Eimas & Quinn, 1994; Oakes, Coppage,
& Dingel, 1997; Quinn, Eimas, & Rosenkrantz, 1993; Quinn & Johnson, 2000). As we discuss in
the following section, the present research is designed to explore another non-trivial source of
environmental input that may drive the formation of links between concepts, including both
thematically related concepts, and concepts belonging to the same taxonomic category: The
7
regularity with which entities or their labels co-occur more reliably with each other than with
others
1
.
1.2 Potential Contributions of Co-Occurrence Regularities
In this section, we review three sources of evidence that underline the possibility that co-
occurrence regularities contribute to forming links between concepts. First, we present
evidence that co-occurrence regularities that can support the formation of semantic relations
are ubiquitous in environmental input, including both language and visual scenes (Asr, Willits, &
Jones, 2016; Frermann & Lapata, 2015; Hofmann, Biemann, Westbury et al., 2018; Huebner &
Willits, 2018; Jones & Mewhort, 2007; Landauer & Dumais, 1997; Rohde, Gonnerman, & Plaut,
2004; Sadeghi, McClelland, & Hoffman, 2015). Second, we underline the plausibility that these
ubiquitous co-occurrence regularities contribute to the development of semantic organization
by reviewing evidence that human learners are sensitive to these regularities from an early age
(e.g., Pelucchi, Hay, & Saffran, 2009; Saffran, Aslin, & Newport, 1996). Finally, we highlight how
co-occurrence regularities may have contributed to the findings from many prior investigations
into the development of semantic organization.
1.2.1 Co-Occurrence regularities are ubiquitous in environmental input. For co-occurrence
regularities to contribute to semantic organization in development, they must both: (1) Be
readily available in environmental input in general, and input to which young learners are
exposed in particular, and (2) Link items that are semantically related. Numerous investigations
1
Phrased statistically: The degree to which the rate of co-occurrence between a given pair of
entities/labels is higher than the rate with which they would co-occur if they occurred randomly in
input, given their respective frequencies.
8
of the regularities present in language and visual scenes suggest that co-occurrence regularities
meet both these criteria.
The richness of co-occurrence regularities in language input is primarily attested by evidence
from numerous computational models that form representations of words based on their
patterns of co-occurrence in language corpora. When given input that captures the language to
which either adults (e.g., Jones & Mewhort, 2007; Landauer & Dumais, 1997) or children (e.g.,
Asr et al., 2016; Huebner & Willits, 2018) are exposed, these models successfully capture much
of the links between words in human semantic networks and perform comparably to humans in
multiple semantic judgment tasks. This evidence demonstrates that co-occurrence regularities
both are ubiquitous in the environmental input available from early in development, and can
support the acquisition of links among semantically-related items. Moreover, recent modeling
research suggests that the same characteristics are true of co-occurrence regularities between
objects in the visual domain (Sadeghi et al., 2015).
This evidence from modeling research is corroborated by analyses of the factors that contribute
to semantic links between words in adults. Specifically, these analyses have shown that the co-
occurrence regularities in linguistic corpora account for a substantial amount of the variance in
the strength of semantic links (Hofmann et al., 2018; Spence & Owens, 1990). Taken together,
these findings highlight co-occurrence regularities as a source of information ubiquitous in
input that has the potential to contribute to the formation of semantic links.
1.2.2 Ability to learn co-occurrence regularities emerges early in development. For co-
occurrence regularities to contribute to semantic organization, it is necessary but not sufficient
9
that these regularities are present in the environment. In addition, young learners must also
possess an ability pick up on these regularities starting early in development.
As attested by findings from extensive research in the field of statistical learning, sensitivity to
the regularity with which separate perceptual inputs co-occur with each other more reliably
than with other inputs is evident in multiple domains from early infancy onwards. For example,
beginning in infancy, humans are sensitive to the regularity with which speech sounds (Pelucchi
et al., 2009; Saffran et al., 1996), tones (Saffran, Johnson, Aslin, & Newport, 1999), and visual
objects (Bulf, Johnson, & Valenza, 2011) sequentially co-occur in time, and with which visual
objects occur together in space (Fiser & Aslin, 2002). Given the evidence for this sensitivity
across many domains, these findings highlight the plausibility that sensitivity to co-occurrence
regularities between words or the objects they denote may contribute to the development of
semantic organization.
Although our understanding of the contribution of this sensitivity to semantic organization is
limited, findings from a handful of studies conducted by Fisher and colleagues (Fisher, 2010;
Fisher, Matlen, & Godwin, 2011; Matlen, Fisher, & Godwin, 2015) suggest that regular co-
occurrences between words do indeed contribute to the degree to which children treat the
entities they denote as semantically related. In these studies, when asked to decide which of
two “Match” entities shared a novel property (e.g., “plaxium blood”) with a “Target”, 4-year-old
children only reliably chose Match items that were taxonomically related to the Target when
their labels also co-occurred in either corpora of infant and child speech input (e.g., puppy-dog;
Fisher et al., 2011), or an experimentally manipulated auditory stream of words (Matlen et al.,
2015). In addition, findings from a study conducted by Wojcik and Saffran (2015) provide
10
evidence that toddlers similarly form links between words that co-occur in sentences. Taken
together, the evidence reviewed in this section underlines the possibility that exposure to co-
occurrence regularities that are ubiquitous in environmental input and link semantically related
items can contribute to the development of semantic organization.
1.2.3 Contribution of co-occurrence regularities to semantic organization. If co-occurrence
regularities contribute to the development of semantic organization, this contribution may be
apparent in the results of prior studies that have investigated semantic organization
development, even when these studies were not designed to uncover the contributions of co-
occurrence regularities. An examination of these prior results is indeed suggestive of co-
occurrence contributions to semantic organization.
First, many prior studies have yielded evidence that, starting early in development, children’s
semantic organization is shaped by thematic relations between items that play complementary
roles in commonly encountered scenarios (also referred to as schematic and script relations),
such as ‘dog’ and ‘bone’ or ‘lock’ and ‘key’(e.g., Fenson et al., 1989; Lucariello et al., 1992;
Walsh et al., 1993). Moreover, this behavior persists into adulthood (Lin & Murphy, 2001; Ross
& Murphy, 1999). The regularity with which items that play complementary roles or their labels
regularly co-occur is a clear candidate for the source of input from which thematic relations are
learned. Moreover, the emergence of thematic relations early in development, and their
persistence into adulthood, is consistent with the ready availability of co-occurrence
regularities in language and visual input.
Second, an inspection of the items used in prior studies suggests that many items selected as
taxonomically related by researchers also tended to co-occur in the (linguistic and/or visual)
11
environment, such as cat-dog, table-chair, lion-elephant, shirt-pants, and sheep-cow (Bauer &
Mandler, 1989; Bergelson & Aslin, 2017; Smiley & Brown, 1979; Styles & Plunkett, 2009;
Waxman & Namy, 1997; Willits, Wojcik, Seidenberg, & Saffran, 2013). This raises the possibility
that children’s treatment of such items as related in prior studies may reflect the contribution
of co-occurrence regularities, knowledge that the items belong to the same taxonomic
category, or some combination of both. Therefore, like the more extensively studied
contributions of perceptual similarity, co-occurrence regularities may also contribute to the
appearance of links between concepts belonging to the same taxonomic category (c.f. Sloutsky
& Fisher, 2004; Sloutsky et al., 2007) or to the emergence of taxonomic relations (c.f. Quinn &
Johnson, 2000) in children’s semantic organization.
1.3 Present Study
The goal of the present study is to directly investigate whether co-occurrence regularities
contribute to the relations between concepts that organize children’s semantic knowledge. To
capture these contributions (and potential changes therein) during a period of development
that has been extensively studied and debated (e.g., Blaye et al., 2006; Lucariello et al., 1992;
Nelson, 1977; Walsh et al., 1993; Waxman & Namy, 1997), we conducted this investigation with
three age groups of children spanning four to seven years of age. The study was designed to
assess whether co-occurrence regularities contribute to relations between concepts in
children’s semantic knowledge both when co-occurrence is the only source of relatedness, and
when the concepts are also taxonomically related. Specifically, we measured both: (1) Whether
children perceived concepts whose labels co-occur in corpora of child speech input as more
related than concepts whose labels do not co-occur, and (2) Whether children perceived
12
concepts belonging to the same taxonomic category as related more strongly when their labels
also co-occur (e.g., cat-mouse) than when they do not (e.g., cat-sheep). Evidence that children
perceive concepts with co-occurring labels as related would highlight co-occurrence as a
potential source of environmental input that can foster semantic relations. In particular, this
evidence would highlight co-occurrence as a key candidate for fostering thematic relations. By
the same token, evidence that children perceive concepts belonging to the same taxonomic
category as more related when their labels co-occur would highlight co-occurrence as also a
potential source of input that can foster relations between members of the same taxonomic
category.
