COGNITIVE PSYCHOLOGY 16, l-27 (1984)

Children’s Sensitivity to Constraints on Word Meaning:

Taxonomic versus Thematic Relations
ELLEN M. MARKMAN AND JEAN E. HUTCHINSON
Stanford University

A major problem in language learning is to figure out the meaning of a word

given the enormous number of possible meanings for any particular word. This
problem is exacerbated for children because they often find thematic relations
between objects to be more salient than the objects’ taxonomic category. Yet
most single nouns refer to object categories and not to thematic relations. How
do children learn words referring to categories when they find thematic relations
so salient? We propose that children limit the possible meanings of nouns to refer
mainly to categorical relations. This hypothesis was tested in four studies. In
each study, preschool children saw a series of target objects (e.g., dog), each
followed by a thematic associate (e.g., bone) and a taxonomic associate (e.g.,
cat). When children were told to choose another object that was similar to the
target (“See this? Find another one.“), they as usual often selected the thematic
associate. In contrast, when the instructions included an unknown word for the
target (“See this fep? Find another fep.“), children now preferred the taxonomic
associate. This finding held up for 2- and 3-year-olds at the basic level of cate-
gorization, for 4- and 5-year-olds at the superordinate level of categorization, and
4- and S-year-olds who were taught new taxonomic and new thematic relations
for unfamiliar objects. In each case, children constrained the meaning of new
nouns to refer mainly to categorical relations. By limiting the hypotheses that
children need to consider, this constraint tremendously simplifies the problem of
language learning.

One of the major problems confronting someone learning a language is

to figure out the meaning of a word given the enormous number of pos-
sible meanings for any particular word. Children commonly learn their

This paper was completed while E. M. Markman was at the Center for Advanced Study
in the Behavioral Sciences, which received support from NSF Grant BNS8206304 and the
Spencer Foundation. This research was supported in part by a Stanford University Fellow-
ship and NIMH Traineeship to J. E. Hutchinson. Portions of the research were presented
at meetings of the Society for Research in Child Development, Detroit, 1983, and the
Western Psychological Association, San Francisco, 1983. We thank the directors and staffs
of the following schools for their participation in these studies: Afternoon Children’s Center,
Bing Nursery School, California Young World (Sunnyvale), College Terrace Preschool,
Community Preschool, Covenant Presbyterian, Darby’s Day Care, Ellen Thacher Nursery
School, Fleumoix Nursery School, Menlo Park Presbyterian, Parents’ Nursery School, and
Stanford Lutheran Toddler’s Center. We are grateful to Diane Edison and Amy Fox for
help in running the studies and to Susan Gelman and Gregory Murphy for advice on the
stimuli for Experiment 4. Requests for reprints should be sent to Ellen M. Markman,
Department of Psychology, Bldg. 420, Stanford University, Stanford, CA 94305.

OOlO-0285/84$7.50
Copyright 0 1984 by Academic Press, Inc.
All rights of reproduction in any form reserved.
2 MARKMAN AND HUTCHINSON

first words (category terms) through ostensive definition: a parent or

other teacher points to an object and labels it. Especially in the early
phases of language acquisition, when children cannot understand a de-
scription of a category, children’s learning of new category terms must
depend heavily on ostensive definition. Once an adult points to an object
and labels it, how does the child settle on an interpretation? At first sight
this would seem to be a simple problem, and in fact children make
hundreds of such inferences correctly when acquiring new vocabulary.
This apparent simplicity, however, belies a complex inferential problem
that was formulated by Quine (1960) in his well-known argument about
translation. Imagine that someone points to a dog and says “chien,” and
our job is to figure out what “chien” means. An obvious hypothesis is
that it means “dog.” But this is not necessary. It could mean “furry
object,” or “brown object,” or “medium-sized object,” and so on. To
decide if the new term refers to dogs, one might set up test situations by
pointing to various objects and asking whether or not “chien” applies.
Quine’s argument is that no matter how many test situations one con-
structs, there will always be more than one hypothesis for the meaning
of a new term that is consistent with the existing evidence.
Young children beginning to acquire their native language continually
face this problem of narrowing down the meaning of a term from an
indefinite number of possibilities. Someone points in some direction and
then utters a word. On what grounds is the child to conclude that a new
unfamiliar word, e.g., “dog”, refers to dogs? What is to prevent a child
from concluding that “dog” is a proper name for that particular dog?
What prevents the child from concluding that “dog” means “four-legged
object” or “black object” or any number of other characteristics that
dogs share? And finally, what prevents the child from concluding that
“dog,” in addition to referring to that particular dog, also refers to the
bone the dog is chewing on or to the tree the dog is lying under? These
last examples of thematic relations pose a particular problem because
children are very interested in such relations and may find them more
salient than categorical relations. Before continuing to discuss how chil-
dren narrow down the possible meanings of terms, we briefly review the
work on classification showing children’s fascination with thematic re-
lations .
One widely used procedure for studying how children form categories
of objects is to ask them to sort objects into groups. Typically children
are presented with objects from several different categories, for example,
vehicles, animals, clothing, and people. They are instructed to put to-
gether the objects that are alike or that go together or are given freedom
to manipulate and group the objects as they like. Another variant of the
sorting procedure is a match to sample task. In this case, children are
 CONSTRAINTS ON WORD MEANING 3

