ArticlePDF Available

Continuity and change in the development of category structure: Insights from the Semantic Fluency task

Authors:

Abstract

Children aged between 5 and 10 years old were tested on a semantic fluency (freelisting) task for two categories: animals and body parts. Additive tree analysis (Sattath & Tversky, 1977) was used to cluster items based upon both their proximity in the generated lists and their frequency of cooccurrence; the resulting trees, together with production frequency data, were compared across three age groups. For the animals category, this analysis revealed that although older children named proportionally more nonmammals, at all ages children tend to cluster animals according to their environmental context. For body parts, the analysis showed more parts, particularly internal organs, named with age and a cluster of face parts generated by all age groups. A novel feature of the current research was the use of statistical measures of additive tree similarity. The results are discussed with respect to theories of developmental change in the organisation of conceptual memory, and are viewed as supporting an assumption of continuity with age in the use of schematic relations in category structure. Insights are drawn from connectionist modelling to help explain the persistence, throughout childhood, of early forms of memory organisation.
Continuity and Change in the Development of Category
Structure: Insights from the Semantic Fluency Task
Running title: Category structure in semantic fluency
Samantha J. Crowe, Department of Psychology, University of Liverpool, UK.
Tony J. Prescott, Department of Psychology, University of Sheffield, UK.
Address for correspondence:
Dr Tony J. Prescott,
Department of Psychology,
University of Sheffield,
Western Bank, Sheffield,
S10 2TP, UK.
Email: t.j.prescott@sheffield.ac.uk
Telephone: +44 114 2226547
Fax: +44 114 2766515
Manuscript no. JUL01-096
1
Continuity and Change in the Development of Category
Structure: Insights from the Semantic Fluency Task
Samantha J. Crowe and Tony J. Prescott
Abstract
Children aged between 5 and 10 years old were tested on a semantic fluency
(freelisting) task for two categories: animals and body parts. Additive tree analysis
(Sattath & Tversky, 1977) was used to clusters items based upon both their proximity
in the generated lists and their frequency of co-occurrence; the resulting trees together
with production frequency data were compared across three age groups. For the
animals category, this analysis revealed that although older children named
proportionally more non-mammals, at all ages children tend to cluster animals
according to their environmental context. For body parts, the analysis showed more
parts, particularly internal organs, named with age and a cluster of face parts generated
by all age groups. A novel feature of the current research was the use of statistical
measures of additive tree similarity. The results are discussed with respect to theories
of developmental change in the organization of conceptual memory, and are viewed as
supporting an assumption of continuity with age in the use of schematic relations in
category structure. Insights are drawn from connectionist modeling to help explain the
persistence, throughout childhood, of early forms of memory organization.
2
Introduction
How does the structure of conceptual memory change during the early school
years? Between the ages of five and ten children show a massive increase in
knowledge—they know more concepts (Clark, 1995), they know much more about
the relationships between concepts (Kail, 1990), they make more inferences on the
basis of their conceptual knowledge (Paris, Lindauer & Cox, 1977), and they make
more sophisticated ontological distinctions (Keil, 1983). The nature of children's
conceptual knowledge also appears to change. For instance, younger children show a
bias towards organizing concepts schematically (or thematically), that is, in relation to
event or scenes (Nelson & Gruendel, 1981; Mandler, 1983; Nelson, 1983), whilst
older children and adults seem to think of concepts primarily in terms of their
similarity (functional or perceptual) to each other, that is, in terms of their taxonomic
relations (Bjorklund, 1985; Lucariello, Kyratzis & Nelson, 1992). As they grow older
children also make different inferences on the basis of their knowledge or ‘theories’
about the world (Carey, 1985), and appear to have more and better strategies for
storing and retrieving conceptual information (Kail, 1990).
An important and enduring question with regard to many of these changes is whether
they involve a radical reorganization of conceptual memory, or whether they arise
from more continuous and incremental forms of change in which new information is
assimilated to earlier memory structures that retain key aspects of their original form
(Mandler, 1983; Eimas, 1994). In this article we use children's production of animal
and body-part terms in the semantic fluency task to investigate these questions and
report findings that support a view of continuity of memory organization over the
period studied.
3
The semantic fluency procedure, also termed conceptual fluency, freelisting, or
category production, requires participants to generate as many exemplars of a category
as possible in a given length of time. When a word or concept is activated in memory
(and then spoken) we assume that it will in turn activate other words or concepts
which are semantically similar or associatively related to it. This assumption, which is
well supported by evidence from numerous semantic priming studies (see Neely,
1991), implies that the order in which words are produced in the fluency task will
provide an indirect measure of the psychological proximity of the items generated.
Henley (1969) and Storm (1980) provide additional evidence for this assumption by
demonstrating that animals of close psychological proximity (as indicated by other
semantic memory tasks such as pair rating, sorting, and verbal association) are named
in close proximity in lists. Performance in the semantic fluency task also shows a
number of consistent characteristics in both children and adults. First, the rate of
production of new items over the duration of the task slows with time following a
hyperbolic decline (Gruenewald & Lockhead, 1980; Kail & Nippold, 1984; Grube &
Hasselhorn, 1996). Second, items that are more typical of a category tend to be
produced with higher relative frequency (i.e. by more subjects) than poorer category
exemplars (Henley, 1969; Uyeda & Mandler, 1980; Kail & Nippold, 1984; Grube &
Hasselhorn, 1996). Third, the most prototypical category members tend to be produced
first, followed by familiar (but not prototypical) items, followed by atypical or unusual
items (Kail & Nippold, 1984). Finally, items are generally produced in short bursts of
conceptually-related words (Bousfield & Sedgewick, 1944; Gruenewald & Lockhead,
1980). Many of these findings support the view that the arrangement of items in the
list reflects important aspects of the underlying conceptual structures.
4
Several studies have employed the semantic fluency task with the aim of
understanding changes in memory organization during childhood (Nelson, 1974;
Storm, 1980; Lucariello et al., 1992; Grube & Hasselhorn, 1996), the category of
animals being most widely studied.
In the earliest study of this kind, Nelson (1974) found that 5-year olds produced, on
average, half as many items as 8-year olds when freelisting nine natural language
categories. For the animal category she found that children predominantly respond by
naming mammals, with older children naming relatively more non-mammals (species
of birds, reptiles, etc.) and members of mammalian sub-classes (e.g. breeds of dogs)
than younger children. This finding was interpreted as showing an expansion of the
hierarchical organization of the animal category with age. The two groups studied also
differed in their most frequent responses with the 8-year olds naming domestic
animals (dog, cat, and horse) most often, and the 5-year olds naming wild animals
(giraffe, lion, elephant, tiger) with greater frequency.
Whereas Nelson looked at conceptual fluency just in children, Storm (1980) studied
the freelisting of the animal category across a much broader ranger or subject
groups— children (6- and 9-year olds), teenagers, college students, and zoology PhD
candidates. Her results confirmed Nelson's finding of increased productivity with age
and educational experience. For each experimental group, and for 25 selected animal
terms, Storm also computed a mean inter-item similarity score for all pairs of concept
words based on their proximity in the generated lists. This measure was used as input
to a hierarchical clustering algorithm that formed tree structures in which the most
proximal items named by each group were clustered together. Surprisingly, Storm
found no obvious differences between the trees generated in this way for the different
5
groups studied. At all ages, subjects clustered animals predominantly according to
their habitat/environmental context. Specifically, farm animals (pig, cow, horse),
domestic animals (cat, dog, mouse), and wild/zoo animals (lion, leopard, elephant,
giraffe), tended to be grouped together. In this respect even the trees generated for PhD
zoology candidates, for whom biological criteria for categorization can be presumed to
be highly salient, were not dissimilar from those produced by 6-year old kindergarten
children—in all subjects the primary mode of mode of organization in freelisting
appears to have been a schematic one.
In a further study comparing young children with adults, Lucariello, Kyratzis, and
Nelson (1992) tested 4- and 7- year old children and college students on a semantic
fluency task for five categories: clothes, animals, food, furniture, and tools. A
hierarchical clustering methodology was again used to elucidate the conceptual
structures underlying the lists produced by different age groups, this time analyzing,
for each age group, all items produced by at least three subjects in that group. For the
4-year-olds this analysis showed a main cluster of zoo animals, and for the 7-year-
olds, clusters composed of domestic animals, zoo animals, forest animals, and aquatic
animals. For the adults, the analysis revealed a substantial number of clusters that
could again be classified in terms of habitat/environmental context, with a cluster of
‘primates’ (not present in the children’s productions) that included humans. Whilst the
use of environmental context in many of these clusterings was again interpreted as
showing conceptual organization by schematic (i.e. event-based) relations, Lucariello
et al. also interpreted their results (together with those for categories) as supporting an
increase in the use of conventional taxonomic relations with age.
6
Finally, Grube and Hasselhorn (1996), looked at the performance of 8-year olds,
compared with older children, on the semantic fluency task for the animal category.
They found that both age groups clustered items in terms of environmental context
(confirming earlier findings), and produced similar typicality structures; that is, the
same animals were named at high frequency at both ages. Interestingly, the degree of
environmental clustering in the lists of these children was related to productivity (the
total number of items generated) only for the older group. This finding was interpreted
as showing that some 10-year olds use strategies that exploit the organization of their
conceptual memory structures to allow more effective memory retrieval. These
researchers also found an increase in the use of strong inter-item associations (e.g.
dog—cat, lion—tiger) in older children, a factor which was again linked with greater
productivity and strategy use.
In summary, much of the previous work on developmental change in the semantic
fluency task has emphasized quantitative rather than qualitative change in the category
lists generated by children—older children generate more items (possibly making
more use of strategic retrieval processes) but appear to cluster semantic knowledge in
similar ways to their younger counterparts. However, data on the youngest children (4-
6 year olds) is limited with only three studies performed to date that have generated
somewhat inconsistent results. Nelson (1974) found changes in typicality ratings
(production frequencies) between 5- and 8-year olds, and has argued for an increase in
the hierarchical organization of conceptual knowledge of this time period, Storm's
results indicated no clear differences in memory organization between 6-year-olds and
older children, while Lucariello et al. (1992) found evidence to support both continuity
in the use of schematic (spatiotemporal) relations with age, and some evidence of
increased use of taxonomic relations with age. Considered within the wider frame of
7
semantic memory research, these studies also stand in interesting contrast to research
that suggests significant changes in the organization of conceptual knowledge over the
same time period (see Bjorklund, 1985; Carey, 1985 for review).