In contrast with the majority of prior research on semantic organization in children that has
used tasks in which children made explicit relatedness judgements (e.g., forced choice tasks
used in Bauer & Mandler, 1989; Fenson et al., 1989; Fisher, 2011; Waxman & Namy, 1997), we
used an implicit measure of relational knowledge designed to attenuate the influence of task
demands or developmental changes in task comprehension. Specifically, we obtained this
implicit measure by asking children to perform a visual search task in which they looked for a
target item, and measuring the degree to which they noticed the presence of distractor items
that are related to the target. Multiple studies with adults have yielded evidence that semantic
relatedness between targets and distractors influence visual search (Huettig & Altmann, 2005;
Moores et al., 2003), and have used this phenomenon to investigate the relations that organize
semantic knowledge (Mirman & Graziano, 2012; Mirman & Magnuson, 2009). According to
recent research, similar effects of semantic distractors are evident in infants and young children
(Bergelson & Aslin, 2017; Chow, Davies, & Plunkett, 2017; Vales & Fisher, 2019; Vales, Unger, &
13
Fisher, 2017), highlighting the appropriateness of this approach for use with a broad
developmental age range. Moreover, in the case of targets and related distractors whose labels
co-occur, a study conducted by Yee, Overton, and Thompson-Schill (2009) provides evidence
that looking at related distractors is likely driven by semantic links between the concepts that
the labels denote. Specifically, many words are polysemous i.e., they can denote multiple
concepts, and only denote a specific concept when used in context. Thus, a semantic link based
on the co-occurrence between the words cat and mouse should be specific to the animal
concepts that these words denote in the contexts in which they co-occur, and not generalize to,
e.g., a computer mouse. Yee et al.’s (2009) study demonstrated that looking behavior is indeed
specific to such semantic links: For targets and related distractors with co-occurring labels,
participants only focused their gaze on a related distractor when it depicted the referent of its
label that is denoted in the context in which it typically co-occurs with the target label. For
example, participants looked at a picture of ham upon hearing the word “egg”, but did not look
at a picture of an iceberg upon hearing the word “lettuce”.
In our implementation of this approach, we used eye tracking during a visual search task to
measure the degree to which participants who were instructed to look for a target item noticed
the presence of a related versus unrelated distractor item. Specifically, we manipulated
whether related distractors were items that reliably co-occur with the target, belong to the
same taxonomic category as the target, or both. For example, if the target item were table, a
co-occurring distractor could be a plate, a taxonomic distractor could be a bed, and a both
taxonomic and co-occurring distractor a chair.
14
The possibility that co-occurrence regularities contribute to relations between concepts in
children’s semantic knowledge, both when concepts are taxonomically unrelated and when
they are taxonomically related, yields two predictions. First, children should look at co-
occurring distractors more than unrelated distractors. Second, the degree to which children
look at related versus unrelated distractors should be greater when the related distractor both
co-occurs with and is taxonomically related to the target relative to when it is taxonomically
related only.
2. Methods
The goal of this experiment was to test the effect of co-occurrence on semantic knowledge. To
accomplish this goal, we used a visual search task in which participants were asked to look for a
Target item, and then shown an array that included the Target, a Related Distractor, and an
Unrelated Distractor. The degree to which participants looked more at the Related versus the
Unrelated Distractor was taken as a measure of the degree to which concepts denoting the
Related Distractor and Target were linked in their semantic knowledge.
To investigate the role of co-occurrence in semantic organization development, we
manipulated the type of relation between the Target and Related Distractor, such that they
either co-occurred, were taxonomically related, or both, and tested the effect of this
manipulation on our measure of links between concepts. Specifically, we tested whether: (1)
Children looked more at co-occurring Related Distractors than Unrelated Distractors, and (2)
The degree to which children looked more at Related versus Unrelated Distractors was greater
15
for Related Distractors that both co-occurred with and were taxonomically related to the target
than for Related Distractors that were only taxonomically related.
2.1 Participants
The final sample included 72 children. There were 24 participants in each of three age groups:
Pre-K (Mage=4.32 years, SD=0.35; 12 female and 12 male; 21 Caucasian, 2 African American, 1
Asian), Kindergarten (Mage=5.40 years, SD=0.57; 9 female and 15 male; 20 Caucasian, 2 African
American, 1 Asian, 1 Caucasian & Hispanic), and 1st Grade (Mage=6.88 years, SD=0.39; 10 female
and 14 male; 19 Caucasian, 2 Caucasian & Hispanic, 2 Other, 1 Caucasian & Asian/Pacific
Islander). Participants were recruited from pre-schools and schools in a middle-class area in a
Northeastern United States city. Data from four children were excluded and replaced with data
from a new participant due to: Early termination of the study (N = 2), noncompliance with
procedures designed to ensure that eye tracking equipment was able to collect looking
behavior data (N = 1; see Procedure), or technical error (N=1).
2.2 Materials
The primary stimuli were Target items and Related Distractor items. Related Distractor items
were related to the Target items in one of three ways: Co-occurrence alone, Taxonomic
relatedness alone, or Both co-occurrence and taxonomic relatedness. All items were depictions
of concrete, real-world entities such as artifacts, foods, animals, and other natural kinds. We
next describe how we operationalized co-occurrence and taxonomic relatedness.
2.2.1 Co-occurrence operationalization. Regular co-occurrence was operationalized as rates of
co-occurrence between pairs words in linguistic input to children (recorded in the corpora in
16
the CHILDES database, a repository of child and child-directed speech; MacWhinney 200) that
were more frequent than would be expected by chance, where chance is the rate at which the
words would co-occur if they occurred randomly in linguistic input. These rates were
determined by calculating a “t-score” for word pairs in CHILDES, which captured the difference
between the observed co-occurrence rates (O) within a given window and the rates of co-
occurrence expected by chance (E) (Evert, 2008):
𝑡 𝑠𝑐𝑜𝑟𝑒 = 𝑂 − 𝐸
𝑂
T-scores calculated from CHILDES were chosen as the co-occurrence criterion both because
they capture a regularity available in environmental input
2
, and because they measure reliable
co-occurrence while controlling for the effect of item frequency. We identified word pairs with
t-scores greater than 3 as regularly co-occurring. In addition, we used a window size of twelve
words (following evidence that co-occurrence regularities within this window size in CHILDES
corpora capture semantic links, e.g., Asr et al., 2016; Willits, Jones, & Landy, 2016).
2.2.2 Taxonomic operationalization. Taxonomic relatedness was determined based on
similarity between the definitions of labels for the items, as measured via WordNet (a database
of words and their definitions). In WordNet, nouns are first grouped into sets of synonyms,
which are in turn linked into a hierarchy according to “IS A” and part-whole relations. Similarity
between pairs of words that label stimuli used in this experiment was measured using the
2
Because co-occurrence patterns in speech are closely mirrored by co-occurrence patterns of real-world
objects (Sadeghi, McClelland, & Hoffman, 2015), this measure may also reflect reliable co-occurrence in
visual input between the entities that words denote.
17
Resnik similarity measure, i.e., the information content (specificity) of the word lowest in the
WordNet hierarchy within which the pair of words is subsumed. For example, the words dog
and cat are subsumed within carnivore, whereas dog and kangaroo are subsumed within
mammal; because the information content of carnivore is greater than the information content
of mammal (i.e., mammal is more abstract), Resnick similarity is higher between dog and cat
versus dog and kangaroo. Similarity in WordNet was chosen as the taxonomic relatedness
criterion because it captures the essence of taxonomic relatedness i.e., close similarity in
meaning without relying on subjective judgments of adult participants that may be influenced
by non-taxonomic relations (Wisniewski & Bassok, 1999). We used a cutoff Resnik similarity
score of 5, such that we identified pairs with a Resnik similarity of greater than 5 as
Figure 1. Each graph shows Resnik similarity between a target item (left: cat, right: shoe)
and both members of the same taxonomic category, and members of different categories.
The same-taxonomic category members that were selected as Taxonomically Related to
the targets for the study are highlighted in gray. These graphs illustrate that the cutoff
Resnik value of 5 distinguished between same- and different-taxonomic category
members.
18
taxonomically related. This cutoff was chosen because it consistently differentiated between
members of the same versus different taxonomic categories (Figure 1).