shown a target object and two choices, one in the same category as the
target and one in a different category. Children must choose which is
most like the target. This task in particular is similar to the one that
children face in ostensive definition, in that someone points to an object
and the child must determine which other objects are like it.
Here is a somewhat oversimplified summary of what is often found in
these studies. Children older than about 7 sort objects on the basis of the
object’s taxonomic category. For example, they place all and only the
vehicles together, all and only the clothing together, and so on. They
perceive the perceptual or functional properties that the objects share
(perhaps in a family resemblance structure (Rosch dz Mervis, 1975)) and
so find the common taxonomic category to be a natural way of organizing
objects. Younger children sort on some other basis. Sometimes, especially
when geometric figures are used, young children create spatial configu-
rations with the objects, arranging them into designs or patterns. When
more meaningful objects are used, children represent causal and temporal
relations among the objects as well as spatial relations. These thematic
relations emphasize events rather than taxonomic similarity. For ex-
ample, children might sort a man and a car together because the man is
driving the car. Or they might place a boy, a coat, and a dog together
because the boy will wear his coat when he takes the dog for a walk.
This attention to thematic relations between objects rather than to how
objects are alike is a common finding replicated in many studies. In ad-
dition to sorting experiments, this thematic bias shows up in studies of
memory clustering and word association (Inhelder & Piaget, 1964;
Denney, 1974; Denney & Ziobrowski, 1972; Nelson, 1977). These findings
indicate that children are more interested in the thematic relations among
objects or that thematic relations are simpler or more readily constructed
than categorical relations.
It is not surprising that children notice these thematic relations. They
are obviously very important for making sense of the world for adults
and children alike. As we move about in our daily life, we observe people
interacting or using tools or other artifacts to accomplish goals. We view
natural occurrences such as storms, and we admire scenery. Much of our
perception is interpretive, trying to figure out what is happening and how.
Even infants tend to place causal interpretations on events they perceive
(cf. Gibson & Spelke, 1983). Thus, these event-like structures are a fun-
damentally important and natural way of organizing information. More-
over, there seem to be fewer developmental and cross-cultural differences
in understanding this type of organization (Mandler, Scribner, Cole, &
DeForest, 1980). This is in marked contrast to the cross-cultural and
developmental differences found in studies of taxonomic classification.
In sum, interest in thematic relations is not limited to young children.
4 MARKMAN AND HUTCHINSON

Nor should attention to thematic relations be viewed as a useless or

nonproductive bias. Noticing causal, spatial, and temporal relations be-
tween objects is essential for understanding the world. It is children’s
attention to categorical relations and not their attention to thematic re-
lations that changes most with development.
When the procedures used in sorting tasks specifically guide children’s
attention toward categories and away from thematic relations, children
do show some understanding of categorical organization (see Carey, in
press; Gelman & Baillargeon, 1983; Markman & Callanan, 1983; Horton,
Note 1, for reviews). To take one example, Smiley and Brown (1979)
tested whether 5- and 6-year-old children could understand taxonomic
relations even though they prefer thematic ones. They presented children
with a target picture and two choice pictures. One of the choices was
thematically related to the target, and one of the choices was taxonomi-
tally related to the target. For example, children were shown a spider
(target), a spider web (thematic choice), and a grasshopper (taxonomic
choice). The experimenter pointed to the spider and asked for “the one
that goes best with this one.” As usual, these young children tended to
pick the spider web, rather than the grasshopper, thereby indicating a
thematic relation. Nevertheless, when they were asked about the grass-
hopper, all of the children could explain the taxonomic relation. Thus,
children have a rudimentary ability to organize objects taxonomically,
but it is often obscured by their attention to thematic relations.
Although children are biased toward organizing objects thematically,
single words, in particular count nouns, do not often encode thematic
relations.’ English does not have a single noun for thematically related
objects such as a boy and his bike, a spider and its web, or a baby and
its bottle. Thus to return to Quine’s problem of induction, we are faced
with a kind of paradox. Children seem to readily learn concrete nouns
like “ball” or “dog” that refer to object categories. Yet they tend to
notice and remember thematic relations between objects more readily
than categorical relations. How is it that children readily learn labels for
categories of objects if they are attending to these thematic relations
between objects instead? To take a concrete example, imagine a mother
pointing to a baby and saying “baby.” Based on the sorting studies, we
should assume that the child will be attending to the baby shaking a rattle
or to the baby being diapered. Why, then, doesn’t the child infer that
“baby” also refers to the rattle or to the diaper, in addition to the par-
ticular baby?

I There are a few exceptions, however. Relational information is contained in the meaning
of kinship terms such as “brother” or “cousin.” The term “friends” also refers to people
who have a particular kind of thematic relationship with each other.
 CONSTRAINTS ON WORD MEANING 5

As a possible solution to this problem, we propose that children have

implicit hypotheses about the possible meaning of words that help them
acquire words for categories. Children may well prefer to construe the
environment in a way that conflicts with the way that language is orga-
nized. But even very young children may be aware of the constraint on
word meaning so that when they believe that they are learning a new
word, they shift their attention from thematic to categorical organization.
This proposal has a strong and a weak form. The strong form is that
sensitivity to the constraint on word meaning can help children discover
and learn new categories. That is, on hearing an unfamiliar word, children
will search for categorical relations. If no familiar, previously unlabeled
category is available, children will analyze the environment to form a
new categorical relation to label. In this way, the constraint on possible
word meanings could help children acquire new categories. The weak
form of the proposal is that children use the constraint to help them link
a new word to a concept they already know. If a familiar categorical
relation is not available, however, children will not attempt to search for
a new one.
The four studies that follow test mainly the weak form of the hypoth-
esis. Experiment 1 focuses on 2- and 3-year-old children’s knowledge of
the constraint on word meaning for basic level categorization. Experi-
ments 2 and 3 focus on 4- and 5-year-olds’ sensitivity to the constraint
for superordinate level categories. And Experiment 4 tests the hypothesis
for 4- and 5-year-olds who were taught new taxonomic and new thematic
relations for unfamiliar objects.
All of the experiments use a match to sample procedure with a target
picture, taxonomic choice, and thematic choice. In the No Word condi-
tion, children are shown the target and asked to simply “find another
one.” In the Novel Word condition, children hear a new label for the
target (e.g., “biv”) and are asked to “find another biv.” The Novel Word
task is quite like ostensive definition: someone points to an object and
labels it, and the child must figure out what else the label refers to. We
predicted that the majority of choices in the Novel Word condition would
be taxonomic. The No Word task also has much in common with osten-
sive definition, except that no label is given. We expected children in this
condition to give many more thematic responses than children in the
Novel Word condition.
EXPERIMENT 1
This first study investigates whether hearing a novel word will cause
2- to 3-year-old children to shift their attention from thematic to categor-
ical relations. Basic level categories (such as “dog” or “chair”) were
used with these young children rather than general superordinate level
6 MARKMAN AND HUTCHINSON

categories (such as “animal” or “furniture”). The basic level, according

to Rosch and her colleagues, is the level of categorization at which cat-
egory members have the most features in common without being confus-
able with members of contrasting categories (Rosch, 1978; Rosch,
Mervis, Gray, Johnson, & Boyes-Braem, 1976; Mervis & Rosch, 1981).2
Methods
Subjects
Forty-one children from nursery schools in Palo Alto, California, participated in the study.
They ranged in age from 2 years 5 months to 3 years 11 months, with a mean age of 3 years
4 months. An additional two children failed to pass a pretest described below and were not
included in the study. Children were randomly assigned to one of two conditions with the
constraint that the conditions be roughly equated for age and sex.