One possible limitation of the research described above is that the pattern of results
obtained with the frequently studied category of animals may not provide a
representative picture of general conceptual change in semantic memory. Although,
Lucareillo et al. (1992) found similar results across three categories (food, animals,
and clothes), they also found ambiguous and somewhat inconclusive results for two
other categories (furniture and tools) that are poorly developed in young children. The
current study therefore investigated a further category of conceptual knowledge that is
well represented in young children’s vocabulary and may undergo significant
development between the ages of 5 and 10. Specifically, we examined semantic
fluency for the category of human body parts. Substantial changes in children's
knowledge of body parts (particularly internal ones) are known to be occur within this
age range (Gellert, 1962), and it has been suggested (Carey, 1985) that a major shift in
the understanding of the body occurs between the ages of eight and ten as children
develop intuitive ‘biological’ theories of how the body works. Since the body parts
category has received little attention in studies of children’s semantic fluency, yet
remains a subject of considerable theoretical interest (see, e.g. Jaakkola & Slaughter,
2002), an investigation into its conceptual structure as revealed by the freelisting task
seems timely.
Methodological innovations in the current study
8
The current study was designed to look for specific evidence of continuity or change,
between the ages of 5 and 10, in the conceptual memory structures underlying
semantic fluency for the animals and body parts categories. Our basic methodology is
similar to Storm (1980) and Lucariello et al. (1992) in that we employ cluster analysis
techniques; however, relative to these other studies we have made a number of
innovations.
First, Storm (1980) analyzed the same 25 animal concepts in all groups. These were
animals that a majority of the youngest children were able to name in a separate
picture naming task rather than those which occurred with high frequency during the
fluency task. This restriction clearly excludes from the analysis any differences
between age groups arising from changes in production frequencies and therefore may
obscure significant developmental change. In contrast, Lucariello et al. (1992)
analyzed different sets of items at each age group, with the selection of items based on
their production frequencies. Relative to Storm’s procedure this methodology is likely
to highlight differences and mask continuity between age groups. The present study
focuses, therefore, on the most frequent responses made during the category fluency
task both within and across age groups, so that the data can be appropriately analyzed
for both similarities and differences between groups.
Second, most previous studies have used the distance between items within lists as
their measure of inter-item similarity for input to a hierarchical clustering algorithm.
However, a second factor of potential significance for estimating similarity, which is
largely over-looked in fluency studies, is the number of times, over a group of
subjects, that two items are mentioned together. In particular, when each subject
generates a relatively small number of items overall, we propose that items which have
9
a high psychological proximity will be more likely to co-occur in lists than items with
low proximity. We have therefore developed a measure of inter-item similarity based
on both within-list proximity, and across-list item co-occurrence. We propose that this
combined measure, which is derived analytically using expected probability
distributions, provides a robust estimate of inter-item similarity that is particularly
appropriate for analyzing fluency data from children of 6 years of age or less whose
category listings often contain fewer than ten items.
Third, to perform the hierarchical cluster analyses we have used the additive tree
(addtree) algorithm developed by Sattath & Tversky (1977). In comparison to some
other hierarchical clustering techniques, the addtree technique places fewer a priori
constraints on the relationships between items and can therefore provide a more
faithful representation of the data. A further limitation of studies using hierarchical
clustering is that comparisons between trees are generally made on a qualitative basis.
A unique feature of the current study is that we have used the Bk measure devised by
Fowlkes and Mallows (1983) to provide appropriate statistical measures of the
similarities between two hierarchical clusterings.
Methods
Participants
The subjects for this study were 155 children from two schools, both in Sheffield, UK.
These children are predominantly from white, urban families of social class II and III.
The children were divided into three groups on the basis of age/school year group: 50
children in year 1, mean age 5:9 years, range 5:1 to 6:4; 55 in year 3, mean age 7:7,
range 7:0 to 8:3; and 50 in year 5, mean age 10:0, range 9:0 to 10:4. All of the children
10
spoke English as their first language. All subjects completed the semantic fluency task
for both the animals and body parts categories, the additional categories of food items,
clothes, vehicles, and plants were also tested, and summary statistics for these other
categories are reported below. The order in which categories were listed was
randomized across subjects.
Administration of Semantic Fluency Tasks
The children were tested individually at school, in a quiet area. For each category,
each child was asked to “Tell me all the _______s you can think of”. A one minute
interval was allowed for each category. Although some previous studies have allowed
a longer time period, the one minute interval was selected here on the basis that
subjects generate items at a much quicker rate early on in the fluency task, and also
tend to generate more typical items toward the start of the task. Pilot studies also
indicated that some of the youngest children found it difficult to focus on the task for
more extended periods. Identical instructions were used with all the children, however,
the younger children sometimes required encouragement. Where encouragement was
given, care was taken not to influence the child’s responses. An example would be
“Can you think of any (more) _______s?” or “Which other ________s do you know?”
Finally, children were asked to shut their eyes throughout the body parts task so that
they would be unable use their own body, or that of the experimenter, as a visual cue.
Obtaining Similarity Scores
The fluency tasks provided separate lists of category items for each child, in
production order. For each category, and each year group, matrices of inter-item
11
similarities were derived using a metric we have developed to control for differences
in production frequency between items and age groups. This metric has two
component measures, termed
and
w below, one based on within-list item proximity
(
) the other on across-list item co-occurrence (
w). In the following we describe how
each measure is calculated and then combined to form the overall inter-item similarity
metric
w. Note that for all of the measures defined here, increasing value indicates
decreasing similarity (or increasing dissimilarity).
The measure,
, of within-list item proximity is defined first. Let a and b be two non-
identical category items that occur at the index positions
i
al
and
i
bl
in the category list
l, containing
n
l
total items, generated by a given participant. Assume, initially, that a
and beach occur only once in list l; in this case we define the normalized inter-item
distance between a and b as the absolute value of the distance between the two items
in the list divided by the total length of the list, i.e.
ial ibl
nl
.
Using
n
l
as a normalization term is appropriate here since the expected probability
distribution for values of
i
al
i
jl has a standard deviation that increases in an
approximately linear manner with list lengthi.
Participants sometimes repeat one or more items when generating a category listing. If
either of the items a or b is repeated in list l we choose our measure of inter-item
distance
abl
to be the smallest of all such distances
dabl mina,bl
ial ibl
nl








.
12
However, the expected probability distribution will also be significantly narrower for
repeated items, so it is appropriate in this situation to include an additional scaling
term. Let
n
al
and
n
bl
be the number of occurrences of a and b respectively in list l.
Based on an analysis of the relevant expected probability distributions, we have found
that a suitable scaling termii is given by
(
n
al
n
bl
)
where
(
1
)
1
.
0
,
(
2
)
0
.
67
,
(
3
)
0
.
5
,
(
4
)
0
.
41
,
(
6
)
0
.
28
. These values cover all the occasions of repeated
items in our data-set.
Let Fab be the total number of subjects in a given group whose responses contained the
pair a and b. We can now define
such that
(a,a)0,
(a,b)1
Fab
dabl
(nal nbl )
l(albl)








.**
In other words,
provides a measure of inter-item similarity that is zero (maximally
similar) between an item and itself, greater than zero for all non-identical items a and
b, and increases with the normalized inter-item distance between a and b averaged
over all participants who named both items at least once in their category list.
Next we define the second measure of inter-item similarity,
w
based on across-list
item co-occurrence, which can be combined multiplicatively with
to form the
overall inter-item similarity metric
w.
Let
f
aand
f
b be the number of participants in the group naming items a and b
respectively, and let N be the total size of the group. The expected number of co-
occurrences of items a an b is then given byiii
** Corrected from the published version.
13
E(Fab)
f
a
f
b
N
.
We now calculate the normalized number of co-occurrences as the difference between
this expected value and the observed value
F
ab ,
C
ab
E
(
F
ab
)
F
ab
.
Let the maximum and minimum obtained values of
C
ab for a given group of
participants be
C
and
C
respectively, and let S1max C,C
.
is now defined
as
w(a,a)0,
w(a,b)1w
C
ab
S
.
Here S provides a scaling factoriv such that
1
C
ab
/
S
1
for all a and b, and w
(
0
w
1
) is a weight used to determine the relative contribution of
w
to the
combined measure of item similarity
w
Note that
w
will be equal to 1 if the
observed number of co-occurrences matches the expected frequency, thus making no
net contribution to
w
. If the number of co-occurrences is greater than expected,
w
will be less than 1 reducing the overall metric proportionately. Finally, if the number
of co-occurrences is less than expected,
w
will be greater than 1 thereby increasing
the size of
w
. As the number of participants naming either item a or b approaches
N (the size of the group),
C
ab becomes a less useful indicator of psychological
proximity (to see this, observe that at
f
a
N
the variance in
C
ab will be zero, every
occurrence of item b will necessarily be a co-occurrence with item a!). However, in
such situations
w
will take a value close to 1, giving a combined similarity metric
w
that depends almost entirely on the within-list proximity measure
. It is
14
therefore appropriate to use
w
as part of a combined measure regardless of the
expected frequency of individual category items.
In the addtree analyses reported below all matrices of similarity data were generated
using the metric
0
.
5
as the value w= 0.5 was found to generate a suitable balance
between the and
for all categories and groupsv.
15
Results
Age related trends in productivity and production frequencies (typicality) are
described first, followed by the addtree analyses of category structure. Except where
stated all results are significant at p<0.01.