2.2.2 Composition of Item Sets. As described at the beginning of the Methods section, this
experiment used a paradigm in which participants searched for Target items in the presence of
Related Distractors that either Co-occurred with the Target, were Taxonomically related to the
Target, or Both. However, there were few candidate Target items for which we could use the
co-occurrence and taxonomic criteria to identify appropriate items for each of the three types
of Related Distractors. Therefore, we composed two sets of items in which Targets were each
paired with only two of the three types of Related Distractors.
The two sets of stimuli (Appendix A) were the Both versus Co-Occurrence only set and the Both
versus Taxonomic only set. Both sets of stimuli consisted of five Target items that were each
paired with two Related Distractors. In both sets, one of the Related Distractors belonged to
the “Both” condition: It both co-occurred with and was taxonomically related to its Target (e.g.,
table-chair: both items belong to the category ‘furniture’ and reliably co-occur in the CHILDES
database) (t-scores: M=12.34, SD=6.75; Resnik Similarity: M= 6.38, SD= 1.60). In the Both versus
Co-Occurrence set, each Target was also paired with a Related Distractor belonging to the “Co-
Occur” condition with which it co-occurred but with which it was not taxonomically related
(e.g., table-plate) (t-scores: M=11.24, SD=5.83; Resnik Similarity: M= 1.28, SD= 0.88). In the
Both versus Taxonomic set, each Target was additionally paired with a Related Distractor
belonging to the “Taxonomic” condition with which it was taxonomically related but with which
it did not co-occur (e.g., table-bed) (t-scores: M=2.28, SD=0.69; Resnik Similarity: M= 5.92, SD=
0.73).
19
Related Distractors were selected to approximately equate: (1) The strength of co-occurrence
(i.e., t-scores) between Targets and their Related Distractors in the Both and Co-Occur
conditions in the Both versus Co-Occurrence set, and (2) The strength of taxonomic relatedness
(i.e., Resnick similarity) between Targets and their Related Distractors in the Both and
Taxonomic conditions in the Both versus Taxonomic set.
For both sets, we additionally paired each Target with Unrelated Distractors. Unrelated
Distractors for each Target were: (1) Approximately matched in frequency to the Related
Distractors for the Target in CHILDES corpora (MacWhinney, 2000), (2) Low in taxonomic
relatedness to the Target (Resnick Similarity: M = 1.70, SD = 0.94), and (3) Low in co-occurrence
with the Target (t-scores: M =-1.15, SD = 3.05) (see Appendix B).
2.2.3 Stimuli. The Both versus Co-Occurrence and Both versus Taxonomic item sets were used
to create triads of stimuli consisting of a Target, one of its Related Distractors, and an Unrelated
Target
Related Distractor
Unrelated Distractor
20
Distractor (Figure 2). For each item in a triad, three photograph images were selected to
maximize the perceptual diversity with which each item was represented (following the
successful implementation of this approach in Moores et al., 2003; see also section 3.1.2 for
evidence that visual similarity did not drive looking behavior in this experiment). All images
were edited to a size of 276px2. These images were used to create three parallel versions of
each triad. The selection of which images to use together for a triad in a given version was
designed to minimize systematic perceptual similarity between any two images in a triad
(Figure 3).
The three parallel versions of the Both versus Co-Occurrence and Both versus Taxonomic triads
were then each arranged into separate pseudo-random sequences, such that one of the parallel
versions was randomly selected to be presented to each participant. In the sequences: 1) Two
triads with the same Target never appeared consecutively, 2) Two triads in which Targets
appeared with the same type of Related Distractor never appeared consecutively, and 3) Each
triad appeared twice, to compensate for transient inattention to the visual search task. The
Figure 3. Example of three versions of a triad of items in the Both condition (Target: Table,
Related Distractor: Chair, Unrelated Distractor: Box) in which images have been chosen to
both maximize the perceptual diversity with which given items are represented, and
minimize perceptual similarity between any two images. Items are each shown in one of the
three equidistant positions in which they appeared to participants.
21
resulting sequences consisted of 20 trials (i.e., five triads in the Both condition and five in either
the Co-Occur or Tax condition, which each appeared twice). When presented to participants,
images in triads appeared in one of three approximately equidistant positions. The assignment
of Target, Related Distractor, and Unrelated Distractor items to positions was counterbalanced
such that 1) Each type of item appeared in each position an approximately equal number of
times, and 2) Across the two presentations of each triad, its constituent items appeared in
different positions.
2.3 Apparatus
Participants were seated approximately 64cm away from a 34.5cm x 19.5cm laptop monitor
with a resolution set to 1920x1080 dpi. At this distance, the 276px2 images each subtended
approximately 8.6˚ of visual angle. Triad presentation and gaze recording was accomplished
using either Tobii Studio (version 3.3, Tobii Technology, Stockholm, Sweden) or SMI Experiment
Center (version 3.7, SensoMotoric Instruments, Teltow, Germany) software. Specifically, 35
children viewed triads presented using SMI software, and the remaining 61 children viewed
triads presented using Tobii software. The appearance of the display and sampling rate (60Hz)
was constant across all participants. Because all pre-processing and analyses were conducted
using code written in the R environment specifically for this experiment, the variation in
software used across participants could not influence the results.
2.4 Procedure
All participants were tested individually in a quiet space, and took part in two sessions: One in
which they viewed one of the three parallel versions of the Both versus Co-Occurrence triad
22
sequence, and one in which they viewed one of the three parallel versions of the Both versus
Taxonomic triad sequence. The order in which participants viewed the two types of triad
sequence, and which parallel version they viewed, was counterbalanced across participants.
Viewing the two types of triad sequence was divided into two sessions in order to reduce the
effects of fatigue on looking behavior.
During each session, participants were seated in front of an eye tracking-equipped laptop (see
Apparatus), and told that they would play a game in which their job was to look for pictures on
the computer screen. The experimenter then pointed out the eye tracking camera, and said
that the game would only work if the camera could see the participant’s eyes. Next, the
experimenter calibrated the eye tracker to track the participants’ gaze.
To familiarize participants with the task and encourage them to avoid behaviors such as
pointing and speaking that can interfere with eye tracking data quality, participants then
completed a practice trial. During the practice trial, the experimenter asked the participant to
look for a picture of a dog, and then showed the participant an array of three pictures
consisting of a dog, balloons, and a clock. If the participants spontaneously pointed at the
picture of the dog and/or said that they had found it, the experimenter told the participants
that, “In this game, your job is to just look for the picture you don’t need to point or tell me
when you see it”. If a participant continued to point or speak following the practice trial, the
experimenter reminded the participant not to do so, and instructed the participant to clasp
their hands in their lap if necessary.
Following the practice trial, the participant viewed either the Both versus Co-Occurrence or
Both versus Taxonomic triad sequence. Each triad was preceded by a blank screen, during
23
which the experimenter told the participant to look for the Target in the upcoming triad. The
triad then appeared on the screen for five seconds, to ensure that participants had the
opportunity to look at each item in the triad multiple times. The outcome data of interest
consisted of measures of how long participants inspected Related and Unrelated Distractors
when asked to look for Targets. Each session took approximately 10 minutes.
2.5 Gaze Data Pre-processing
The raw data recorded by the eye tracker consisted of samples, collected at a rate of 60Hz, of
the X and Y coordinates measured in pixels of the left and right eye gaze on the screen, along
with a timestamp for each sample. Prior to analyzing looking behavior, these raw data were
first pre-processed using custom code written in R (R Core Development Team, 2008) to
average the positions of the eyes, interpolate over short (<75ms) gaps of missing data (Wass,
Smith, & Johnson, 2013), attenuate noise (Stampe, 1993), and detect fixations. Based on prior
evaluations of fixation detection parameters (Blignaut, 2009; Manor & Gordon, 2003), the gaze
recorded during a given sample was identified as a fixation if it was part of a gaze event that
lasted at least 100ms within which the dispersion of gaze positions did not exceed
approximately 1˚ of visual angle (35px). Both raw data and scripts used for pre-processing can
be found on the Open Science Framework (https://osf.io/9de7p/).
3. Results
Analyses were conducted in the R environment using functions in base R, the lme4 package for
mixed-model regression (Bates, Maechler, Bolker, & Walker, 2015), and the car package for
calculating significance values of main effects and interactions from mixed-model regression
24
(Fox & Weisberg, 2011). Both pre-processed data and scripts for these analyses can be found on
the Open Science Framework (https://osf.io/9de7p/).
3.1 Contributions of Co-Occurrence and Taxonomic Relations to Children’s Semantic
Knowledge
To evaluate the contributions of co-occurrence and taxonomic relations to semantic
organization, we first calculated the total duration of fixations for Related and Unrelated
Distractors in each trial (see also Supplemental Materials for exploratory analyses of
contributions to the time course of looking at Distractors within trials). We then used these
values to calculate a “Related Looking” difference score for each trial that captured the degree
to which a participant looked longer at the Related versus Unrelated Distractor (i.e., total
fixation duration to Related Distractor - Unrelated Distractor). We then calculated the mean
value of the Related Looking score for each Relation Type (i.e., Taxonomic, Co-Occurring, and
Both) for each participant.