Procedure
Pretest. A simple pretest was given to ensure that the children understood instructions
to find an object that is the same as another. A target picture was propped up against a
frame and the child’s attention was drawn to the picture. Then another identical picture
and a distractor were placed on the table in front of the child, and the child was asked to
“find one that is the same as this one” (the target picture). Three sets of pictures were
used: two identical circles with a squiggle as the distractor, two rectangles with a z as the
distractor, and two arrows with a U-shaped figure as the distractor. Children were scored
as passing the pretest only if they answered all three items correctly.
No Word condition. The procedure used in this condition was very similar to the
pretest procedure. Children were first introduced to a hand puppet and were told to put
the picture they chose in the puppet’s mouth. On each trial, the experimenter propped the
target picture against the frame and told the child, “Look carefully now. See this?” as she
pointed to the picture. Then the experimenter placed the two choice pictures on the table
and told the child to “find another one that is the same as this,” as she continued to point
to the target picture. The instructions were designed to make it as clear as possible to these
young children that we were looking for a taxonomic match.
One of the choice pictures was a member of the same basic level category as the target:
for example, the target might be a poodle and the choice a German shepherd (both dogs).

2 Rosch, Mervis, Gray, Johnson, and Boyes-Braem (1976) showed that 3-year-old children
are capable of using category membership to sort objects at the basic level of categorization,
even though they fail to sort objects taxonomically at the superordinate level. In the Rosch
et al. (1976) study, children were presented with two objects related at the basic level, along
with an unrelated distractor, and were asked to find the two that are alike. Three-year-olds
almost always selected the two category members over the unrelated distractor. Because
this study failed to include any competing thematic relations, however, it did not establish
the relative salience of thematic and categorical relations. In a preliminary study, we dem-
onstrated that when a competing thematic relation is present (e.g., a baby and a bottle), 2-
and 3-year-olds often select it over the basic level category (e.g., two babies). When an
unrelated distractor was used, children selected the categorical associate 94% of the time,
as in the Rosch et al. (1976) study. When a thematically related distractor was used, how-
ever, children selected the categorical associate only 56% of the time. This finding allows
us to address the main question about the role of a word in inducing categorical organization.
 CONSTRAINTS ON WORD MEANING 7

We attempted to make the two category exemplars fairly dissimilar yet still readily identi-
fiable to these young children. The other choice picture was a strong thematic associate to
the target-in this case, dog food. There were 10 such triads in all. They are listed in Table
1. The left-right placement of the thematic and category choices was randomly determined
for each subject with the constraint that half of the thematic choices be on the left and half
be on the right. The presentation order of the 10 items was also randomly determined for
each subject.
Novel Word condition. The materials and procedure for this condition were iden-
tical to those of the No Word condition, with one change. Children in this condition were
told that the puppet could talk in puppet talk. They were instructed to listen carefully to
find the right picture. The puppet gave the target picture an unfamiliar name and used the
same name in the instructions for picking a choice picture. For example, the puppet might
say, “See this? It is a sud. Find another sud that is the same as this sud.” Ten meaningless
one-syllable words were used and a different random assignment of words to pictures was
made for each child.

Results
When children in the No Word condition had to select between another
category member and a thematically related object, they chose the the-
matic relation almost half of the time. They selected other category mem-
bers a mean of 5.95 times out of 10 (59%), SD = 2.28. This was not
significantly different from chance. When the target picture was labeled
with an unfamiliar word children were much more likely to select cate-
gorically. They now chose the other category member a mean of 8.29
times out of 10 (83%), SD = 1.82. This was significantly different from
chance, t(20) = 8.08, p < .Ol, and was significantly different from the
No Word condition, t(39) = 3.63, p < .OOl. The effect held up over every
item and was significant when items rather than subjects were treated as
a random factor, paired t(9) = 8.40, p < .OOl. As predicted, when children
think they are learning a new word they look for categorical relationships
between objects and pay less attention to thematic relations. These re-

TABLE 1
Stimulus Materials for Experiment 1
Standard object Taxonomic choice Thematic choice
Police car Car Policeman
Tennis shoe High-heeled shoe Foot
Dog m Dog food
Straight backed chair Easy chair Man in sitting position
Crib Crib Baby
Birthday cake Chocolate cake Birthday present
Blue jay Duck Nest
Outside door Swinging door Key
Male football player Man Football
Male child in swimsuit Female child in overalls Swimming pool
8 MARKMAN AND HUTCHINSON

suits support the hypothesis at least for very young children and basic
level categories. The next three studies examine whether the effect holds
up for familiar superordinate level categories and for newly learned object
categories.
EXPERIMENT 2
This study tests the hypothesis that hearing a new word will induce
older preschoolers to look for superordinate level taxonomic relations
rather than thematic relations. The superordinate level of categorization,
according to Rosch and her colleagues, is more abstract than the basic
level. Members of superordinate level categories share fewer attributes,
especially perceptual attributes, than members of basic level categories
(Rosch, 1978; Rosch et al., 1976; Mervis & Rosch, 1981).
Methods
Subjects
Sixty children attending nursery schools in Palo Alto, California, and
surrounding towns participated in the study. They ranged in age from 4
years 4 months to 5 years 3 months, with a mean age of 4 years 10 months.
An additional two children were dropped from the study, one who did
not cooperate with the experimenter and one who did not understand the
task. The children were randomly assigned to four conditions, 15 per
condition, with the constraint that the conditions be roughly equated for
age and sex.
Procedure
No Word condition. The procedure used in this condition was very
similar to that used in the No Word condition of Experiment 1, except
that superordinate level categories were used instead of basic level cat-
egories. The experimenter saw each child individually for one 15 to 20-
min session. Subjects were shown 30 colorful pictures of common ob-
jects. Ten of the pictures served as targets. Associated with each of the
target pictures were two choice pictures. One of the choice pictures was
related in a thematic way to the target (e.g., milk/cow). The other choice
picture was a member of the same superordinate category as the target
(e.g., pig/cow). An attempt was made to use a variety of thematic rela-
tions rather than just one, so as not to limit the generality of the results.
A complete list of the stimulus materials appears in Table 2. As in Ex-
periment 1, the instructions in the No Word condition were designed to
make it as clear as possible that we were looking for a taxonomic match.
On each trial in the No Word condition, the experimenter, using a hand
puppet, said, “I’m going to show you a (new) picture. Then you’ll have
 CONSTRAINTS ON WORD MEANING 9