Productivity
One-way ANOVA revealed substantial overall increases with age in the mean number
of responses generated for all categories. The means, standard deviations and F-values
are shown in table 1. Post hoc tests (Fisher’s PLSD) showed significant increases in
the mean number of items produced with increasing age in all categories, except
between year 3 and year 5 groups for the clothes category (for the vehicles category
the difference between year 3 and year 5 was significant at p<0.05). There were also
significant differences between the number of items listed in the different categories
for all three year groups (repeated measures ANOVA: Year 1, F(5)= 39.82; Year 3
F(5)= 51.86; Year 5, F(5)= 68.95). In all years, the body parts category generated the
most responses followed by the animals category, clothes and foods generated an
intermediate numbers of responses, and vehicles and plants the fewest. In view of the
higher productivity for animals and body parts, and the theoretical importance of
these two categories as outlined in the introduction, the remaining analyses focus on
these two categories exclusively.
**** TABLE 1 ABOUT HERE ****
16
Production frequencies
The production frequency for any given category item is calculated as the proportion
of children in a selected group who named that item at least once. Tables 2 and 3
show, for each of the three year groups, the production frequencies of items in the
animals and body parts categories generated by at least 20% of the children in that
group.
For each category, product-moment correlations were calculated for the production
frequencies of the items named by a least 5% of participants in all age-groups (there
were 34 such items in the animals category, 28 in the body parts category). For the
animals tasks this gave the correlations: Year 1 and Year 3 r(32) = 0.84, Year 1 and
Year 5 r(32) = 0.77, Year 3 and Year 5 r(32) = 0.78. For the body parts task Year 1
and Year 3 r(26) = 0.89, Year 1 and Year 5 r(26) = 0.63, Year 3 and Year 5= 0.83.
The strength of these correlations indicates considerable similarity between the typical
responses generated by the different year groups.
Significant correlations were also found with the results of previous semantic fluency
studies (Posnansky, 1978; Grube & Hasselhorn, 1996). For instance, Posnansky
(1978) calculated production frequencies for 25 different categories including animals
and body parts for children in grades 2, 3, 4, and 6, for a 1-minute, written, freelisting
task. Comparing the year 3 and year 5 groups in the current study with the nearest
equivalent groups in Posnansky's study (Grades 2 and 4 respectively), gave
correlations for the animals task of r(17) = 0.80 (year 3 and grade 2) and r(20) = 0.74
(year 5 and grade 4), and for the body parts task of r(19) = 0.79 (year 3 and grade 2)
and r(19) = 0.59 (year 5 and grade 4).
17
Morrison, Chappell, and Ellis (1997) have provided estimates of the age of acquisition
of many of the animal terms investigated here. Production frequencies of animals
terms from the current study showed a significant negative correlation with their
'objective age of acquisition (75%)' measure for all three age groups—year 1 r(22) = -
.48 (p<0.05), year 3 r(22) = -0.57, and year 5 r(22) = -.57. This result indicates that
some of the most frequently produced responses for the animal category are also those
that are acquired first.
Finally, comparisons were made with adult typicality ratingsvi, collected by Uyeda and
Mandler (1980), for items in the categories ‘four-footed animals’ and ‘parts of the
human body’. This gave the correlations: for the animals category, year 1 r(16)= -0.68,
year 3 r(16)= -0.57 (p<0.05), and year 5 r(16)= -0.59; and for the body parts category,
year 1 r(18)= -0.73, year 3 r(18)= -0.82 (p<0.05), and year 5 r(18)= -0.67. Thus,
production frequencies, at all ages, appear to provide a good indication of item
typicality at least with respect to adult norms.
**** TABLE 2 ABOUT HERE ****
**** TABLE 3 ABOUT HERE ****
Composition of category data
Following Nelson's (1974) finding that 8-year-olds name more non-mammals than 5-
year-olds, the ratio of non-mammals to total animals was compared across age groups.
This measure, with group medians (ranges) of 0.20 (0–0.67) for year 1, 0.22 (0–0.83)
for year 3, and 0.29 (0–0.77) for year 5, indicated that all age groups named mammals
18
predominantly, and that older children named significantly more non-mammals
(Kruskal-Wallisvii H(2)=6.17p<0.05, sum of ranks= 3362.0, 4254.0, and 4474.0 for
years, 1, 3, and 5 respectively).
Nelson also noted a predominance of ‘wild’ animals in the top five responses of 5 year
olds (giraffe, lion, elephant, tiger, horse). In the current study the top five of the year 1
group (mean age 5.9) includes two wild animals (lion, tiger, cat, dog, horse).
However, from table 2 it is clear that the older children also name some ‘wild’ animals
with high frequency. An interesting trend in the current data-set is the apparent
increase in typicality with age for two domestic animals, cat and dog, relative to other
animal terms. Specifically, in the year 1 group, the average production frequency for
these animals (0.46) is only slightly greater (0.06) than the average production
frequency for the overall top ten items (0.40). In years 3 and 5, however, the same
comparison shows a difference of 0.22 (0.68 - 0.46), and 0.26 (0.87 - 0.61)
respectively. This difference suggests that between the ages of 6 and 8 there may be a
substantial increase in the status of cat and dog as prototypical animals.
Carey (1985), following Gellert (1962) and others, has argued for a significant change
in children’s conceptual understanding of internal body parts between the ages of 8
and 10. Data from the current study was therefore analyzed to determine the ratio of
internal organsviii (i.e. heart, brain, lung, etc.) to total body parts named by each age
group. This measure, with group medians (ranges) of 0.0 (0–0.5) for year 1, 0.07 (0–
0.75) for year 3, and 0.21 (0–0.64) for year 5, indicated a significant increase with age,
with year 5 children in particular, naming many more internal organs than the
youngest group (Kruskal-Wallis: H(2)=38.78, p<0.01, sum of ranks= 2749.0, 3963.5,
and 5377.5 for years, 1, 3, and 5 respectively).
19
Representing category structure
Several addtree analyses were carried out for each of the category/group
w
similarity matrices using the ADDTREE/P program written by Corter (1982). First, we
were interested in analyses that revealed the conceptual structure of children’s
knowledge, taking into account increasing productivity and changes in production
frequency with age. Target concepts for these within-group analyses were chosen on
the basis that they had been produced by at least 20% of the children in the particular
age group being considered. In order to perform these analyses items were excluded if
they failed to co-occur at least once with each of the other items in the set, since, under
these circumstances, a full matrix of inter-item distances cannot be computed. For this
reason a small number of the items that appear in Tables 1 and 2 are excluded from
the corresponding addtree analyses. Second, we wanted to allow direct comparison of
related conceptual structures across age groups. Target concepts for these analyses
were chosen on the basis that they had been produced by at least 20% of participants
in the youngest age group (year 1), and co-occurred at least once in all three groups.
Interpreting Addtrees
Several factors are of note when interpreting addtrees.
First, a tree’s ‘goodness of fit’ to the data is expressed as a stress value (the measure
reported is Kruskal’s stress measure) ranging from 0 (best possible fit) to 1 (worst
possible fit). High stress values may indicate that a tree is not an adequate
representation of the similarity structure of the data. The statistic r2 is also reported,
which provides a measure of the proportion of the variance of raw distances (i.e. the
matrix of similarities used as input to the analysis) that is accounted for by the distances
shown in the tree, thus tree representations which most accurately fit the data will have
low stress values and high r2 values.
20
A second factor for consideration is the division of concepts into clusters. Each addtree
is composed of nodes (horizontal lines in the figures displayed below) and arcs (vertical
lines). Concepts are represented by external nodes and are formed into clusters by
internal nodes; all nodes are joined by arcs. The distinctiveness of a cluster and the
degree of similarity within a cluster are both indicated by the length of nodes, with the
shortest lengths indicating the greatest similarity. For any two items in the tree, their
inter-item similarity is indicated by the sum of the lengths of the nodes in the path
between the two items (again shorter= more similar). Note that the length of arcs has no
significance.
Finally, in comparing across age groups, some measure is needed of the similarity of
two trees with identical sets of external nodes but different internal structure (note, no
statistical measures exist for comparison of trees with different sets of external nodes).
Two trees may be similar at one level of clustering but dissimilar at another, for this
reason we have used the statistic Bk developed by Fowlkes and Mallows (1983), which
provides multiple measures of similarity at different levels of clustering. Further
explanation of the calculation and interpretation of the Bk statistic is given below.
Analysis of Conceptual Structure for the Animals Task
Figure 1 shows addtree analyses of animal names for all three year groups, where items
were selected, for each year group, on the basis that they were listed by at least 20% of
the children in that group (year 1 stress = 0.094, r2 = 0.712; year 3 stress = 0.072, r2 =
0.702; year 5 stress = 0.079, r2 = 0.660).
**** FIGURE 1 ABOUT HERE ****
21
In accordance with previous findings (Storm, 1980; Lucariello et al., 1992; Grube &
Hasselhorn, 1996) these analyses show that typical environmental context is a key
organizing factor in children's freelisting of the animal category. In year 1, two major
clusters are evident containing domestic/farm animals (cat, dog, cow, horse, sheep, pig)
and wild/zoo animals (elephant, giraffe, monkey, lion, tiger). Farm animals form their
own distinct sub-cluster within the larger domestic/farm animal group. Year 3 shows a
cluster containing domestic pets (cat, dog, rabbit, hamster, mouse), and, within the
second branch of the main tree, sub-clusters of wild/zoo animals (elephant, giraffe,
lion, tiger, kangaroo, monkey), farm animals (horse, cow, sheep, pig), and aquatic
animals (fish, dolphin, shark, whale). Finally, the year 5 tree again has a large cluster of
predominantly wild/zoo animals (giraffe, elephant, monkey, lion, tiger), a cluster of
farm animals (horse, cow, pig, sheep), and several smaller clusters including the pairs
(cat, dog), (mouse, rat), (hamster, snake), and the triplet (fish, fox, rabbit). Whilst the
interpretation of some of these smaller clusters is less obvious (fish, fox, rabbit, for
instance could be a collection of British wild animals), the general trend toward
clustering by environmental context seems clear. The term bird appears in all three
trees but is not consistently clustered with a specific group, perhaps because of its status
as a super-ordinate class that includes animals falling into many different schematic
categories.
Figure 2 shows addtree analyses of animal names for all three year groups, where items
were selected, for all year groups, on the basis that they were listed by at least 20% of
the children in the year 1 group and co-occurred at least once in all three year groups
(year 1 stress = 0.094, r2 = 0.712; year 3 stress = 0.064, r2 = 0.854; year 5 stress =
22
0.079, r2 = 0.862). Note that, for the year 1 group, the trees generated for this analysis
and the previous one in Figure 1 are identical.