3.1.1 Relations in children’s semantic knowledge. Using the Related Looking data, we first
tested whether the magnitude of Related Looking differed from chance (i.e., 0) for each
Relation Type. One-sample t-tests, Bonferroni-corrected for multiple comparisons, revealed
that Related Looking was greater than chance in all Relation Type conditions (all ps < .01).
Therefore, our paradigm detected contributions of all Relation Types to children’s semantic
knowledge.
3.1.2 Comparison of relations in children’s semantic knowledge. Next, to compare the relative
contributions of the Relation Types, we generated an omnibus mixed effects model with
25
Relation Type and Age Group (Pre-Kindergarten, Kindergarten and 1st Grade) as fixed effects,
and a random intercept for participant. This analysis revealed main effects of Age Group
(F(2,69)=5.219, p=.008) and Relation Type (F(2,138)=7.985, p=.0005) that did not interact (F(4,
138)=1.711, p=.151; though see Supplemental Materials and available code for exploratory
analyses of changes in effects of Relation Type across Age Groups). To examine the main effects
of these factors, we used a pairwise comparisons approach in which we re-generated the
omnibus model with each level of a given factor as the reference level to which the others were
compared.
Pairwise comparisons of the Age Group factor revealed significantly lower overall Related
Looking magnitudes in 1st Grade than in both Pre-Kindergarten and Kindergarten (Figure 4).
Inspection of the fixation durations to both Related and Unrelated Distractors suggests that this
result was due to the fact that the total amount of time and variance of looking at either
Figure 4. Related Looking (calculated from total duration of fixations to Related minus
Unrelated Distractors in each trial) in the Relation Type conditions. Data are shown collapsed
across Age Group in the left panel (error bars represent standard errors of the mean). Data
are separated by Age Group and depicted with data points for individual participants on the
right.
26
Distractor was much lower in 1st Grade children (M=169.89, SD=308.79) than in either
Kindergarten (M=403.01, SD=512.12) or Pre-Kindergarten children (M=438.90, SD=533.18). The
degree to which fixation durations for Related versus Unrelated Distractors could differ from
was therefore smaller in 1st Grade children.
Critically, pairwise comparisons of the Relation Type factor (Figure 4) revealed significantly
greater Related Looking magnitudes in both the Co-Occur (M=131.99ms, SD=206.41ms,
p<.0001) and Both (M=132.30ms, SD=149.06ms, p=.004) conditions than in the Taxonomic
condition (M=51.56ms, SD=136.75ms). There was no significant difference between the Co-
Occur and Both conditions (p=.241). In sum, we observed that children noticed Related
Distractors in the Both and Co-Occur conditions to a greater extent than Related Distractors in
the Taxonomic condition. These patterns suggest that children’s semantic knowledge contained
stronger links between items that co-occurred, regardless of whether they were also
taxonomically related, than between items that were only taxonomically related and did not co-
occur.
We additionally note that although the stimuli were designed to minimize visual similarity
between Targets and Related Distractors, we conducted a separate control study to address the
possibility that the patterns of looking observed in the present experiment were driven by
visual similarity. This study collected adult visual similarity ratings between Targets and their
Related and Unrelated Distractors, and is reported in full in Supplemental Materials (see also
materials on the Open Science Framework for data and analyses). The results of this study
revealed that visual similarity was similarly greater in the Taxonomic and Both conditions than
in the Co-Occur condition. Thus, visual similarity was high for pairs that belong to the same
27
taxonomic category, regardless of whether they also co-occurred. This pattern contrasts
strongly with looking behavior in the present experiment, in which Related Looking was highest
when Related Distractor labels co-occurred with Target labels, regardless of whether they also
belonged to the same taxonomic category. Thus, visual similarity cannot explain why Related
Looking was greater in the Co-Occur versus Taxonomic conditions, or the Both versus
Taxonomic conditions.
3.1.3 Co-Occurrence contributions to individual differences in semantic knowledge. The
results reported in the previous section suggest that co-occurrence contributes to semantic
links between concepts. Specifically, co-occurrence contributes both to semantic links on its
own, and to semantic links between concepts that belong to the same taxonomic category (as
Figure 5. Scatterplots depicting the degree to which individuals’ Both Related Looking
magnitudes were associated with their Co-Occur Related Looking magnitudes (left) and their
Taxonomic Related Looking magnitudes (right). Dashed lines represent regression lines from a
simple linear regression model in which Both Related Looking was predicted by Co-Occur and
Taxonomic Related Looking.
28
shown by greater Related Looking in the Both than the Taxonomic Relation Type condition).
The finding that co-occurrence contributes to links between concepts that belong to the same
taxonomic category is particularly noteworthy, because many concepts that belong to the same
taxonomic category are denoted by labels that co-occur, such as cat-dog, table-chair, lion-
elephant, and shirt-pants. Moreover, such concepts have often been used in prior research on
the development of semantic organization (Bauer & Mandler, 1989; Bergelson & Aslin, 2017;
Smiley & Brown, 1979; Styles & Plunkett, 2009; Waxman & Namy, 1997; Willits et al., 2013).
What are the respective contributions of co-occurrence and membership in the same
taxonomic category to semantic links between these concepts? . To address this question, we
investigated whether co-occurrence not only contributes to links between members of the
same taxonomic category, but moreover, contributes to a larger extent than the taxonomic
relation between them. Specifically, we tested whether individual differences in the magnitude
of children’s Related Looking in the Both Relation Type condition were more associated with
their Related Looking in the Co-Occur than in the Taxonomic conditions. Plots of these
associations are depicted in Figure 5.
To conduct this test, we generated a simple linear regression model in which Related Looking in
the Both condition was independently predicted by Related Looking in the Co-Occur condition,
and in the Taxonomic condition. This model revealed that Related Looking in the Co-Occur
condition was a significant predictor of Related Looking in the Both condition (Estimate=0.325,
SE=0.080, p=.0001), whereas Related looking in the Taxonomic condition was not
(Estimate=0.039, SE=0.120, p=.744).
3.2 Summary
29
The results of this study revealed: (1) Stronger links between concepts whose labels co-occur
than those whose labels do not co-occur, and (2) Stronger links between concepts belonging to
the same taxonomic category when their labels also co-occur than when they do not.
Moreover, we observed evidence that co-occurrence contributes more than taxonomic
relatedness to the degree to which children perceive entities that are linked in both ways as
related. These findings provide new evidence that co-occurrence contributes to both relations
between concepts in general, and relations between concepts belonging to the same
taxonomic category in particular.
4. General Discussion
The purpose of this study was to investigate the contributions of co-occurrence regularities to
the organization of children’s semantic knowledge during a substantial period of semantic
development from age four to seven that has been extensively studied and debated in prior
work (e.g., Blaye et al., 2006; Deák & Bauer, 1996; Tversky, 1985; Walsh et al., 1993; Waxman &
Namy, 1997). Using an implicit, eye-gaze measure of the relations that organize children’s
semantic knowledge, we observed that co-occurrence contributed both independently to links
between concepts, and to links between concepts belonging to the same taxonomic category.
Specifically, children both: (1) Perceived concepts whose labels co-occur in child-directed
speech input as more related than those whose labels do not co-occur, and (2) Perceived
concepts belonging to the same taxonomic category as more strongly related when their labels
also co-occur than when they do not. Moreover, we observed that the degree to which children
perceived concepts that both co-occur and belong to the same taxonomic category as related
was predicted by the degree to which they perceived co-occurring concepts as related, but not
30
by their perception of taxonomically related concepts as related. This outcome suggests that
children’s perception of relatedness between concepts such as cat and dog, which both co-
occur and belong to the same taxonomic category, is driven by their sensitivity to co-
occurrence to a greater extent than by their taxonomic knowledge.
These findings highlight key contributions of co-occurrence regularities to semantic
organization during development, a source of environmental input that has been largely
overlooked in prior research. First, these findings underline co-occurrence regularities as a key
candidate source of input for fostering thematic relations, which contribute to organizing
semantic knowledge from early in development (Fenson et al., 1989) into adulthood (Lin &
Murphy, 2001). By the same token, the ready availability of these regularities in input may help
explain both why the contributions of co-occurrence were greater than those of taxonomic
relations in the present study, and evidence for the early emergence of thematic relations in
semantic organization in numerous prior studies . Importantly, the present findings also suggest
that co-occurrence regularities may contribute to relations between members of the same
taxonomic category. Specifically, the degree to which children perceived members of the same
taxonomic category as related was significantly increased when their labels co-occurred. Thus,
even relations between members of the same taxonomic category may be driven in part by co-
occurrence regularities.