TABLE 2
Stimulus Materials for Experiments 2 and 3
Standard object Taxonomic choice Thematic choice
cow Pig Milk
Ring Necklace Hand
Door Window Key
Crib Adult bed Baby
Bee Ant Flower
Hanger Hook Dress
Cup Glass Kettle
Car Bicycle Car tire
Sprinklern Watering can Grass
Paintbrush” Crayons Easel
Trainb Bus Tracks
Dab Cat Bone
(1This set was used only in Experiment 2.
b This set was used only in Experiment 3.

to find another one that is the same kind of thing.” The experimenter
then placed the target picture face up on the table directly in front of the
child, and said, “See this?” She placed the two choice pictures to the
left and right of the target, then said, “Can you find another one that is
the same kind of thing as this one? Find another one that is the same
kind of thing as this one.” The left-right position of the choice pictures
was randomized for each child in such a way that thematic and taxonomic
choices each appeared half the time on the left and half the time on the
right, across the 10 stimulus triads. The order of presentation of triads
was also randomized for each subject. After children made a choice, they
were asked to justify their response: “How do you know these two are
the same kind of thing?”
Novel Word condition. The materials and procedure for this condition
were identical to those of the No Word condition, except that the target
picture was now labeled with a novel word. Children were told that the
puppet could talk in puppet talk, and that they were to listen carefully to
what he said. The instructions now included an unfamiliar label for the
target: “I’m going to show you a kind of dax. Then you’ll have to find
another kind of dax. See this? It’s a kind of dax. Can you find another
kind of dax?” A different meaningless one-syllable word was used for
each target picture. Children again were asked to justify their choices.
Because they were given a label, we expected children in the Novel Word
condition to choose the taxonomically related picture more often than
children in the No Word condition.
Two additional control conditions were included to attempt to rule out
10 MARKMAN AND HUTCHINSON

one alternative explanation for increased taxonomic responding in the

Novel Word condition. We are arguing that when children hear a word,
they focus on categorical relationships because of general knowledge of
what nouns encode, and not because of specific knowledge about the
word’s meaning. But children already knew real word names for the
target pictures and they conceivably could have translated the unfamiliar
labels. Translation of the unfamiliar words into known words might help
children choose taxonomically. We could not control for the possibility
that children were translating, but we did run one condition to determine
what word children might translate into if they were translating, and
another condition to determine what effect such translation would have.
Translation condition. Children might translate the unfamiliar word
either into a basic level word or into a superordinate level word. For
instance, given a cow as a target picture and told that it was a “dax,”
they might translate “dax” into “cow” (basic level) or they might trans-
late “dax” into “animal” (superordinate level). In the Translation con-
dition, we looked at the kinds of translations children make when they
are explicitly asked to translate the unfamiliar word. The procedure was
identical to that of the Novel Word condition except children were also
asked “What do you think dax means?“-once right after the target
picture was introduced (but before children saw the choice pictures), and
again after they made a choice.
When we analyzed the children’s translations of these novel words, we
found that on the first translation, before seeing the choice pictures, chil-
dren almost never translated the unfamiliar labels into superordinate
terms (only 6.7% of the time). Forty-seven percent of the first translations
were basic level words. The rest of the translations were descriptive
phrases. Even after making a choice, subjects still produced superordi-
nates only 7.3% of the time, while the mean percentage of basic level
translations decreased to 19%. Thus if children in the Novel Word con-
dition spontaneously translated the unfamiliar word, they would be very
unlikely to translate it into a superordinate level word.
Basic Word condition. If children are going to translate into a single
known word, it will be into a basic level category term. The Basic Word
condition tested whether translation into a basic level word would facil-
itate taxonomic responding.
The materials and procedure for this condition were identical to those
in the Novel Word condition. The instructions simply substituted the
basic level word for the target, in place of the unfamiliar label. For the
example with the cow target, the experimenter said: “I’m going to show
you a picture of a cow. Then you’ll have to find another picture that is
the same kind of thing. See this? It’s a cow. Can you find another one
 CONSTRAINTS ON WORD MEANING 11

that is the same kind of thing as this one?” As in the other three condi-
tions, children were asked to justify their choices.
Results
As is typical for children this age, when no word was present they
made a number of thematic choices. When children in the No Word
condition had to select between another member of the same superordi-
nate category and a thematically related object, they chose the categorical
relation only 4.93 times out of 10 (49%), SD = 2.88. This was not different
from chance. As predicted, the presence of a new word caused children
to focus more attention on taxonomic relations. When the target picture
was labeled with an unfamiliar word, children now chose the other cat-
egory member a mean of 6.87 times out of 10 (69%), SD = 2.55. This
was more often than would be expected by chance, t(14) = 2.75, p <
.05. The difference between the conditions was significant by a one-tailed
t test, t(28) = 1.88, p < .05.
Children who heard the target described with a basic level word chose
the categorically related object 57% of the time. This was not significantly
different from either the Novel Word or the No Word condition. It was
also not significantly different from chance.
The results were more clear-cut when items were treated as a random
factor and conditions as a within groups factor. Pictures labeled with
unfamiliar words elicited significantly more taxonomic responses than the
same pictures not labeled, paired t(9) = 4.95, p < .002. This effect held
up for 9 out of the 10 stimulus triads. Pictures labeled with a basic object
word elicited an intermediate number of taxonomic responses: signifi-
cantly more than in the No Word condition, paired t(9) = 4.14, p < .Ol,
but significantly fewer than in the Novel Word condition, paired t(9) =
2.49, p < .05.
Converging evidence for the hypothesis came from the justifications
that children gave for their choices. Two raters coded the justifications
as thematic, categorical, or irrelevant, and agreed on 90% of the classi-
fications. Thematic justifications expressed an interactive relationship be-
tween the target and the object chosen. An example is “The cow makes
milk.” Children justified more of their choices thematically when they
heard no word (51%) than when they heard an unfamiliar word (19%),
t(28) = 3.17, p < .005. Likewise, they justified more of their choices
thematically when they heard a familiar basic level word (38%) than when
they heard an unfamiliar word, t(28) = 1.92, p = .062. There was no
difference in the numbers of thematic justifications given in the No Word
and Basic Word conditions.
Justifications and choices were not perfectly correlated, as the kind of
12 MARKMAN AND HUTCHINSON