**** FIGURE 2 ABOUT HERE ****
In each tree in Figure 2 the internal nodes are numbered according to the number of
clusters (k) formed if the tree is split below that point. For example, for the Year 1 tree,
the dotted line in Figure 2 shows the decomposition of the tree into k= 2 clusters. The
comparison statistic Bk, devised by Fowlkes and Mallows (1983), provides a measure of
the similarity of two trees at each level of clusteringix for k= 2, …, n-1, where n is the
number of objects in the tree. Bk varies between 0, maximum dissimilarity, and 1,
maximum similarity. The graphs at the base of Figure 2 show plots of Bk for the
comparisons, from left to right, Year 1 with Year 3, Year 1 with Year 5, and Year 3
with Year 5. Values of Bk are shown by the diamond-shaped point plots for k= 2, …,11.
The solid line and the two dotted lines on each graph show, respectively, the expected
value of Bk and its upper and lower limits. Values of Bk outside the range indicated by
these limits can be considered significantx. On average, two-thirds of the Bk scores
(7/10, 8/10, and 5/10 in the three comparisons) lie above the upper limit of their
expected values, indicating that the trees generated for all three groups in this task are
similar at many levels of clustering.
Analysis of Conceptual Structure for the Body Parts Task
Figure 3 shows addtree analyses of body parts for all three year groups, where items
were selected, for each year group, on the basis that they were listed by at least 20% of
23
the children in that group ( Year 1 stress = 0.080, r2 = 0.740; Year 3 stress = 0.085, r2 =
0.800; Year 5 stress = 0.072, r2 = 0.708).
**** FIGURE 3 ABOUT HERE ****
The children in all years distinguished a cluster of internal parts (e.g bone, heart) and a
cluster of face parts (e.g. eye, nose, mouth), The addtree analysis clearly illustrates
children’s increasing knowledge with age of internal parts of the body. Year 5 children
named many more internal body parts than the younger age groups; furthermore, the
addtree analysis for this group shows some internal organization within the cluster of
internal parts, specifically, associated pairs of internal organs (heart—lung and liver—
kidney), and muscular-skeletal parts (bone—muscle). The associative pair (arm—leg)
appears consistently in all year groups, digits (finger, toe) either cluster with each other
or within the appendages to which they attach (hand, foot). A group of body parts than
can be thought of as joints or connectors form a distinct cluster in the year 3 group
(elbow, shoulder, knee, and neck), but appear as two separate pairs at year 5 (elbow,
knee) and (shoulder, neck). In the year 1 group neck is clustered with ‘tummy’ (the most
popular term amongst younger children for the abdomen) perhaps on a similar basis that
they are both ‘connecting’ parts.
Figure 4 shows addtree analyses of body parts for all three year groups, where items
were selected, for all year groups, on the basis that they were listed by at least 20% of
the children in the year 1 group and co-occurred at least once in all three year groups
(year 1 stress = 0.082, r2 = 0.736; year 3 stress = 0.082, r2 = 0.828; year 5 stress =
0.081, r2 = 0.862). Graphs of the Bk statistic are also shown for the comparisons, from
24
left to right, year 1 with year 3, year 1 with year 5, and year 3 with year 5. As with the
animals category, a two-thirds majority of Bkscores lie above the upper limit of their
expected values (8/13, 7/13, and 11/13 in the three comparisons) indicating similarities
at many levels of clustering between all three year groups.
**** FIGURE 4 ABOUT HERE ****
Discussion
The semantic fluency task cannot provide exhaustive access to a children’s category
knowledge, and, in the somewhat abbreviated form used here, is unlikely to provide
access to even a majority of the concepts a child knows for any given category. What it
can reveal, however, is what items of a category spring most easily to mind, and thus, in
some sense are most typical of that category, and what items within a category are most
strongly linked and are therefore likely to be recalled together. In this discussion we
first summarize the findings for the two different categories studied (animals and body
parts), and then consider the validity and generality of the results as reflections of the
organization of conceptual structure in memory. We then consider the implications of
these findings for a number of influential theories of conceptual memory development.
Finally, we suggest a possible explanation for the pattern of results shown based on
insights from connectionist modeling of memory acquisition.
Development of semantic fluency for the animals category
The degree of similarity across age groups both in terms of production frequencies (as a
measure of typicality) and clustering (as shown by the addtree analysis) suggests an
underlying continuity in the organization of memory for animal concepts. This study
25
also confirms the finding that children from an early age tend to cluster animals
primarily in terms of their typical environmental context (i.e. where animals are
experienced or might be expected to be experienced). Although older children know
and list significantly more animals, this primary mode of clustering is maintained. Thus,
8- and 10-year-olds named more non-mammals, but these were clustered together with
mammals found in similar contexts, for instance, kangaroo was found clustered with
monkey (both wild/zoo animals), while fish and shark shared an aquatic cluster with
dolphin and whale. There was little support for Nelson’s (1974) suggestion of general
change, between 5 and 8 years, in the most frequently produced items from wild
animals to domestic animals. The current data suggest instead, a more focused increase
with age in the relative frequency of two specific domestic animals—cat and dog
implying that these animals are increasingly considered to be prototypical. Since
Nelson’s subjects were slightly younger than those investigated here (mean age 5.1
years compared to 5.9 years for the current year 1 group), the prototypicality of
domestic animals for young children may warrant some further investigation.
Development of semantic fluency for the body parts category
Production frequency data for the body parts task also showed substantial similarities
across age groups. Children at all ages generated a cluster of internal body parts, and a
cluster of face parts separate from clusters of other body parts. In addition to the
expected increase in the number of body parts named with age, the oldest children also
generated a much large cluster of internal body parts and organs than the younger
children. The basis of clustering for body parts suggest two underlying dimensions of
organization. The first is a topological basis for clustering (that is naming parts that are
26
close together), based around a principle distinction between the head and the trunk.
The second is organization according to function, this is evident in the clustering of
limbs (arm—leg) and in older children of ‘joints’ (elbow, knee, shoulder), digits
(finger—toe), and related internal parts (heart—lung, kidney—liver, and bone—muscle).
The presence of an increasing number of such functional associations in the older
children suggests that this dimension of organization may become of greater importance
with age, although further research is needed to establish whether there is a reliable,
age-related trend.
Validation of results
One of the most effective techniques for validating cluster analyses is replication
(Aldenderfer & Blashfield, 1984). Similar results for the animal category found in other
studies (Storm 1980; Lucariello et al, 1992; Grube & Hasselhorn, 1996) demonstrate
that the finding of clustering by environmental context in the animals task is a robust
and repeatable one. A second group of children tested by the authors on both the
animals and body parts tasks also showed similar results to those reported here (Hartley,
1999).
A second means of validation is by comparison with other methods of analysis. Several
studies have looked at semantic fluency data for the animals category using
multidimensional scaling (MDS) techniques (Henley, 1969; Chan et al., 1993). MDS
attempts to find a small number of principle dimensions that are able to provide a good
fit to a matrix of inter-item distances. Results from studies that have been successful in
using MDS with adult animal fluency data generally complement the findings of
hierarchical clustering. Thus, for instance, Chan et al. (1993) found a principle
27
dimension corresponding to a wild/domestic distinction in semantic fluency data
(1993), and a second dimension corresponding to size, while Henley (1969) found
principle dimensions of ‘ferocity’ and ‘size’, but also noted clusters of wild and
domestic animals in animal semantic space.
Generality of results
An important question is whether the results of these semantic fluency studies reflect
generic properties of the structural organization of conceptual memory or merely
indicate characteristics of memorial processes that are specific to freelisting tasks.
Support for a generic, rather than task-specific, view comes from the studies of Henley
(1969) and Storm (1980) who have shown that semantic memory tasks such as pair
rating, sorting, and verbal association produce similar results to the semantic fluency
task for the animals category. To further investigate the generality of the current study,
Roberts and Hartley (described in Hartley, 1999) asked children to provide similarity
ratings for the twelve animals with highest production frequency in the semantic
fluency task. Three groups of children were tested, taken from years 1, 3, and 5 of a UK
primary school (i.e. the same age groups as in the fluency study). Addtree analyses of
the resulting similarity matrices showed that the principle basis for clustering in each
age group was environmental context, showing good agreement with the results of the
freelisting task. A tendency to sub-cluster items by size, as described by Chan et
al.(1993), was also evident, particularly for the oldest group (e.g. elephant—giraffe,
horse—cow). Full details of this study are given in Hartley (1999).
Production frequencies in the current study were consistent with those of other fluency
studies and, more interestingly, showed good correlations with norms for ‘age of
acquisition’ and adult typicality. These findings therefore provide strong support for the
28
assumption that production frequencies in the semantic fluency task reflect the
typicality of category members, and are not simply an artifact of the freelisting
paradigm. Bjorklund and co-workers (Bjorklund, Thompson & Ornstein, 1983;
Bjorklund, 1985) found that children’s judgments of typicality become more similar to
those of adults with age. Such a trend is not evident here in the comparison of
production frequency data to adult typicality ratings. Bjorklund et al.’s findings have,
however, been called into question by more recent evidence showing that young
children often misunderstand the task of providing typicality judgments (Maridaki
Kassotaki, 1997).
A possible source of task-specific effects in the semantic fluency task is the use of
meta-memorial strategies. For instance, Grube and Hasselhorn (1996) have suggested
that increased productivity in word generation behavior could reflect the use, in older
children, of specific strategies that would make their recall more efficient. Thus, for
instance, children may purposefully decide to produce as many farm animals as they
can, or name parts of the face as they can think of, etc. Although there is good evidence
that older children are more able to use strategies to aid their recall, Bjorklund (1985)
has pointed out the use of such strategies depends upon, and is supported by, conceptual
knowledge. Thus evidence of strategy use is not incompatible with the assumption that
freelisting behavior reflects category structure.