4.1 Implications for Semantic Development Research
Together, these findings have several important implications for our understanding and future
investigation of the development of semantic organization. First, when interpreting the results
of prior investigations of semantic organization development, it is important to consider the
31
degree to which co-occurrence may have contributed to the results of previous studies that
aimed to examine the contribution of taxonomic relations to semantic knowledge. For example,
the present findings highlight challenges for interpreting the results of prior studies that have
used taxonomically related items that also co-occur such as chair-table, cat-dog, lion-elephant,
etc. (Bauer & Mandler, 1989; Smiley & Brown, 1979; Styles & Plunkett, 2009; Waxman & Namy,
1997; Willits et al., 2013). By the same token, future investigations of semantic organization
development should measure and account for contributions of co-occurrence regularities
much in the same way in which researchers measure and account for contributions of
perceptual similarity (Nguyen, 2007; Waxman & Namy, 1997).
Finally, existing theoretical accounts of the drivers of semantic organization development have
highlighted roles for some sources of information in the environment, with a particular focus on
perceptual features (Hills, Maouene, Maouene, Sheya, & Smith, 2009; McClelland & Rogers,
2003; Sloutsky, 2010; Smith & Heise, 1992). The results of the present study suggest that a full
understanding of the origins of semantic organization in development will also require
accounting for the contributions of co-occurrence regularities.
4.2 Expansion of Statistical Learning Research
The evidence for a contribution of co-occurrence to semantic organization is consistent with
results from prior statistical learning studies that showed evidence that a sensitivity to
environmental regularities, such as reliable co-occurrence, can be found in multiple domains
from early in development onward (Bulf et al., 2011; Fiser & Aslin, 2002; Saffran et al., 1996).
Although the body of research on statistical learning is sizable, the possibility that sensitivity to
environmental regularities contributes to semantic knowledge has received only limited
32
investigation to date (Matlen et al., 2015; Wojcik & Saffran, 2015). The present results
therefore build upon and extend prior research by providing evidence that exposure to such
regularities contributes to linking concepts in semantic organization. By the same token, the
present findings highlight a potential new avenue for research into how exposure to co-
occurrence regularities may foster the formation of semantic relations.
4.3 Conclusion
Due to the impact that the organization of semantic knowledge has on a multitude of facets of
cognition (including memory retrieval, reasoning, and knowledge learning), understanding its
origins and how it may be shaped through experience is a key endeavor in cognitive science. A
potential role for learning from environmental regularities, such as reliable co-occurrence, has
received little attention to date. In this study, we found evidence that young children’s
semantic relational knowledge in general, and knowledge of relations between members of the
same taxonomic category in particular, is influenced by a sensitivity to reliable co-occurrence in
the environmental input. These findings have both significant methodological implications, such
that they highlight the importance of accounting for reliable co-occurrence when attempting to
manipulate semantic relations between items for use in studies, and key theoretical
implications, such that they highlight a source of environmental input that may play a
substantive role in the emergence of semantic organization.
33
References
Asr, F. T., Willits, J. A., & Jones, M. N. (2016). Comparing predictive and co-occurrence based models of
lexical semantics trained on child-directed speech. Paper presented at the 38th Annual Meeting
of the Cognitive Science Society, Philadelphia, PA.
Bates, D., Maechler, M., Bolker, B., & Walker, S. (2015). Fitting linear mixed-effects models using lme4.
Journal of Statistical Software, 67, 1-48.
Bauer, P. J., & Mandler, J. M. (1989). Taxonomies and triads: Conceptual organization in one-to two-
year-olds. Cognitive Psychology, 21, 156-184.
Bergelson, E., & Aslin, R. N. (2017). Nature and origins of the lexicon in 6-mo-olds. Proceedings of the
National Academy of Sciences, 114, 12916-12921.
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.
Blignaut, P. (2009). Fixation identification: The optimum threshold for a dispersion algorithm. Attention,
Perception, & Psychophysics, 71, 881-895.
Bower, G. H., Clark, M. C., Lesgold, A. M., & Winzenz, D. (1969). Hierarchical retrieval schemes in recall
of categorized word lists. Journal of Verbal Learning and Verbal Behavior, 8, 323-343.
Bulf, H., Johnson, S. P., & Valenza, E. (2011). Visual statistical learning in the newborn infant. Cognition,
121, 127-132.
Carey, S. (1985). Conceptual change in childhood. Cambridge, Massachusetts: MIT Press.
Chow, J., Davies, A. A., & Plunkett, K. (2017). Spoken-word recognition in 2-year-olds: The tug of war
between phonological and semantic activation. Journal of Memory and Language, 93, 104-134.
34
Coley, J. D. (2012). Where the wild things are: Informal experience and ecological reasoning. Child
Development, 83, 992-1006.
Deák, G., & Bauer, P. (1996). The dynamics of preschoolers' categorization choices. Child Development,
67, 740-767.
Eimas, P. D., & Quinn, P. C. (1994). Studies on the formation of perceptually based basic-level categories
in young infants. Child development, 65, 903-917.
Fenson, L., Vella, D., & Kennedy, M. (1989). Children's knowledge of thematic and taxonomic relations at
two years of age. Child Development, 60, 911-919.
Fiser, J., & Aslin, R. N. (2002). Statistical learning of new visual feature combinations by infants.
Proceedings of the National Academy of Sciences, 99, 15822-15826.
Fisher, A. V. (2010). What’s in the name? Or how rocks and stones are different from bunnies and
rabbits. Journal of Experimental Child Psychology, 105, 198-212.
Fisher, A. V. (2011). Processing of perceptual information is more robust than processing of conceptual
information in preschool-age children: Evidence from costs of switching. Cognition, 119, 253-
264.
Fisher, A. V., Matlen, B. J., & Godwin, K. E. (2011). Semantic similarity of labels and inductive
generalization: Taking a second look. Cognition, 118, 432-438.
Fox, J., & Weisberg, S. (2011). An r companion to applied regression (Second Edition ed.). Thousand
Oaks, CA: Sage.
Frermann, L., & Lapata, M. (2015). Incremental bayesian category learning from natural language.
Cognitive Science, 40, 13331381.
Gobbo, C., & Chi, M. (1986). How knowledge is structured and used by expert and novice children.
Cognitive Development, 1, 221-237.
Heit, E. (2000). Properties of inductive reasoning. Psychonomic Bulletin & Review, 7, 569-592.
35
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, 381-396.
Hofmann, M. J., Biemann, C., Westbury, C., Murusidze, M., Conrad, M., & Jacobs, A. M. (2018). Simple
co‐occurrence statistics reproducibly predict association ratings. Cognitive Science, 42, 2287-
2312.
Huebner, P. A., & Willits, J. A. (2018). Structured semantic knowledge can emerge automatically from
predicting word sequences in child-directed speech. Frontiers in Psychology, 9.
Huettig, F., & Altmann, G. T. (2005). Word meaning and the control of eye fixation: Semantic competitor
effects and the visual world paradigm. Cognition, 96, B23-B32.
Imai, M., Gentner, D., & Uchida, N. (1994). Children's theories of word meaning: The role of shape
similarity in early acquisition. Cognitive Development, 9, 45-75.
Inhelder, B., & Piaget, J. (1964). The early growth of logic in the child. New York: Norton.
Jimura, K., Hirose, S., Wada, H., Yoshizawa, Y., Imai, Y., Akahane, M., . . . Konishi, S. (2016). Relatedness-
dependent rapid development of brain activity in anterior temporal cortex during pair-
association retrieval. Neuroscience Letters, 627, 24-29.
Jones, M. N., & Mewhort, D. J. (2007). Representing word meaning and order information in a
composite holographic lexicon. Psychological Review, 114, 1-37.
Landauer, T. K., & Dumais, S. T. (1997). A solution to plato's problem: The latent semantic analysis
theory of acquisition, induction, and representation of knowledge. Psychological Review, 104,
211.
Lin, E. L., & Murphy, G. L. (2001). Thematic relations in adults' concepts. Journal of Experimental
Psychology: General, 130, 3-28.
Lucariello, J., Kyratzis, A., & Nelson, K. (1992). Taxonomic knowledge: What kind and when? Child
Development, 63, 978-998.
36
Manor, B. R., & Gordon, E. (2003). Defining the temporal threshold for ocular fixation in free-viewing
visuocognitive tasks. Journal of Neuroscience Methods, 128, 85-93.