justification children gave did not always match the kind of choice they
made. Children seemed to explain their choices in terms of the task as
they saw it. For example, when children chose thematically in the Novel
Word condition, they seemed reluctant to justify the thematic choice with
a thematic explanation. When they heard a novel word, they justified
thematic choices with a thematic explanation an average of only 44% of
the time. The children seemed to be in conflict between having chosen
thematically but believing that the word implies a taxonomic relation.
After choosing a cow and milk as being the same kind of thing, for ex-
ample, many children did not justify their selection in the most natural
way (stating that milk comes from a cow). They had to manufacture some
justification to satisfy the experimenter, and so ended up giving a rela-
tively high proportion of irrelevant justifications for their thematic
choices such as “I don’t know” (mean = 39%), compared to children in
the No Word condition (mean = 6%), t(24) = 2.52, p < .02. When
children in the No Word condition made a thematic choice they did not
have the same conflict and were quite willing to give thematic justifica-
tions. These children, who did not hear a word, justified their thematic
choices thematically the majority (a mean of 84%) of the time. The dif-
ference between the Novel and No Word conditions in propensity to
justify thematic choices thematically was significant, t(24) = 2.71, p <
.02. Children in the Basic Word condition justified thematic choices the-
matically an average of 67% of the time. This was not significantly dif-
ferent from either the Novel Word or No Word condition.
In sum, when an object was labeled with an unfamiliar word, children
were more likely to look for another object from the same superordinate
level category than when the object was not labeled. Children almost
certainly were not translating the novel word into a superordinate level
term, so that cannot account for the effect. If children were translating
the term into a basic level word for the object, that would have helped
them to choose a categorically related object. However, the justifications
from children hearing a novel word differed from those of children hearing
a basic level word, suggesting that translation into known terms was not
accounting for the results.
EXPERIMENT 3
Experiment 3 is a modified replication of Experiment 2. In Experiment
2 the No Word instructions (“Find another one that is the same kind of
thing”) were designed to promote as much taxonomic responding as pos-
sible. The No Word condition was the baseline measure of taxonomic
responding, and we hypothesized that the presence of a new word would
elevate taxonomic responses above even this baseline. Thus, this was a
conservative test of the hypothesis.
 CONSTRAINTS ON WORD MEANING 13

In Experiment 3 we attempted to make the No Word instructions more

neutral. A neutral instruction is more like the natural language learning
context since in both cases, children view objects and hear them labeled
without instructions about what relations to attend to. In this study, chil-
dren were asked to “Find another one.” This is less explicit than the
earlier instructions, but for adults still clearly implies that taxonomic
similarity is called for. In this way we could compare what children nat-
urally found salient to their choices when they heard a novel word. In
Experiment 3 we also used slightly younger children, who would be ex-
pected to show a greater baseline preference for thematic relations.

Methods

Subjects
Sixty children attending nursery schools in Palo Alto, California, and surrounding towns
participated in the study. They ranged in age from 4 years 0 months to 4 years 10 months,
with a mean age of 4 years 5 months. An additional seven children were dropped from the
study because they showed response bias, always choosing the picture on the same side.
The children were randomly assigned to three conditions, 20 per condition, with the con-
straint that the conditions be roughly equated for age and sex.

Procedure
NO Word condition. The procedure used in this condition was very similar to that
used in the No Word condition of Experiment 2. Subjects were shown 30 colorful pictures
of common objects. Ten of the pictures served as targets. Associated with each of the target
pictures were two choice pictures, one thematically related to the target and one taxon-
omically related to the target. Eight out of ten of the picture sets were the same as those
used in Experiment 2. The two new sets are noted in Table 2.
On each trial in the No Word condition, the experimenter said, “I’m going to show you
something. Then I want you to think carefully, and find another one. See this? Can you
find another one?” After children selected a picture, the experimenter asked them to justify
their choice: “How did you know it was this one.7” The left-right position of the choice
pictures was again randomized for each child in such a way that thematic and taxonomic
choices each appeared half the time on the left and half the time on the right. The order of
presentation of triads for each subject was also randomized.
Novel Word condition. The materials and procedure for this condition were iden-
tical to those of the No Word condition, except that a novel word was used to describe the
target picture. Children were told that the puppet could talk in puppet talk, and that they
were to listen carefully to what he said. The instructions included an unfamiliar label for
the target, which the child was asked to repeat, in order to ensure attention to the word:
“I’m going to show you a dax. Then I want you to think carefully, and find another dax.
See this dax? Can you say dax? Can you find another dax?” Children were again asked to
justify their choices.
hsic Word condition. The materials and procedure for this condition were identical
to those in the Novel Word condition. The instructions simply substituted the basic level
word for the target in place of the unfamiliar word. For the example with the cow target,
the experimenter said: “I’m going to show you a cow. Then I want you to think carefully,
and find another cow. See this cow? Can you say cow? Can you find another cow?”
14 MARKMAN AND HUTCHINSON

Children seemed willing to make a choice, despite the fact that neither of the choice pictures
was a cow (the two choices were pig and milk). As in the other two conditions, children
were asked to justify their choices.