Implications for theories of conceptual memory development
Mandler (1983) has reviewed evidence suggesting that the conceptual knowledge of
young children is organized schematically, that is, in terms of the sort of relationships
(such as temporal or spatial contiguity) typically found in events, stories, or real-world
29
scenes. Nelson and her co-workers (Nelson & Gruendel, 1981; Nelson, 1983; Nelson,
Fivush, Hudson & Lucariello, 1983; Lucariello et al., 1992) have made a similar
proposal, suggesting that young children’s conceptual memory is organized around
representations of events, or life experiences, termed scripts. A substantial body of
research (see Mandler, 1983; Bjorklund, 1985 for review) suggests, however, a shift in
knowledge organization with age away from the use of schematic relations and towards
taxonomic ones, that is relations based on the similarity of perceptual and functional
attributes (though see Lin and Murphy (2001) for evidence of the use of schematic
relations in adults).
A further proposal for significant change in conceptual organization between ages 5 and
10 has been made by Carey (1985). Carey argues that children below the age of 8 lack a
biological framework with which to structure their understanding of either the body,
and the operation of its parts, or the animal kingdom, and the relations between its
members. Instead, young children rely on ‘psychological’ theories of the body, in
which bodily processes are accounted for in terms of what a person ‘wants or thinks’,
and make inferences about animals on the basis of their perceived similarity to people
rather than any understanding of the biological relatedness of different animal kinds.
By age 10, Carey suggests that children have come to regard themselves as biological
organisms and have attained a new functional understanding of internal bodily
processes. The development of this ‘intuitive biology’ also results in an understanding
of the animal kingdom based on biological inter-relatedness rather than similarity to
humans.
The issue with respect to the current study is not whether there is an increased use of
taxonomic relations with age, or a significant change in children’s theoretical
30
understanding of biological categories. There is, indeed, persuasive evidence in favor of
both of these proposals. Instead we are concerned here with what these changes might
imply for the organization of conceptual knowledge structures.
With regard to the category of animals, the finding of clustering by environmental
context is consistent with a primary basis for category organization based on schematic
relations. Thus, animals appeared to be encoded in terms of the places in which they are
experienced (i.e. in the home, on a farm, in a zoo, etc.), and are most closely related to
other animals experienced in similar locations. Significantly, this same basic structure
seems to underlie the animal knowledge of all age groups. Therefore, although older
children may have more sophisticated knowledge of animals, that may include
increased understanding of taxonomic relations, there is little to suggest a radical
upheaval with age in the way that this category knowledge is organized. The findings of
increased proportions of non-mammals and invertebrates in the animal freelistings of
older children are consistent with suggestions of the expansion (Nelson, 1974; Anglin,
1977) or redrawing (Carey, 1985) of the boundaries of this category with age. Changes
in category boundaries might themselves be expected to bring about re-organization of
internal category structures. The current study suggests, however, that redrawing the
boundaries of the animal category has a more or less incremental effect on its internal
organization; that is, such changes are accommodated through the addition of new
environmental contexts, or the inclusion of new category members in existing context
groupings.
For the body parts category, the current study suggests a primary organization of
external parts based on the differentiation of face/head parts from trunk/limb parts.
Since a schema for the face is thought to develop in infancy (Gibson & Spelke, 1983),
31
this finding is also consistent with the proposal that schematic relations form the main
substrate of conceptual representations of young children. There was also some
evidence that taxonomic (functional) relations may play a role in the sub-cluster
organization of external body parts especially in the older age group.
Younger children in our study generated only a small number of internal body parts,
and there was insufficient information in the fluency data to determine whether these
are organized in any structured way. In contrast the oldest children in our sample (9 and
10 year olds) generated a substantial cluster of internal body showing some evidence of
internal structure according to function. This data would appear to consistent with
Carey’s assertion that between the ages of eight and ten children development new
representations of the insides of the human body based on a functional understanding of
bodily processes.
Continuity and Change in Category Structure
How can we make sense of the apparent persistence of early forms of memory
organization in the face of significant theory change and increased familiarity and use
of taxonomic relations with age? Research on connectionist models of human memory
(Rumelhart & Todd, 1992; Elman, 1993; McClelland, 1994; Hartley, Prescott &
Nicolson, 1998), since it allows the investigation of learning trajectories, suggests one
possibility. The landscape of possible network configurations for a neural network
usually contains many local optima—configurations that are better than other nearby
solutions but do not provide a global optimum (the solution that provides the best
possible overall ‘fit’ to the training data). For this reason the early training phase for a
network, which is critical in establishing an initial organization, may set limits for later
32
learning (Elman, 1993). Further training, that involves additional or more complex
training patterns, will generally promote convergence to a nearby, locally optimal
configuration (relative to the previously established pattern of organization), that will
accommodate both the old and new data. However, this configuration will generally be
quite different from that of a network trained from scratch with the full and final
training set. In other words, an early phase of training on a sub-set of data, that
emphasizes some properties and not others, is likely to bias the eventual outcome of the
learning process.
Viewing the development of conceptual memory from this perspective would suggest a
parallel between the early phase of network learning and the establishment of memory
structures based on schematic relations. Later learning about taxonomic/functional
relations, or changes in the theoretical understanding of concepts, could add
considerable refinement and complexity to the pattern of organization without
necessarily effecting a major upheaval to its basic structure. A further implication is
that if, as Mandler (1983) has suggested, schematic relations "remain the predominant
form of organization throughout life" (p. 473), then adult conceptual memory may not
be optimally organized with respect to the representation of taxonomic relations. That
is, the super-imposition of taxonomic data on a substrate defined by schematic
relationships could lead to a degree of compromise in how effectively the taxonomic
relations are encoded.
Conclusion
In this article we have examined the development of semantic fluency between the ages
of 5 and 10, both for the frequently studied category of animals, and for the relatively
33
unexplored category of human body parts. The use of a hierarchical clustering
technique (additive tree analysis) revealed interesting patterns of both continuity and
change in children’s conceptual structures. For the animals category we have
confirmed a tendency to cluster items by environmental context at all ages (resolving
some uncertainty particularly with regard to the behavior of the youngest age group
investigated). Thus, although there is a substantial increase in children’s knowledge
over this age range, accompanied by a deepening understanding of biological kinds, this
pattern of memory acquisition could be described, following Eimas (1994, p. 85), as “a
quantitative enrichment, and not a qualitative transformation of {...} early category
representation.” For the body parts category the analysis provided here suggests a
mixture of continuity and change that we have provisionally interpreted as showing a
primarily schematic organization modified by a growing functional understanding of
the body (and, in particular, its internal parts). Uniquely, for a study of hierarchical
clustering in the development of semantic fluency, these comparisons have been made
both within and across age groups and have used quantitative measures of addtree
similarity. Finally, we have suggested that the tendency of connectionist models to
retain their early organizational structure whilst adapting to represent more complex or
detailed information, could provide plausible models for understanding continuity and
change in the development of category structure.
34
Acknowledgements
The authors are grateful to the teachers and children of the Sacred Heart and
Hillsborough primary schools in Sheffield, UK, for their assistance in this study, and to
Rod Nicolson of the University of Sheffield, for his advice and guidance.
i The probability distribution for
i
al
i
jl is a discrete triangular distribution with
mean 0 and range
(
n
l
1
)
to
(
n
l
1
)
. This distribution can be approximate by the
equivalent continuous triangular distribution (range –n to +n) which has a standard
deviation of 6n, and therefore increases linearly with n.
iiValues of
(
n
al
n
bl
)
were calculated by (i) generating the relevant probability
distributions for different values of
n
al
,
and
n
bl
, (ii) calculating the standard deviations
over a broad range of values for
n
l
, and (iii) using linear regression to find a suitable
coefficient to estimate the standard deviation as a linear function of
n
l
. Note, that a
linear approximation was found to produce a good fit to this data for
n
al
n
bl
6
.
iii The set of possible values for
F
ab forms a standard hypergeometric distribution, for
which the expected value is given by the equation shown.
iv In theory the standard deviation of the hypergeometric distribution could be used to
scale values of
C
ab , however, this has two drawbacks. First, we would still need to
ensure that the values of the scaled
C
ab fell within a bounded range. Second, as the
number of participants naming either item a or b approaches N (the size of the group),
the standard deviation of the distribution approaches 0, making it an unsuitable
scaling factor for items with very high production frequencies.
v Additional analyses were also performed using and
1
as separate metrics, the
results of these analyses were broadly consistent with the trends described here for the
combined
0
.
5
measure.
35
vi Uyeda and Mandler (1980) used a rating scale of 1 (high typicality) to 7 (low
typicality), hence the correlations with the production frequencies in the current study
are all negative.
vii Note, the non-parametric Kruskal-Wallis test is used since many children in all age
groups named no non-mammals and consequently the distribution of ratios is highly
skewed. Values for H are after correction for ties.
viii Internal organs were selected for this analysis because of the difficulty of defining
internal body parts. For instance, it is not clear whether parts of the mouth (teeth,
tongue, etc.) should count as internal or external. Gellert (1962) noted a similar
problem in her analysis of children’s freelisting of things ‘inside’ the body.
ix The division of a tree into k clusters is dependent on the choice of the root node of
the tree. The root node is usually selected to produce a balanced tree, and, in the case
of all the trees used in the current Bk analyses, one which minimizes the variance of
the distances from the root to the external nodes of the tree. However, other choices
for the root node are possible. Clearly, the choice of root node can effect the
computation of statistic the Bk statistic. However, experiments with alternative
choices of root node (that also produced reasonably balanced trees) showed only a
small effect on Bk for the comparisons reported here.
x Fowlkes and Mallows (1983) state that these limits provide only an approximate
indication of the significance of Bk (i.e. it is not possible to specify a p value) since
the distribution of the measure is not normal and successive values are generally
correlated.
36
Figure Legends
Figure 1. Addtree analyses of inter-item similarity data for animal terms using items,
for each year group, that were listed by at least 20% of the children in that group.