Matlen, B. J., Fisher, A. V., & Godwin, K. E. (2015). The influence of label co-occurrence and semantic
similarity on children’s inductive generalization. Frontiers in Psychology, 6, 1146.
McClelland, J. L., & Rogers, T. T. (2003). The parallel distributed processing approach to semantic
cognition. Nature Reviews Neuroscience, 4, 310-322.
Medin, D. L., Lynch, E. B., Coley, J. D., & Atran, S. (1997). Categorization and reasoning among tree
experts: Do all roads lead to rome? Cognitive Psychology, 32, 49-96.
Mirman, D., & Graziano, K. M. (2012). Individual differences in the strength of taxonomic versus
thematic relations. Journal of Experimental Psychology: General, 141, 601.
Mirman, D., & Magnuson, J. S. (2009). Dynamics of activation of semantically similar concepts during
spoken word recognition. Memory & Cognition, 37, 1026-1039.
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, 929-
941.
Moores, E., Laiti, L., & Chelazzi, L. (2003). Associative knowledge controls deployment of visual selective
attention. Nature Neuroscience, 6, 182-189.
Nelson, K. (1977). The syntagmatic-paradigmatic shift revisited: A review of research and theory.
Psychological Bulletin, 84, 93.
Nguyen, S. P. (2007). Cross-classification and category representation in children's concepts.
Developmental Psychology, 43, 719-731.
Oakes, L. M., Coppage, D. J., & Dingel, A. (1997). By land or by sea: The role of perceptual similarity in
infants' categorization of animals. Developmental Psychology, 33, 396.
37
Pelucchi, B., Hay, J. F., & Saffran, J. R. (2009). Statistical learning in a natural language by 8‐month‐old
infants. Child Development, 80, 674-685.
Quinn, P. C., Eimas, P. D., & Rosenkrantz, S. L. (1993). Evidence for representations of perceptually
similar natural categories by 3-month-old and 4-month-old infants. Perception, 22, 463-475.
Quinn, P. C., & Johnson, M. H. (2000). Global‐before‐basic object categorization in connectionist
networks and 2‐month‐old infants. Infancy, 1, 31-46.
Rohde, D. L., Gonnerman, L. M., & Plaut, D. C. (2004). An improved method for deriving word meaning
from lexical co-occurrence. Cognitive Psychology, 7, 573-605.
Ross, B. H., & Murphy, G. L. (1999). Food for thought: Cross-classification and category organization in a
complex real-world domain. Cognitive Psychology, 38, 495-553.
Sadeghi, Z., McClelland, J. L., & Hoffman, P. (2015). You shall know an object by the company it keeps:
An investigation of semantic representations derived from object co-occurrence in visual scenes.
Neuropsychologia, 76, 52-61.
Saffran, J. R., Aslin, R. N., & Newport, E. L. (1996). Statistical learning by 8-month-old infants. Science,
274, 1926-1928.
Saffran, J. R., Johnson, E. K., Aslin, R. N., & Newport, E. L. (1999). Statistical learning of tone sequences
by human infants and adults. Cognition, 70, 27-52.
Sloutsky, V. M. (2010). From perceptual categories to concepts: What develops? Cognitive Science, 34,
1244-1286.
Sloutsky, V. M., & Fisher, A. V. (2004). Induction and categorization in young children: A similarity-based
model. Journal of Experimental Psychology: General, 133, 166-187.
Sloutsky, V. M., Kloos, H., & Fisher, A. V. (2007). When looks are everything appearance similarity versus
kind information in early induction. Psychological Science, 18, 179-185.
38
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, 249-257.
Smith, L. B., & Heise, D. (1992). Perceptual similarity and conceptual structure. In B. Burns (Ed.),
Advances in psychology: Percepts, concepts, and categories (Vol. 93, pp. 233-272). North-
Holland: Elsevier.
Spence, D. P., & Owens, K. C. (1990). Lexical co-occurrence and association strength. Journal of
Psycholinguistic Research, 19, 317-330.
Stampe, D. M. (1993). Heuristic filtering and reliable calibration methods for video-based pupil-tracking
systems. Behavior Research Methods, Instruments, & Computers, 25, 137-142.
Styles, S. J., & Plunkett, K. (2009). How do infants build a semantic system? Language and Cognition, 1,
1-24.
R development core team. (2008). R: A language and environment for statistical computing. Vienna,
Austria: R Foundation for Statistical Computing.
Tse, D., Langston, R. F., Kakeyama, M., Bethus, I., Spooner, P. A., Wood, E. R., . . . Morris, R. G. (2007).
Schemas and memory consolidation. Science, 316, 76-82.
Tversky, B. (1985). Development of taxonomic organization of named and pictured categories.
Developmental Psychology, 21, 1111-1119.
Unger, L., Fisher, A. V., Nugent, R., Ventura, S. L., & MacLellan, C. J. (2016). Developmental changes in
semantic knowledge organization. Journal of Experimental Child Psychology, 146, 202-222.
Vales, C., & Fisher, A. V. (2019). When stronger knowledge slows you down: Semantic relatedness
predicts children's co‐activation of related items in a visual search paradigm. Cognitive Science,
43, e12746.
39
Vales, C., Unger, L., & Fisher, A. V. (2017). Looking for the cat and seeing the dog: Using visual search to
study semantic knowledge in children Proceedings of the 39th annual conference of the
cognitive science society. London, UK.
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, 85-102.
Wass, S. V., Smith, T. J., & Johnson, M. H. (2013). Parsing eye-tracking data of variable quality to provide
accurate fixation duration estimates in infants and adults. Behavior Research Methods, 45, 229-
250.
Waxman, S. R., & Namy, L. L. (1997). Challenging the notion of a thematic preference in young children.
Developmental Psychology, 33, 555-567.
Willits, J. A., Jones, M. N., & Landy, D. (2016). Learning that numbers are the same, while learning that
they are different Proceedings of the 38th annual conference of the cognitive science society.
Philadelphia, PA.
Willits, J. A., Wojcik, E. H., Seidenberg, M. S., & Saffran, J. R. (2013). Toddlers activate lexical semantic
knowledge in the absence of visual referents: Evidence from auditory priming. Infancy, 18, 1053-
1075.
Wisniewski, E. J., & Bassok, M. (1999). What makes a man similar to a tie? Stimulus compatibility with
comparison and integration. Cognitive Psychology, 39, 208-238.
Wojcik, E. H., & Saffran, J. R. (2015). Toddlers encode similarities among novel words from meaningful
sentences. Cognition, 138, 10-20.
Yee, E., Overton, E., & Thompson-Schill, S. L. (2009). Looking for meaning: Eye movements are sensitive
to overlapping semantic features, not association. Psychonomic bulletin & review, 16, 869-874.
40
Appendix A
Table A1.
Taxonomic relatedness (Resnik) and co-occurrence (t-score) between Targets and Related
Distractors in the Both versus Co-Occurrence set
Both Distractor
Co-Occur only Distractor
Target
Item
Resnik
T-score
Item
Resnik
T-Score
Car
Bus
5.46
11.83
Street
2.49
19.69
Pig
Cow
7.00
21.53
Farm
1.17
13.83
Monkey
Tiger
5.61
10.24
Banana
1.37
6.37
Carrot
Peas
6.35
3.73
Rabbit
1.37
11.29
Cake
Cookie
10.67
4.95
Present
0
5.03
Average
7.02
10.46
1.28
11.24
Table A2.
Taxonomic relatedness (Resnik) and co-occurrence (t-score) between Targets and Related
Distractors in the Both versus Taxonomic set
Both Distractor
Tax only Distractor
Target
Item
Resnik
T-Score
Item
Resnik
T-Score
Table
Chair
6.19
26.01
Bed
6.19
2.85
Cat
Mouse
5.61
10.69
Sheep
5.61
1.27
Shoe
Sock
5.27
11.09
Coat
5.27
1.84
Giraffe
Zebra
6.22
11.91
Goat
7.07
2.82
Burger
Fries
5.46
11.47
Pretzels
5.46
2.59
Average
5.75
14.23
5.92
2.28
41
Appendix B
Table B1.
Taxonomic relatedness (Resnik) and co-occurrence (t-score) between Targets and Unrelated
Distractors
Both versus Co-Occur Only Set
Both versus Taxonomic Only Set
Target
Item
Resnik
T-score
Target
Item
Resnik
T-score
Car
Phone
3.45
0.38
Table
Box
3.45
1.45
Button
3.45
-3.20
Hat
2.49
-3.99
Pig
Tape
1.37
0
Cat
Rock
1.37
-4.36
Brush
1.37
0
Pencil
1.37
-11.38
Monkey
Blocks
1.37
1.82
Shoe
Window
2.49
0
Crayon
1.37
1.02
Key
2.49
0
Carrot
Ladder
1.37
0
Giraffe
Fan
1.37
0
Door
1.37
0
Towel
1.37
0
Cake
Tree
0.614
-4.63
Burger
Mittens
0.614
0
Dress
0.614
0
Sweater
0.614
0
Average
1.63
-0.46
1.58
-1.83
... For example, children may conceptualize items by thematic relations which share a similar environment. Specifically, young children (age 4-5) prefer to match a sock with a foot (i.e., co-occurrence in language) rather than with pajamas (i.e., taxonomic), suggesting the preference for thematic relation for conceptualization (Unger et al., 2020). ...