Results
As usual, when children in the No Word condition had to choose be-
tween another member of the same superordinate category and a the-
matically related object, they often chose the thematic relation. They
selected the other category member a mean of only 2.50 times out of 10
(25%), SD = 2.11. This was less often than would be expected by chance,
t(19) = 5.17, p < .OOl. When the target picture was labeled with an
unfamiliar word, children were much more likely than children hearing
no label to select categorically. They now chose the other category
member a mean of 6.45 times out of 10 (65%), SD = 1.60. This was more
often than would be expected by chance, t(19) = 3.96, p < .OOl. Children
in the Novel Word condition selected the other category member signif-
icantly more often than children in the No Word condition, t(38) = 6.50,
p < .OOl.
When the data were analyzed with items as a random factor and con-
ditions as a within groups factor, we again found the predicted difference.
Pictures labeled with novel words elicited significantly more taxonomic
responses than the same pictures not labeled, paired t(9) = 8.90, p <
.OOl, and the difference held up for every item.
As in Experiment 2, converging evidence for the hypothesis came from
the justifications that children gave for their choices. Children who heard
a novel word tended to give more justifications that referred to the cat-
egorical relations between the objects, while children who did not hear
a label for the objects referred more to thematic relations. Children were
credited with referring to a categorical relationship if they attempted to
find a common property for the objects (e.g., “They are both round”) or
a common function (e.g., “You can wear both of them”), or if they gave
the two objects a common label (e.g., “They are both animals”). Children
were scored as giving a thematic justification, if they described an inter-
active relationship between the two objects (e.g., “The cow makes
milk”). Two raters coded the justifications into categorical, thematic, and
miscellaneous categories and agreed on 94% of the classifications.
Children who heard a novel label for the picture justified 34% of their
choices by referring to a categorical relation while children who did not
hear a label gave categorical justifications only 10% of the time, t(38) =
2.99, p < .005. Children justified fewer of their choices thematically when
they heard an unfamiliar word (25%) than when they heard no word
(67%), t(38) = 4.34, p < .OOl. Thus, children focused more on categorical
relationships when they heard an unfamiliar word than when they heard
 CONSTRAINTS ON WORD MEANING 15

no word. Moreover, as in Experiment 2, even when children chose the-

matically in the Novel Word condition, they seemed reluctant to justify
the thematic choice with a thematic explanation. When they heard an
unfamiliar label, they justified thematic choices with a thematic expla-
nation only an average of 44% of the time. Children in the No Word
condition who did not hear a label justified thematic choices thematically
an average of 79% of the time, t(38) = 2.02, p < .06.
In sum, when young children are asked to classify things, they often
classify them thematically. But simply hearing a new word induces chil-
dren to focus more on categorical relationships. These results, although
supporting the hypothesis, need to be interpreted in light of the results
from the Basic Word condition. Children in the Basic Word condition
gave just as many taxonomic responses as children in the Novel Word
condition. The mean percentage of taxonomic responses in the Basic
Word condition was 62%, as compared to 65% for the Novel Word con-
dition. The percentage of choices justified thematically in the Basic Word
condition (38%) also did not differ significantly from the percentage of
choices justified thematically in the Novel Word condition (25%). The
question is, were children in the Novel Word condition really translating
into a basic level word?
A closer examination of the data suggests that translation into a basic
level word is not accounting for the advantage of the novel word. That
is, the Novel Word condition and Basic Word condition differ in several
ways. One piece of evidence for a difference is a significant condition x
item interaction, F(18,513) = 2.34, p < .002. The relative difficulty of
different items, in terms of how many taxonomic responses they elicit,
stays approximately the same for the Novel Word and No Word condi-
tions, but changes for the Basic Word condition. Some items received a
high number of taxonomic responses in the Basic Word condition and a
low number of taxonomic responses in the Novel Word condition, or vice
versa. For example, one target picture was a cup. The taxonomic choice
was a glass. Children who heard the basic level word “cup” chose the
glass much more often (95% of the time) than children who heard the
unfamiliar word “biv” (40% of the time). It seems that children were
very willing to overgeneralize the word “cup” to mean “glass,” too. Cup
functioned almost as a superordinate term for both cup and glass. Chil-
dren were probably not translating “biv” into “cup,” or they would have
chosen the glass a high percentage of the time-in the Novel Word con-
dition as well.
16 MARKMAN AND HUTCHINSON

It may be that the basic word can only help children to choose tax-
onomically when they can easily overgeneralize it. We examined the jus-
tifications children gave to see if there was evidence for overgeneraliza-
tion of basic level terms. Children’s justifications in the Novel Word and
Basic Word conditions, and also their translations in the Translation con-
dition of Experiment 2, were coded for the number of times children
spontaneously called both the target and the taxonomic choice by the
same basic level word. We had tried to choose the target pictures so that
their basic word names could not function as names for both the target
and the taxonomic choice. But apparently, we were not totally successful,
as we found three items for which children naturally seemed willing to
overgeneralize the basic word name: cup overgeneralized to glass, hanger
overgeneralized to hook, and bee overgeneralized to ant.
The data were reanalyzed without these three ill-chosen items. When
this was done the condition x item interaction disappeared. The new
mean percentages of taxonomic responses for each condition were 64%
for the Novel Word condition, 50% for the Basic Word condition, and
24% for the No Word condition. The difference between the Novel Word
condition and the Basic Word condition in number of taxonomic re-
sponses now approached significance, t(38) = 1.81, p = .075..The Basic
Word condition still had significantly more taxonomic responses than the
No Word condition, t(38) = 3.48, p < .002. When the results were rean-
alyzed with items as a random factor and conditions as a within group
factor, the differences were more striking. Items described by basic level
words elicited significantly fewer taxonomic responses than items de-
scribed by unfamiliar words, paired t(6) = 2.46, p < .05. More taxonomic
responses were still given to items in the Basic Word condition than to
the same items in the No Word condition, paired t(6) = 3.90, p < .Ol.
One argument that children were not translating the unfamiliar word,
then, is that the pattern of responses to items in the Novel Word and
Basic Word conditions was different. Moreover, the pattern for the Basic
Word condition suggests that the basic word was more likely to lead to
taxonomic responding when children could overgeneralize it.
Other evidence that children in the Novel Word condition were not
simply translating the word into a basic level term comes from the jus-
tifications for thematic choices. Children who heard a basic level word
did not seem as reluctant to justify thematic responses thematically as
children who heard an unfamiliar word. The proportion of thematic re-
sponses justified thematically was 70% in the Basic Word condition, as
opposed to 44% in the Novel Word condition, t(38) = 2.28, p < .05. Thus
the justification data show that the basic level word did not shift chil-
dren’s attention away from thematic relations as much as the unfamiliar
word did. One possible explanation for this is that children in the Novel
Word condition were trying to figure out the meaning of the new word,
 CONSTRAINTS ON WORD MEANING 17