Figure 2. Top: Addtree analyses of inter-item similarity data for animal terms using
items, for all year groups, that were listed by at least 20% of the children in the year 1
group and co-occurred in all three year groups. Internal nodes are labeled according to
the number of clusters (k) formed when the tree is split below that node. Bottom:
Graphs of Bk (x-axis) vs. k (y-axis) for the comparisons, from left to right, year 1 with
year 3, year 1 with year 5, and year 3 with year 5. Values of Bk are indicated by diamond
symbols, while the plain solid line and the two dotted lines show the expected value
and its upper and lower limits for each value of Bk. Values of Bk that exceed the upper
limit can be considered significant.
Figure 3. Addtree analyses of inter-item similarity data for body parts using items, for
each year group, that were listed by at least 20% of the children in that group.
Figure 4. Top: Addtree analyses of inter-item similarity data for body parts using
items, for all year groups, that were listed by at least 20% of the children in the year 1
group and co-occurred in all three year groups. Bottom: Graphs of Bk for the
comparisons, from left to right, year 1 with year 3, year 1 with year 5, and year 3 with
year 5 (see Figure 2 legend for key).
37
38
References
Aldenderfer, M. S., & Blashfield, R. K. (1984). Cluster Analysis. London: Sage
Publications.
Anglin, J. M. (1977). Word, object, and conceptual development. New York:
Norton & Co.
Bjorklund, D. F. (1985). The role of conceptual knowledge in the development
of organization in children's memory. In C. J. Brainerd (Ed.), Basic Processes in
Memory Development. New York: Springer-Verlag.
Bjorklund, D. F., Thompson, B. E., & Ornstein, P. A. (1983). Developmental-
trends in children's typicality judgments. Behavior Research Methods &
Instrumentation, 15(3), 350-356.
Bousfield, W. A., & Sedgewick, C. H. W. (1944). An analysis of sequences of
restricted associative responses. Journal of General Psychology, 30, 149-165.
Carey, S. (1985). Conceptual change in childhood. Cambridge, MA: MIT Press.
Chan, A. S., Butters, N., Paulsen, J. S., Salmon, D. P., Swenson, M. R., &
Maloney, L. T. (1993). An assessment of the semantic network in patients with
Alzheimer’s- disease. Journal of Cognitive Neuroscience, 5(2), 254-261.
Clark, E. V. (1995). The lexicon in acquisition. Cambridge: Cambridge
University Press.
Corter, J. E. (1982). Addtree/P - a Pascal program for fitting additive trees based
on Sattath and Tversky Addtree algorithm. Behavior Research Methods &
Instrumentation, 14(3), 353-354.
39
Eimas, P. D. (1994). Categorization in early infancy and the continuity of
development. Cognition, 50, 83–93.
Elman, J. L. (1993). Learning and development in neural networks - the
importance of starting small. Cognition, 48(1), 71-99.
Fowlkes, E. B., & Mallows, C. L. (1983). A method for comparing two
hierarchical clusterings. Journal of the American Statistical Association, 78, 553-569.
Gellert, E. (1962). Children's conceptions of the context and functions of the
human body. Genetic Psychology Monographs, 65, 293-405.
Gibson, J. J., & Spelke, E. S. (1983). Infant Perception. In J. H. Flavell & E. M.
Markman (Eds.), Cognitive Development(Vol. 3, ). New York: Wiley.
Grube, D., & Hasselhorn, M. (1996). Children's freelisting of animal terms:
Developmental changes in activating categorical knowledge. Zeitschrift Fur
Psychologie, 204(2), 119-134.
Gruenewald, P. J., & Lockhead, G. R. (1980). The Free Recall of Category
Examples. Journal of Experimental Psychology: Human Learning and Memory, 6(3),
225-241.
Hartley, S. J. (1999). Development of conceptual knowledge: Connectionist and
experimental insights. PhD Thesis, University of Sheffield.
Hartley, S. J., Prescott, T. J., & Nicolson, R. (1998, ). Experimental and
connectionist perspectives on semantic memory development. Paper presented at the
Proceedings of Twentieth Annual Conference of the Cognitive Science Society,
Wisconsin, USA.
40
Henley, N. M. (1969). A psychological study of the semantics of animal terms.
Journal of Verbal Learning and Verbal Behaviour, 8, 176-184.
Jaakkola, R. O. & Slaughter, V. (2002). Children's body knowledge:
Understanding 'life' as a biological goal. British Journal of Developmental Psychology,
20, 325–342.
Kail, R., & Nippold, M. A. (1984). Unconstrained retrieval from semantic
memory. Child Development, 55(3), 944-951.
Kail, R. V. (1990). The development of memory in children. New York:
Freeman.
Keil, F. C. (1983). On the emergence of semantic and conceptual distinctions.
Journal of Experimental Psychology-General, 112(3), 357-385.
Lin, E.L. & Murphy, G.L. (2001). Thematic Relations in Adult Concepts.
Journal of Experimental Psychology: General, 130(1), 2-28.
Lucariello, J., Kyratzis, A., & Nelson, K. (1992). Taxonomic knowledge - what
kind and when. Child Development, 63(4), 978-998.
Mandler, J. M. (1983). Representation. In J. H. Flavell & E. M. Markman
(Eds.), Cognitive Development(Vol. 3, ). New York: Wiley.
Maridaki Kassotaki, K. (1997). Are rating-based procedures reliable for
derivation of typicality judgments from children? Behavior Research Methods
Instruments & Computers, 29(3), 376-385.
McClelland, J. L. (1994). The organization of memory: A parallel distributed
processing perspective. Revue Neurologique, 150(8-9), 570-579.
41
Morrison, C. M., Chappell, T. D., & Ellis, A. W. (1997). Age of acquisition
norms for a large set of object names and their relation to adult estimates and other
variables. Quarterly Journal of Experimental Psychology Section a-Human
Experimental Psychology, 50(3), 528-559.
Neely, J. H. (Ed.). (1991). Semantic priming effects in visual word recognition:
A selective review of current findings and theories. Hillsdale, N.J: Lawrence Erlbaum
Associates.
Nelson, K. (1974). Variations in children's concepts by age and category. Child
Development, 45, 577-584.
Nelson, K. (1983). The derivation of concepts and categories from event
representation. In E. K. Scholnick (Ed.), New trends in conceptual representation:
Challenges to Piaget's theory? Hillsdale, NJ: Lawrence Erlbaum Associates.
Nelson, K., Fivush, R., Hudson, J., & Lucariello, J. (1983). Scripts and the
development of memory. In M. T. H. Chi (Ed.), Contributions to human development.
Vol. 9. Trends in memory development research. Basel: S. Karger.
Nelson, K., & Gruendel, J. M. (1981). Generalized event representations: Basic
building blocks of cognitive development. In A. Brown & M. Lamb (Eds.), Advances in
Developmental Psychology(Vol. 1, ). Hillsdale, NJ: Lawrence Erlbaum Associates.
Paris, S. G., Lindauer, B. K., & Cox, G. L. (1977). The development of
inferential comprehension. Child Development, 48, 1728-1733.
Posnansky, C. J. (1978). Category norms for verbal items in 25 categories for
children in grades 2-6. Behavior Research Methods and Instrumentation, 10, 819-832.
Rumelhart, D. E., & Todd, P. M. (Eds.). (1992). Learning and connectionist
representations. Cambridge: MA:MIT Press.
42
Sattath, S., & Tversky, A. (1977). Additive similarity trees. Psychometrika,
42(3), 319-345.
Storm, C. (1980). The semantic structure of animal terms - a developmental-
study. International Journal of Behavioral Development, 3(4), 381-407.
Uyeda, K. M., & Mandler, G. (1980). Prototypicality norms for 28 semantic
categories. Behavior Research Methods and Instrumentation, 12, 587-595.
43
Table 1
Mean number of responses in category fluency tasks (standard deviations in brackets).
One-way ANOVA shows a significant increase in mean number of responses with
age, F-values (all significant at p<0.01) are given in column 5.
Category Year 1 Year 3 Year 5 F(2, 152)
Animals 8.86 (3.02) 12.29 (4.55) 15.76 (3.99) 38.61
Body Parts 9.88 (4.20) 13.02 (5.1) 17.12 (4.53) 30.45
Clothes 8.40 (2.76) 10.99 (3.93) 12.28 (4.36) 14.94
Foods 8.08 (2.90) 11.87 (4.93) 14.28 (4.08) 29.24
Plants 3.74 (2.69) 5.51 (2.48) 7.40 (3.73) 18.62
Vehicles 5.28 (2.20) 8.26 (3.04) 9.66 (3.31) 29.97
44
Table 2.
Relative frequency of items in the ‘animals’ category named by at least 20% of the
participants in each age group.
animals Year 1 animals Year 3 animals Year 5
lion 0.48 dog 0.69 dog 0.88
tiger 0.48 cat 0.67 cat 0.86
cat 0.46 elephant 0.49 fish 0.68
dog 0.46 monkey 0.47 tiger 0.66
horse 0.44 horse 0.45 lion 0.58
pig 0.38 cow 0.42 monkey 0.52
monkey 0.34 rabbit 0.36 mouse 0.50
cow 0.32 giraffe 0.35 rabbit 0.48
elephant 0.32 lion 0.35 hamster 0.46
giraffe 0.28 pig 0.35 horse 0.46
sheep 0.28 tiger 0.33 pig 0.44
rabbit 0.22 fish 0.31 cow 0.42
bird 0.20 kangaroo 0.24 zebra 0.40
mouse 0.24 bird 0.34
sheep 0.24 butterfly 0.32
dolphin 0.22 rat 0.32
whale 0.22 sheep 0.32
bird 0.20 fox 0.26
hamster 0.20 elephant 0.22
shark 0.20 giraffe 0.20
guinea pig 0.20
snake 0.20
45
Table 3.
Relative frequency of items in the ‘body parts’ category named by at least 20% of
participants in each age group.