... On the basis of our findings, we argue for the need for further research on potential learning experiences that might facilitate children's integration of knowledge across distinct domains. Specifically, co-occurrence in languages predicts children's integration of knowledge (e.g., sock and foot versus sock and pajamas; Unger et al., 2020). It is also possible that children's integration of quantities in rational numbers across notations would benefit by having more co-occurring experiences. ...
Article
Knowledge integration is a crucial part of learning as concepts are built over time and modality. In mathematics, rational numbers are a particularly difficult concept that requires integration across notations to understand the quantity represented by fractions and decimals (e.g., ½ and 0.5). We investigated how 5- to 14-year-old children conceptualize rational numbers, whether it differs across development (Study 1), and whether the conceptualization is related to math achievement (Study 2). A novel Numerical SpAM task was created to document children’s unprompted rational number conceptualization. We found that most children organized rational numbers either by notation or quantity, with quantity becoming more common after age 10. Quantitative conceptualization predicted higher math achievement compared to notation, regardless of age. Moreover, conceptualizing rational numbers by quantity across distinct notations (e.g., ¹/2 and 0.5) was related to children’s math achievement. Implications for numerical understanding as well as general cognition are discussed.
... Indeed, subsequent studies (Fisher, 2010;Fisher et al., 2011) found that when Target and Taxonomic match items were referred to by semantically similar labels, four-year-old children only reliably chose Taxonomic matches when semantically similar labels reliably directly co-occurred in language input to children. In contrast, young children responded at chance when labels for Taxonomic matches and targets did not reliably co-occur (e.g., lamb and sheep) (see also Unger, Vales, & Fisher, 2020b, for further evidence that co-occurrence contributes relations between members of the same Taxonomic category in children). ...
... This insight can instead be gained by investigating the development of semantic relations that can be formed from measurable information available in the environment. Specifically, investigating the contributions of direct and shared co-occurrence to semantic development could involve testing the degree to which direct and shared co-occurrence regularities (measured from, e.g., recordings of child language input, MacWhinney, 2000) predict the development of semantic relations (Unger et al., 2020a;Unger et al., 2020b). ...
Article
Full-text available
As adults, we draw upon our ample knowledge about the world to support such vital cognitive feats as using language, reasoning, retrieving knowledge relevant to our current goals, planning for the future, adapting to unexpected events, and navigating through the environment. Our knowledge readily supports these feats because it is not merely a collection of stored facts, but rather functions as an organized, semantic network of concepts connected by meaningful relations. How do the relations that fundamentally organize semantic concepts emerge with development? Here, we cast a spotlight on a potentially powerful but often overlooked driver of semantic organization: Rich statistical regularities that are ubiquitous in both language and visual input. In this synthetic review, we show that a driving role for statistical regularities is convergently supported by evidence from diverse fields, including computational modeling, statistical learning, and semantic development. Finally, we identify a number of key avenues of future research into how statistical regularities may drive the development of semantic organization. Keywords: semantic development; semantic organization; semantic knowledge; statistical learning; taxonomic relations; association
... Indeed, subsequent studies (Fisher, 2010;Fisher et al., 2011) found that when Target and Taxonomic match items were referred to by semantically similar labels, four-year-old children only reliably chose Taxonomic matches when semantically similar labels reliably directly co-occurred in language input to children. In contrast, young children responded at chance when labels for Taxonomic matches and targets did not reliably co-occur (e.g., lamb and sheep) (see also Unger, Vales, & Fisher, 2020, for further evidence that co-occurrence contributes relations between members of the same Taxonomic category in children). ...
... This insight can instead be gained by investigating the development of semantic relations that can be formed from measurable information available in the environment. Specifically, investigating the contributions of direct and shared co-occurrence to semantic development could involve testing the degree to which direct and shared co-occurrence regularities (measured from, e.g., recordings of child language input, MacWhinney, 2000) predict the development of semantic relations , Unger, Vales, et al., 2020. ...
Preprint
As adults, we draw upon our ample knowledge about the world to support such vital cognitive feats as using language, reasoning, retrieving knowledge relevant to our current goals, planning for the future, adapting to unexpected events, and navigating through the environment. Our knowledge readily supports these feats because it is not merely a collection of stored facts, but rather functions as an organized, semantic network of concepts connected by meaningful relations. How do the relations that fundamentally organize semantic concepts emerge with development? Here, we cast a spotlight on a potentially powerful but often overlooked driver of semantic organization: Rich statistical regularities that are ubiquitous in both language and visual input. In this synthetic review, we show that a driving role for statistical regularities is convergently supported by evidence from diverse fields, including computational modeling, statistical learning, and semantic development. Finally, we identify a number of key avenues of future research into how statistical regularities may drive the development of semantic organization.
... To evaluate the co-occurrence, several metrics exists in the literature the most popular being the co-occurrence frequency, the t-score [67] and the mutual information or enhanced versions of it [68]. The mutual information metric does not perform well in low co-occurrence frequencies while the t-score performs best in well attested pairings. ...
Thesis
Full-text available
In this thesis, a bibliographic meta-analysis of the risk management field in the space sector is presented and a new robust and reproducible hybrid approach for space mission risk classification is proposed. The method consists of: a) text mining of bibliographic data and b) an Impact-Matrix-Cross-Reference- Multiplication-Applied-to-a-Classification (MICMAC)-inspired analysis that measures the degree of interrelation of the defined classification parameters and uses it to calculate automatically a weight factor for each one of them. The proposed method, thus, eliminates the error of a potentially subjective judgment from the otherwise required panel of experts that was tasked with the evaluation of the degree of interrelation of the various parameters in traditional MICMAC methods or with the assignment of weights to the parameters. Eight classification parameters are defined and assessed including: a) Criticality to the Agency, b) Objective importance, c) Cost, d) Lifetime, e) Complexity, f) Manned mission, g) Destination, and h) Ω Factor. Finally, the NASA Mars mission Perseverance is used to demonstrate the classification algorithm.
Article
Full-text available
With development knowledge becomes organized according to semantic links, including early-developing associative (e.g., juicy-apple) and gradually-developing taxonomic links (e.g., apple-pear). Word co-occurrence regularities may foster these links: Associative links may form from direct co-occurrence (e.g., juicy-apple), and taxonomic links from shared co-occurrence (e.g., apple and pear co-occur with juicy). Four experiments (2017-2020) investigated this possibility with 4-8-year-olds (N =148, 82 female) and adults (N =116, 35 female) in a US city with 58.6% White; 29.0% Black, and 5.8% Asian demographics. Results revealed earlier development of the abilities to form direct (d s>0.536) than the abilities to form shared co-occurrence-based links (d s>1.291). We argue that the asynchronous development of abilities to form co-occurrence-based links may explain developmental changes in semantic organization.
Article
Full-text available
According to the Declarative/Procedural Model, the lexicon depends on declarative memory while grammar relies on procedural memory. Furthermore, procedural memory underlies the sequential processing of language. Thus, this system is important for predicting the next item in a sentence. Verb processing represents a good candidate to test this assumption. Semantic representations of verbs include information about the protagonists in the situations they refer to. This semantic knowledge is acquired implicitly and used during verb processing, such that the processing of a verb preactivates its typical patients (e.g., the window for break). Thus, determining how the patient typicality effect appears during children’s cognitive development could provide evidence about the memory system that is dedicated to this effect. Two studies are presented in which French children aged 6–10 and adults made grammaticality judgments on 80 auditorily presented sentences. In Experiment 1, the verb was followed by a typical patient or by a less typical patient. In Experiment 2, grammatical sentences were constructed such that the verb was followed either by a typical patient or by a noun that could not be a patient of that verb. The typicality effect occurs in younger children and is interpreted in terms of developmental invariance. We suggest that this effect may depend on procedural memory, in line with studies that showed that meaning is necessary to allow procedural memory to learn the sequence of words in a sentence.