and were therefore more aware that a word was involved. Consequently,
they were more aware than children in the Basic Word condition that
they should be searching for categorical relations.
There is one last piece of evidence that something different was going
on in the Novel Word and Basic Word conditions. If children in the Novel
Word condition were translatinig into a basic level word, then we might
suppose that the basic word should be salient to them and should appear
fairly often in their justifications. But the proportion of justifications in
which a basic word name for the target appeared was only 28% for the
Novel Word condition, compared to 49% for the Basic Word condition,
t(38) = 2.78, p < .Ol.
In sum, the results for Experiment 3 replicated the findings of Exper-
iment 2, providing even stronger support for the hypothesis. Hearing a
new word diminishes children’s tendency to look for thematic relation-
ships, and causes them to look for categorical relationships instead. There
is at least some evidence that children focus on categorical relationships
because of the sheer presence of the word, and not because of any par-
ticular knowledge about the meaning of the word.
EXPERIMENT 4
Post hoc analyses from Experiment 3 were used to argue that trans-
lation of the novel word into a basic level word could not account for
children’s categorical choices. Experiment 4 is designed to provide ad-
ditional evidence that children use abstract knowledge about words rather
than specific known meanings to facilitate taxonomic responding. In this
study, pictures of artificial objects were used instead of real objects.
Children are not likely to translate unfamiliar names for these pictures
into known words, because they do not know real word names for them.
If the presence of an unfamiliar word still causes children to shift from
thematic to taxonomic responding when the materials are also unfamiliar,
then this would rule out translation as an explanation for the effect.
Methods
Subjects
Thirty-two children attending nursery schools in Palo Alto, California, and surrounding
towns participated in the study. The children ranged in age from 4 years 6 months to 5
years 11 months, with a mean age of 5 years 2 months. An additional three children were
dropped from the study because of response bias. The children were randomly assigned to
two conditions, 16 per condition, with the constraint that the conditions be roughly equated
for age and sex.

Procedure
The design and procedure for this study are essentially the same as that of Experiment
3. The main difference is that the experimenter fust described the taxonomic and thematic
relations for the artificial objects before asking children to select the picture that was like
the target.
18 MARKMAN AND HUTCHINSON

FIG. 1. Sample taxonomic training picture in Experiment 4.

NO Word condition. Each child was seen individually for one 20-min session. Sub-
jects were shown eight sets of pictures in random order. Each set included a target picture,
and two choice pictures, one thematically related and one taxonomically related to the
target. Before children saw the target picture and the two choices, they were shown two
training pictures that illustrated how the target picture related to each of the choice pictures.
One picture showed the target object and the taxonomic choice, side by side. For these
pairs, children were told a common function that the two objects shared. An example
taxonomic training picture is shown in Fig. 1. For this example, the experimenter said,
“This swims in the water” (pointing to the left hand object). “This swims in the water”
(pointing to the right hand object).
A second training picture showed the target and the thematic choice in an interactive
relationship. The experimenter told the children how the two objects interacted. The the-
matic training picture for the set just given is shown in Fig. 2. For this example, the
experimenter said, “This catches this” (pointing to the objects she was referring to as she
said the sentence). Children were asked to repeat the spoken information to make sure that
they were paying attention. The first training picture was left on the table as the second
training picture was introduced, so that children could see the connection between the
target in the first picture and the target in the second picture. The order of presentation of
training pictures was randomized so that taxonomic and thematic training pictures were
each presented first half of the time.
A second example taxonomic training picture is shown in Fig. 3. For this example the
experimenter said, “This pokes holes in things” (pointing to the left hand object). “This
pokes holes in things” (pointing to the right hand object). The thematic training picture for
the same set is shown in Fig. 4. For this picture, the spoken information was “You keep
this in here.”
After children saw the two training pictures in a set, the pictures were removed from the
table. The rest of the trial was a match to sample task following the same procedure as the
No Word condition of Experiment 3. The experimenter said, “I’m going to show you

FIG. 2. Sample thematic training picture in Experiment 4.

 CONSTRAINTS ON WORD MEANING 19

FIG. 3. Sample taxonomic training picture in Experiment 4.

something. Then I want you to think carefully, and find another one.” The experimenter
then placed the target picture face up on the table directly in front of the child and said,
“See this?” She placed the two choice pictures to the left and right of the target and said,
“Can you find another one?” Note that the choices were pictures of the individual objects
as in the previous studies, rather than pictures of two objects together. After children made
a choice, the experimenter asked them to justify their response: “How did you know it
was this one?” The left-right order of choices was randomized so that taxonomic choices
each appeared half the time on the left and half the time on the right.
Novel Word condition. The materials and procedure for this condition were iden-
tical to those of the No Word condition, except that a novel word was used to label the
target picture during the match to sample task. After children saw the training pictures, the
experimenter said, “I’m going to show you a dax. Then I want you to think carefully, and
find another dax. See this dax? Can you say dax? Can you find another dax?” A different
unfamiliar word was used for each set. Children again were asked to justify their choices.
Because of the unfamiliarity of the materials, children in both conditions found it difficult
to justify their responses, so the justifications will not be discussed.

Results
The results for the choices were parallel to those found for Experiments
2 and 3. As usual, when children in the No Word condition had to select
between another member of the same category and a thematically related
object, they often chose the thematic relation. They selected the other
category member a mean of only 3.00 times out of 8 (37%), SD = 1.79.
This was significantly less than chance, t(l5) = 2.24, p < .05. When the

FIG. 4. Sample thematic training picture in Experiment 4.