Body Part Year 1 Body Part Year 3 Body Part Year 5
leg 0.78 leg 0.80 foot 0.80
head 0.62 foot 0.76 nose 0.72
eye 0.56 arm 0.65 arm 0.68
nose 0.54 eye 0.62 brain 0.68
foot 0.52 hand 0.60 eye 0.68
arm 0.50 nose 0.56 heart 0.68
mouth 0.46 mouth 0.53 leg 0.66
tummy 0.42 head 0.51 hand 0.56
hand 0.38 ear 0.44 mouth 0.56
ear 0.34 finger 0.42 finger 0.54
hair 0.32 toe 0.35 lung 0.54
neck 0.32 heart 0.31 toe 0.50
bone 0.26 brain 0.29 ear 0.48
finger 0.26 elbow 0.29 elbow 0.44
toe 0.22 hair 0.29 liver 0.40
heart 0.20 knee 0.27 head 0.38
neck 0.27 hair 0.34
shoulder 0.27 knee 0.34
tummy 0.22 rib 0.34
bone 0.28
neck 0.28
shoulder 0.28
kidney 0.24
stomach 0.24
ankle 0.22
skin 0.22
back 0.20
muscle 0.20
thigh 0.20
46
Year 3
cat
dog
rabbit
bird
hamster
mouse
horse
cow
sheep
pig
fish
dolphin
shark
whale
elephant
giraffe
lion
tiger
kangaroo
monkey
Year 1
tiger
lion
monkey
bird
giraffe
elephant
pig
sheep
cow
horse
dog
cat
Figure 1 (part 1)
47
cat
dog
butterfly
bird
giraffe
elephant
monkey
lion
tiger
zebra
mouse
rat
fish
fox
rabbit
hamster
snake
horse
cow
pig
sheep
Year 5
Figure 1 (part 2)
48
Year 5
sheep
pig
cow
tiger
horse
lion
monkey
dog
cat
giraffe
bird
elephant
2
34
5
8
6
7
9
10
11
bird
cat
dog
elephant
giraffe
lion
tiger
cow
sheep
horse
monkey
pig
2
3
6
4
5
7
8
9
10
11
Year 3
Year 1
tiger
lion
monkey
bird
giraffe
elephant
pig
sheep
cow
horse
dog
cat
2
3
45
6
7
8
9
10
11
Figure 2 (part 1)
49
0
0.25
0.5
0.75
1
2
4
6
8
10
0
0.25
0.5
0.75
1
2
4
6
8
10
0
0.25
0.5
0.75
1
2
4
6
8
10
B(k): Year 1 / Year 3 B(k): Year 1 / Year 5 B(k): Year 3 / Year 5
Figure 2 (part 2)
50
Year 3
brain
heart
arm
leg
hand
head
ear
hair
eye
mouth
nose
foot
finger
toe
elbow
shoulder
knee
neck
tummy
Year 1
bone
heart
arm
leg
finger
hand
tummy
eye
mouth
head
neck
hair
foot
toe
ear
nose
Figure 3 (part 1)
51
Year 5
brain
bone
muscle
heart
lung
kidney
liver
rib
foot
hand
finger
toe
ear
eye
mouth
nose
elbow
knee
arm
leg
shoulder
hair
stomach
head
neck
Figure 3 (part 2)
52
Year 3
bone
heart
neck
finger
toe
ear
arm
leg
hair
eye
mouth
hand
foot
head
nose
2
3
4
5
6
7
8
9
10
11
12
13 14
Year 1
bone
heart
nose
head
neck
foot
toe
eye
mouth
hand
arm
leg
finger
hair
ear
2
3
4
5
67
8
9
10
11
12
13
14
Figure 4 (part 1)
53
Year 5
bone
heart
arm
leg
foot
hand
ear
eye
mouth
nose
finger
toe
head
hair
neck
2
3
4
5
6
7
8
9
10
11
12
13
14
0
0.25
0.5
0.75
1
2
4
6
8
10
12
14
0
0.25
0.5
0.75
1
2
4
6
8
10
12
14
0
0.25
0.5
0.75
1
2
4
6
8
10
12
14
B(k): Year 1 / Year 3 B(k): Year 1 / Year 5 B(k): Year 3 / Year 5
Figure 4 (part 2)
Crowe, S. J. and T. J. Prescott (2003). Continuity and Change in the Development
of Category Structure: Insights from the Semantic Fluency Task. International
Journal of Behavioral Development, 27, 467-479..
Erratum: On p. 470. In the equation for
!
"
(a,b)
the parameter
!
"
(nal nbl )
should be a
divisor not a multiplier of
!
dabl
, hence the equation should read
!
"
(a,a)=0,
"
(a,b)=1
Fab
dabl
#
(nal nbl )
l(a$l%b$l)
&
'
(
)
)
*
+
,
,
.
... To our knowledge, this is the first time this has been done in this way (cf. Crowe and Prescott 2003). We believe this opens interesting possibilities for future cross-linguistic work on body part categorisation as it does not require asking people to make explicit linguistic judgements or presuming a specific type of relation between words. ...
... Subsystems also correspond to the end points of cross-linguistic tendencies in semantic shifts (Wilkins 1981(Wilkins , 1996. Finally, there are some correspondences between the subgroups found here and those found in free-listing data from English speaking children (e.g., a cluster of digits-see Crowe and Prescott 2003). Future work on a broader sample of language families will have to confirm whether the observations reported here are specific to Japonic or whether they have wider applicability. ...
Article
Full-text available
The human body is central to myriad metaphors, so studying the conceptualisation of the body itself is critical if we are to understand its broader use. One essential but understudied issue is whether languages differ in which body parts they single out for naming. This paper takes a multi-method approach to investigate body part nomenclature within a single language family. Using both a naming task (Study 1) and colouring-in task (Study 2) to collect data from six Japonic languages, we found that lexical similarity for body part terminology was notably differentiated within Japonic, and similar variation was evident in semantics too. Novel application of cluster analysis on naming data revealed a relatively flat hierarchical structure for parts of the face, whereas parts of the body were organised with deeper hierarchical structure. The colouring data revealed that bounded parts show more stability across languages than unbounded parts. Overall, the data reveal there is not a single universal conceptualisation of the body as is often assumed, and that in-depth, multi-method explorations of under-studied languages are urgently required.
... Word prototypicality in general has been related to how quickly a word may be retrieved from the mental lexicon, which is known as the typicality effect (Jerger & Damian, 2005). Indeed, more typical exemplars show an advantage in processing with respect to less typical exemplars in semantic tasks such as category verification (Jerger & Damian, 2005), semantic fluency (Crowe & Prescott, 2003), animacy decision (Räling et al., 2017) and category naming (Hampton, 1995). The typicality effect has been observed also in tasks involving both lexical and semantic processes like reading (Garrod & Sanford, 1977), sentence production (Kelly et al., 1986) and picture naming (Holmes & Ellis, 2006. ...
Article
Full-text available
Emotional words differ in how they acquire their emotional charge. There is a relevant distinction between emotion-label words (those that directly name an emotion, e.g., “joy” or “sadness”) and emotion-laden words (those that do not name an emotion, but can provoke it, e.g., “party” or “death”). In this work, we focused on emotion-label words. These words vary in their emotional prototypicality, which indicates the extent to which the word refers to an emotion. We conducted two lexical decision experiments to examine the role played by emotional prototypicality in the recognition of emotion-label words. The results showed that emotional prototypicality has a facilitative effect in word recognition. Emotional prototypicality would ease conceptual access, thus facilitating the retrieval of emotional content during word recognition. In addition to the theoretical implications, the evidence gathered in this study also highlights the need to consider emotional prototypicality in the selection of emotion-label words in future studies.
... A category production task simply requires participants to name concepts that belong to a given category, such as ANIMALS 1 or EMOTIONS. It is used in a wide range of research, but particularly to investigate underlying categorical and conceptual structure (e.g., Crowe & Prescott, 2003;Hampton & Gardiner, 1983;Rosch, 1975;Troyer, 2000), semantic memory (e.g., Binney et al., 2018;Ryan et al., 2008), and executive function (e.g., Baldo & Shimamura, 1998;Fisk & Sharp, 2004;Shao et al., 2014). The task is also an important tool in clinical research (e.g., Bokat & Goldberg, 2003;Henry & Crawford, 2004) and diagnosis (e.g., Quaranta et al., 2016;Zhao et al., 2013). ...
Article
Full-text available
We present a database of category production (aka semantic fluency) norms collected in the UK for 117 categories (67 concrete and 50 abstract). Participants verbally named as many category members as possible within 60 seconds, resulting in a large variety of over 2000 generated member concepts. The norms feature common measures of category production (production frequency, mean ordinal rank, first-rank frequency), as well as response times for all first-named category members, and typicality ratings collected from a separate participant sample. We provide two versions of the dataset: a referential version that groups together responses that relate to the same referent (e.g., hippo, hippopotamus) and a full version that retains all original responses to enable future lexical analysis. Correlational analyses with previous norms from the USA and UK demonstrate both consistencies and differences in English-language norms over time and between geographical regions. Further exploration of the norms reveals a number of structural and psycholinguistic differences between abstract and concrete categories. The data and analyses will be of use in the fields of cognitive psychology, neuropsychology, psycholinguistics, and cognitive modelling, and to any researchers interested in semantic category structure. All data, including original participant recordings, are available at https://osf.io/jgcu6/.
... In free-listing tasks, participants are presented with target words (e.g., "kinship"), and are asked to list as many exemplars related to the word as they can in a given timeframe. Free-listing tasks have often been deployed to gain insights in the organization of conceptual knowledge by cognitive and developmental psychologists too, that by means of free-listing described the representation of several conceptual categories, such as animals (Crowe and Prescott 2003), landscape (van Putten et al. 2020), food (Hough and Ferraris 2010), and-more importantly for our purposes-gender (Mazzuca et al. 2020a). Like other semantic fluency tasks, free-listing tasks are thought to provide an indirect measure of psychological proximity of concepts: the basic assumption underlying these kinds of tasks is that concepts that are mentioned earlier and more frequently are more psychologically salient for the target concept. ...
Article
Full-text available
Biases in cognition are ubiquitous. Social psychologists suggested biases and stereotypes serve a multifarious set of cognitive goals, while at the same time stressing their potential harmfulness. Recently, biases and stereotypes became the purview of heated debates in the machine learning community too. Researchers and developers are becoming increasingly aware of the fact that some biases, like gender and race biases, are entrenched in the algorithms some AI applications rely upon. Here, taking into account several existing approaches that address the problem of implicit biases and stereotypes, we propose that a strategy to cope with this phenomenon is to unmask those found in AI systems by understanding their cognitive dimension, rather than simply trying to correct algorithms. To this extent, we present a discussion bridging together findings from cognitive science and insights from machine learning that can be integrated in a state-of-the-art semantic network. Remarkably, this resource can be of assistance to scholars (e.g., cognitive and computer scientists) while at the same time contributing to refine AI regulations affecting social life. We show how only through a thorough understanding of the cognitive processes leading to biases, and through an interdisciplinary effort, we can make the best of AI technology.