Article
Purpose We examined the association of Stroke Belt birth state with late-life cognition in The Study of Healthy Aging in African Americans (STAR). Methods STAR enrolled 764 Black Americans ages 50+ who were long-term Kaiser Permanente Northern California members. Participants completed Multiphasic Health Check-ups (MHC;1964-1985) where early-life overweight/obesity, hypertension, diabetes, and hyperlipidemia were measured. At STAR (2018), birth state, self-reported early-life socioeconomic status (SES), and executive function, verbal episodic memory, and semantic memory scores were collected. We used linear regression to examine the association between Stroke Belt birth and late-life cognition adjusting for birth year, gender, and parental education. We evaluated early-life SES and cardiovascular risk factors (CVRF) as potential mechanisms. Results Twenty-seven percent of participants were born in the Stroke Belt with a mean age of 69(SD=9) at STAR. Stroke Belt birth was associated with worse late-life executive function (β(95% CI):-0.18(-0.33,-0.02)) and semantic memory (-0.37(-0.53, -0.21)), but not verbal episodic memory (-0.04(-0.20, 0.12)). Adjustment for SES and CVRF attenuated associations of Stroke Belt birth with cognition (executive function (-0.05(-0.25, 0.14)); semantic memory (-0.28(-0.49, -0.07))). Conclusion Black Americans born in the Stroke Belt had worse late-life cognition than those born elsewhere, underscoring the importance of early-life exposures on brain health.
Article
Full-text available
A large literature suggests that the organization of words in semantic memory, reflecting meaningful relations among words and the concepts to which they refer, supports many cognitive processes, including memory encoding and retrieval, word learning, and inferential reasoning. The co‐activation of related items has been proposed as a mechanism by which semantic knowledge influences cognition, and contemporary accounts of semantic knowledge propose that this co‐activation is graded—that it depends on how strongly related the items are in semantic memory. Prior research with adults yielded evidence supporting this prediction; however, there is currently no evidence of graded co‐activation early in development. This study provides the first evidence that in children the co‐activation of related items depends on their relational strength in semantic memory. Participants (N = 84, age range: 3–9 years) were asked to identify a target (e.g., bone) amid distractors. Children's responses were slowed down by the presence of a related distractor (e.g., puppy) relative to unrelated distractors (e.g., flower)—suggesting that children co‐activated related items upon hearing the name of the target. Importantly, the degree of this co‐activation was predicted by the strength of the target–distractor relation, such that distractors more strongly related to the targets slowed down children to a larger extent. These findings have important implications for understanding how organized semantic knowledge affects other cognitive processes across development.
Article
Full-text available
What determines human ratings of association? We planned this paper as a test for association strength (AS) that is derived from the log likelihood that two words co‐occur significantly more often together in sentences than is expected from their single word frequencies. We also investigated the moderately correlated interactions of word frequency, emotional valence, arousal, and imageability of both words (r's ≤ .3). In three studies, linear mixed effects models revealed that AS and valence reproducibly account for variance in the human ratings. To understand further correlated predictors, we conducted a hierarchical cluster analysis and examined the predictors of four clusters in competitive analyses: Only AS and word2vec skip‐gram cosine distances reproducibly accounted for variance in all three studies. The other predictors of the first cluster (number of common associates, (positive) point‐wise mutual information, and word2vec CBOW cosine) did not reproducibly explain further variance. The same was true for the second cluster (word frequency and arousal); the third cluster (emotional valence and imageability); and the fourth cluster (consisting of joint frequency only). Finally, we discuss emotional valence as an important dimension of semantic space. Our results suggest that a simple definition of syntagmatic word contiguity (AS) and a paradigmatic measure of semantic similarity (skip‐gram cosine) provide the most general performance‐independent explanation of association ratings.
Article
Full-text available
Previous research has suggested that distributional learning mechanisms may contribute to the acquisition of semantic knowledge. However, distributional learning mechanisms, statistical learning, and contemporary “deep learning” approaches have been criticized for being incapable of learning the kind of abstract and structured knowledge that many think is required for acquisition of semantic knowledge. In this paper, we show that recurrent neural networks, trained on noisy naturalistic speech to children, do in fact learn what appears to be abstract and structured knowledge. We trained two types of recurrent neural networks (Simple Recurrent Network, and Long Short-Term Memory) to predict word sequences in a 5-million-word corpus of speech directed to children ages 0–3 years old, and assessed what semantic knowledge they acquired. We found that learned internal representations are encoding various abstract grammatical and semantic features that are useful for predicting word sequences. Assessing the organization of semantic knowledge in terms of the similarity structure, we found evidence of emergent categorical and hierarchical structure in both models. We found that the Long Short-term Memory (LSTM) and SRN are both learning very similar kinds of representations, but the LSTM achieved higher levels of performance on a quantitative evaluation. We also trained a non-recurrent neural network, Skip-gram, on the same input to compare our results to the state-of-the-art in machine learning. We found that Skip-gram achieves relatively similar performance to the LSTM, but is representing words more in terms of thematic compared to taxonomic relations, and we provide reasons why this might be the case. Our findings show that a learning system that derives abstract, distributed representations for the purpose of predicting sequential dependencies in naturalistic language may provide insight into emergence of many properties of the developing semantic system.
Article
Full-text available
In our recent book, we present a parallel distributed processing theory of the acquisition, representation and use of human semantic knowledge. The theory proposes that semantic abilities arise from the flow of activation amongst simple, neuron-like processing units, as governed by the strengths of interconnecting weights; and that acquisition of new semantic information involves the gradual adjustment of weights in the system in response to experience. These simple ideas explain a wide range of empirical phenomena from studies of categorization, lexical acquisition, and disordered semantic cognition. In this précis we focus on phenomena central to the reaction against similarity-based theories that arose in the 1980's and that subsequently motivated the "theory-theory" approach to semantic knowledge. Specifically, we consider i) how concepts differentiate in early development, ii) why some groupings of items seem to form "good" or coherent categories while others do not, iii) why different properties seem central or important to different concepts, iv) why children and adults sometimes attest to beliefs that seem to contradict their direct experience, v) how concepts reorganize between the ages of 4 and 10, and vi) the relationship between causal knowledge and semantic knowledge. The explanations for these phenomena are illustrated with reference to a simple feed-forward connectionist model; and the relationship between this simple model, the broader theory, and more general issues in cognitive science are discussed.
Article
Full-text available
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, thematic). The goal of the current research was 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 experiments, 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.
Article
Recent research reported the surprising finding that even 6-mo-olds understand common nouns [Bergelson E, Swingley D (2012) Proc Natl Acad Sci USA 109:3253-3258]. However, is their early lexicon structured and acquired like older learners? We test 6-mo-olds for a hallmark of the mature lexicon: cross-word relations. We also examine whether properties of the home environment that have been linked with lexical knowledge in older children are detectable in the initial stage of comprehension. We use a new dataset, which includes in-lab comprehension and home measures from the same infants. We find evidence for cross-word structure: On seeing two images of common nouns, infants looked significantly more at named target images when the competitor images were semantically unrelated (e.g., milk and foot) than when they were related (e.g., milk and juice), just as older learners do. We further find initial evidence for home-lab links: common noun "copresence" (i.e., whether words' referents were present and attended to in home recordings) correlated with in-lab comprehension. These findings suggest that, even in neophyte word learners, cross-word relations are formed early and the home learning environment measurably helps shape the lexicon from the outset.
Article
Functional MRI studies have revealed that the brain activity in the anterior temporal cortex during memory retrieval increases over months after memory encoding. Behavioral evidence has demonstrated that long-term memory can sometimes be consolidated more rapidly in one or two days. In the present functional MRI study, we manipulated the relatedness between paired faces to be retrieved in a pair-association task. The brain activity in the anterior temporal cortex during retrieval of paired associates increased rapidly in one day, as shown in previous studies. We found that the speed of the brain activity development was dependent on the level of semantic relatedness of paired faces. The results suggest that the semantic relatedness enhances the speed of formation of memory representation in the anterior temporal cortex.
Article
Models of category learning have been extensively studied in cognitive science and primarily tested on perceptual abstractions or artificial stimuli. In this paper, we focus on categories acquired from natural language stimuli, that is, words (e.g., chair is a member of the furniture category). We present a Bayesian model that, unlike previous work, learns both categories and their features in a single process. We model category induction as two interrelated subproblems: (a) the acquisition of features that discriminate among categories, and (b) the grouping of concepts into categories based on those features. Our model learns categories incrementally using particle filters, a sequential Monte Carlo method commonly used for approximate probabilistic inference that sequentially integrates newly observed data and can be viewed as a plausible mechanism for human learning. Experimental results show that our incremental learner obtains meaningful categories which yield a closer fit to behavioral data compared to related models while at the same time acquiring features which characterize the learned categories. (An earlier version of this work was published in Frermann and Lapata .).