... During childhood, as lexical and phonological processing skills develop (Gândara & Befi-Lopes, 2010), linguistic correlates of performance on VF tasks might be different from those that are observed in adults for whom these linguistic skills are already more established. Additionally, the many developmental changes that semantic memory networks undergo during childhood (Crowe & Prescott, 2003) imply variations in the cognitive functions that are recruited, particularly for SVF performance (Acevedo et al., 2000). ...
Article
Despite the widespread use of verbal fluency (VF) tasks in child neuropsychological research and clinical practice, the contribution of executive and linguistic processes to variability in children's fluency performance is still unclear. This is particularly important when considering the development of orthographic knowledge and semantic network during childhood. The present study investigated the contributions of executive functions and linguistic skills to performance in VF tasks in children. We examined the contributions of basic executive functions (i.e.,inhibitory control, working memory, and flexibility) and high-order executive functions (i.e., planning), vocabulary, lexical access speed, and phonological awareness to VF performance in 111 typically developing children (8-10 years old). Multiple regression analyses showed that phonological awareness was a predictor of performance in phonemic verbal fluency (PVF), and lexical access speed was the best predictor of performance in semantic verbal fluency (SVF). Among the executive function components, working memory was a predictor of performance in PVF and most categories of SVF (except animal fluency). In addition to working memory resources (i.e., a basic executive function), planning (i.e., a high-level executive function) was also recruited in the clothing category of SVF. These results highlight the importance of phonological processing skills in children's performance on VF tasks and show similarities and differences in the contributions of various linguistic and executive skills to PVF and SVF. These findings have implications for interpreting the results of these measures in research and clinical practice.
... We wanted to capture the whole pattern of elicited conceptual relations, avoiding constraining participants to produce only properties that are true of the concept. Importantly, free-listing tasks have also been used to understand how people represent concepts 35,36 . In free-listing tasks, concepts that are mentioned earlier and more frequently in a given list are thought to be more psychologically salient for the target concept. ...
Article
Full-text available
Several studies have highlighted the flexible character of our conceptual system. However, less is known about the construction of meaning and the impact of novel concepts on the structuring of our conceptual space. We addressed these questions by collecting free listing data from Italian participants on a newly–and yet nowadays critical–introduced concept, i.e., COVID-19, during the first Italian lockdown. We also collected data for other five illness-related concepts. Our results show that COVID-19’s representation is mostly couched in the emotional sphere, predominantly evoking fear—linked to both possible health-related concerns and social-emotional ones. In contrast with initial public debates we found that participants did not assimilate COVID-19 neither completely to severe illnesses (e.g., tumor) nor completely to mild illnesses (e.g., flu). Moreover, we also found that COVID-19 has shaped conceptual relations of other concepts in the illness domain, making certain features and associations more salient (e.g., flu-fear; disease-mask). Overall, our results show for the first time how a novel, real concept molds existing conceptual relations, testifying the malleability of our conceptual system.
... Over the course of development, semantic organization both expands to incorporate new concepts and new relations between concepts (Bjorklund & Jacobs, 1985;Blaye, Bernard-Peyron, Paour, & Bonthoux, 2006;Coley, 2012;Crowe & Prescott, 2003;Howard & Howard, 1977;Nguyen, 2007;Storm, 1980;Tversky, 1985;Unger, Fisher, Nugent, Ventura, & MacLellan, 2016;Vales, Stevens, & Fisher, 2020;Walsh, Richardson, & Faulkner, 1993). The trajectory of semantic organization development has been extensively studied in prior developmental research. ...
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
... However, this demographic bias may not have affected the comparisons of semantic memory structures, because the knowledge about animals is acquired in the early stage of the development (55). Furthermore, the primitive structures, e.g., clustering, are already present in early childhood (56)(57)(58); basic semantic structures should be relatively invariant across ages and educational backgrounds. ...
Article
Full-text available
Background: Beneficial effects of transcranial direct current stimulation (tDCS) are relevant to cognition and functional capacity, in addition to psychiatric symptoms in patients with schizophrenia. However, whether tDCS would improve higher-order cognition, e.g., semantic memory organization, has remained unclear. Recently, text-mining analyses have been shown to reveal semantic memory. The purpose of the current study was to determine whether tDCS would improve semantic memory, as evaluated by text-mining analyses of category fluency data, in patients with schizophrenia. Methods: Twenty-eight patients entered the study. Cognitive assessment including the category fluency task was conducted at baseline (before tDCS treatment) and 1 month after t administration of tDCS (2 mA × 20 min, twice per day) for 5 days, according to our previous study. The category fluency data were also obtained from 335 healthy control subjects. The verbal outputs (i.e., animal names) from the category fluency task were submitted to singular valued decomposition (SVD) analysis. Semantic memory structures were estimated by calculating inter-item cosines (i.e., similarities) among animal names frequently produced in the category fluency task. Data were analyzed longitudinally and cross-sectionally to compare the semantic structure within the patient group (i.e., baseline vs. follow-up) and between groups (patients vs. healthy controls). In the former, semantic associations for frequent items were compared in the form of cosine profiles, while in the latter, the difference in the magnitude of the correlations for inter-item cosines between healthy controls and patients (baseline, follow-up) was examined. Results: Cosine profiles in the patient group became more cluster-based (i.e., pet, carnivores, and herbivores) at follow-up compared to those at baseline, yielding higher cosines within subcategories. The correlational coefficient of inter-item cosines between healthy controls and patients was significantly greater at follow-up compared to baseline; semantic associations in patients approached the normality status after multi-session tDCS. Conclusions: To our knowledge, this is the first study to demonstrate the facilitative effect of tDCS on semantic memory organization in patients with schizophrenia. Text-mining analysis was indicated to effectively evaluate semantic memory structures in patients with psychiatric disorders.
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.
Conference Paper
Full-text available
We describe an experimental investigation of the development of children’s knowledge structures which aims to provide data for connectionist modelling. 167 children between 5 and 11 years of age completed two category fluency tasks where they were asked to produce as many names of a) animals and b) parts of the body, as they could in one minute. Similarity scores were derived based on distances between concepts in the lists produced. These were analysed using the ADDtree algorithm (Sattath & Tversky, 1977) to build structures representing the organisation of the children’s knowledge of animals and body parts. The results showed that animal knowledge was generally organised in terms of environmental context/habitat, however, there was evidence for subtle changes in knowledge organisation between age groups. More pronounced changes were observed in the organisation of knowledge of body parts which gave some... support to the assertion that children progress from making coarser to making finer distinctions between concepts (see Keil, 1979) and reflected the progression observed in knowledge structure development
Article
The development of organization in children’s memory has received substantial research attention over the past 10 years (for reviews see Lange, 1978; Mandler, 1979; Moely, 1977; Ornstein & Corsale, 1979). Organization is customarily defined as the structure discovered or imposed upon a set of items by a learner, with this structure facilitating retrieval of items from memory (see Chapter 1 by Ackerman for a discussion of factors affecting retrieval from children’s memory). Organization has generally been thought of as a strategic process (i.e., deliberate and effortful), with age changes in the use of organization being attributed to age differences in strategic functioning. Such a conclusion is supported by the results of numerous studies demonstrating that young children, who show little evidence of spontaneous organization, can be trained to cluster their recall according to adult categorizations (e.g., Bjorklund, Ornstein, & Haig, 1977; Moely & Jeffrey, 1974; Moely, Olson, Halwes, & Flavell, 1969). In this chapter, I dispute this position, arguing instead that most of the age changes in the organization of children’s recall are not strategic, but rather can be attributed to developmental changes in the structure and content of children’s conceptual representations. Organization in memory does become strategic, I believe, sometime during adolescence, resulting in a qualitatively different type of memory functioning. However, I argue that the regular improvements observed in memory organization over the course of the preschool and elementary school years can most parsimoniously be attributed to developmental differences in the structure of semantic memory and the ease with which certain types of semantic relationships can be activated.
Article
The present study employed multidimensional scaling and ADDTREE clustering analyses to derive the cognitive maps and clustering representations of normal elderly controls (NC), patients with Alzheimer's disease (AD), and patients with Hun-tington's disease (HD); the analyses were performed on subjects' responses in a category fluency task that involved generating animal names for 60 sec. A measure of the proximity of animal names was used as an index of associational strength; MDS and ADDTREE estimates were based on this measure. A comparison of the NC, AD, and HD subjects' cognitive maps suggests that the semantic network of AD patients is abnormal in two ways. First, the organization of the semantic network is disrupted. Second, new abnormal associations and clusterings are formed. These results support the notion that AD is characterized by a breakdown in the structure of semantic knowledge and not primarily by a deficiency in the accessibility of semantic information.
Article
121 second graders and 119 fourth graders were asked to enumerate as many animals as they could within 5 minutes. About half of the second graders carried out this task again six months later. Data analyses revealed the following results: (a) The time course of the increasing number of enumerated animals (i.e., the productivity slope) differs only slightly between the grades. Fourth graders, however, produced more animal terms than second graders, and with regard to the temporal course of children's productivity there was a greater deviation from individually calculated exponential or hyperbolic functions among the older children. The productivity among second graders proves to be very stable over six months, (b) The probability of producing specific animal terms increases with age without substantial changes in the rank order of production frequencies, (c) At both grades comparable tendencies to cluster the animal terms according to typical environmental contexts can be observed. Among fourth graders the amount of clustering is related to productivity, (d) The use of strong associative interitem relations (e.g. lion-tiger) increases with age. The more associative relations are used the more animals are produced within the given time. It is suggested that the results reflect both a quantitative increase of the knowledge base between ages 8 and 10 and qualitative changes in the activation of information from one's own knowledge base.