20 MARKMAN AND HUTCHINSON

target picture was labeled with an unfamiliar word, children were more
likely to select categorically. They now chose the other category member
a mean of 5.06 times out of 8 (63%), SD = 2.38. This was more than
chance by a one-tailed t test, t(H) = 1.78, p < .05. Children hearing a
novel word were significantly more likely to select an object from the
same category than children not hearing a label, t(30) = 2.77, p < .Ol.
When the results were analyzed with items as a random factor and
conditions as a within groups factor, we again found the predicted differ-
ence. Pictures labeled with unfamiliar words elicited significantly more
taxonomic responses than the same pictures not labeled, paired t(7) =
4.07, p < .005, and the difference held up for every item.
DISCUSSION
The hypothesis tested by these studies is that children place an abstract
constraint on what single nouns might mean. Children limit words to refer
mainly to objects that share some property or function rather than al-
lowing words to refer to objects that are united by thematic relations.
This would help explain how children acquire words that refer to cate-
gories even though, in many other situations, they seem to find the the-
matic associations between objects to be more salient. The simple pres-
ence of a noun, even an, unfamiliar one such as “dax,” should cause
children to search for objects that share perceptual or functional prop-
erties. Thus, labeling a picture as “a dax” and asking children to find
“another dax” should help override their preference for choosing the-
matically.
Overview of Results
The results from four studies supported the hypothesis. The main re-
sults from all of the studies are summarized in Table 3.
As can be seen from the results of Experiment 1, even children as
young as 2 and 3 years place constraints on what an unfamiliar word
might mean. When presented with two basic level objects-for example,
two different kinds of dogs-and a third object that was thematically
related, such as dog food, very young children often selected a dog and
dog food as being the same kind of thing. If, however, one of the dogs
was called by an unfamiliar label, e.g., “dax,” and children were told to
find another dax, they were now much more likely to select the two dogs.
By 4 or 5 years of age, children have set further constraints on what a
word might mean, as Experiments 2 and 3 demonstrate. A word induces
them to search for categorical relations even among objects that can only
be related at the superordinate level of categorization. For example, with
no word present, children often selected a dog and dog bone as being the
 CONSTRAINTS ON WORD MEANING 21

same kind of thing because of the strong thematic association between

dog and bone. When one of the dogs was called a “dax,” however, and
children were asked to find another dax, they more often selected a dog
and a cat as being the same, because they are both in the same superor-
dinate category, animals.
At the superordinate level of categorization, the same pattern of
choices was obtained for two different sets of instructions. Even when
the No Word instructions emphasized taxonomic relations, as in Exper-
iment 2 (“Find another one that is the same kind of thing as this one.“),
we found an increase in taxonomic responding in the Novel Word con-
dition. In Experiment 3 when the No Word instructions were more neutral
(“Find another one.“), there was an even bigger shift toward more tax-
onomic responding in the presence of a new word.
The justification data corroborated the choices. Children who heard a
novel word tended to give more justifications that referred to the cate-
gorical relations between the objects, whereas children who did not hear
a label for the objects referred more to thematic relations (see column 2
of Table 3). Even when children chose thematically in the Novel Word
condition, they seemed reluctant to justify the thematic choice with a
thematic explanation (see column 3 of Table 3). For example, when chil-
dren select a dog and a dog bone as being the same, they ordinarily justify

TABLE 3
Main Results for Experiments 1 to 4
Percent of all choices Percent of thematic
Percent of taxonomic justified choices justified
choices thematically thematically
Experiment 1: Basic level categories in 2-3-year-olds (taxonomically biased instructions)
No Word 59% - -
Novel Word 83% - -
Experiment 2: Superordinate level categories in 4-S-year-olds (taxonomically biased in-
structions)
No Word 49% 51% 84%
Novel Word 69% 19% 44%
Basic Word 57% 38% 67%
Experiment 3: Superordinate level categories in 4-year-olds (neutral instructions)
No Word 25% 67% 7%
Novel Word 65% 25% 44%
Bask Word 62% 38% 70%
Experiment 4: Unfamiliar categories in 4-S-year-olds (neutral instructions)
No Word 37% - -
Novel Word 63% - -
22 MARKMAN AND HUTCHINSON

this by saying that the dog eats the bone. However, those children who
had heard the dog labeled with an unfamiliar term, yet nevertheless se-
lected the dog bone, now justified their choice by saying that the dog and
the bone were both white, for example, or refused to explain their selec-
tion. There was no such reluctance to justify thematic choices themati-
cally when no label was given.
The hypothesis is that the presence of an unfamiliar word shifts chil-
dren’s attention to taxonomic relations because of an abstract constraint
children place on possible word meanings, and not because they know
the meaning of the word. Thus, we would like to rule out translation into
a known word as accounting for the effect. Based on our results and on
other research on language acquisition (e.g., Anglin, 1977; Clark, 1973;
Mervis & Rosch, 1981), it is extremely unlikley that children would trans-
late the unfamiliar terms into superordinate level terms. Children seeing
a dog and hearing the word “dax,” for example, would be very unlikely
to translate “dax” into “animal” or “mammal” or even “pet.” If they
were translating at all, they would think that “dax” meant “dog.” Even
when children were specifically asked to translate the novel word in the
Translation condition of Experiment 2, they rarely translated it into a
superordinate level term. Thus translation into a superordinate level word
cannot account for the increase in taxonomic responding for children who
heard the picture labeled with an unfamiliar word.
Had children been translating into a basic level term, it would have
helped them to select taxonomically. When the target picture was labeled
with a basic object word, children selected more taxonomically related
pictures than children who heard no label (see column 1, Experiments 2
and 3 in Table 3). However, the increase in taxonomic responding when
children heard a novel word cannot be fully accounted for by translation
of the novel word into a basic level word. Children’s justifications provide
one source of evidence that they were not simply translating the word
into a basic level term. When children heard the novel words they seemed
reluctant to justify their choices thematically, even in those cases where
they had selected a thematically related picture. In contrast, when chil-
dren heard familiar basic level words, they were happy to justify their
choices in terms of thematic relationships (see columns 2 and 3 of Table
3). In the Basic Word condition, children heard, for example, the dog
called “dog” and were told to “find another dog.” When these children
selected the dog and dog bone, they showed no reluctance to justify their
choice by saying that the dog eats the bone. Perhaps when children heard
the basic level term they tried to generalize it to other items. When they
failed, the fact that the familiar term was a count noun may not have
been salient enough to prevent them from claiming that it referred to a
 CONSTRAINTS ON WORD MEANING 23

thematic relationship. In contrast, children who heard the novel word

tried to figure out what it might mean. This may have heightened their
awareness that a word was involved. As a consequence, children in the
Novel Word condition may have felt more reluctant than children in the
Basic Word condition to describe the word as referring to a thematic
relation.
The most compelling evidence that translation into known terms cannot
account for the results comes from the fourth study, where unfamiliar
objects were used as well as unfamiliar words. Here children were shown
three novel objects. They were taught a taxonomic relation for two of
the objects and a thematic relation for two. When no label was used
children often selected the two objects that were related thematically as
being the same. When an unfamiliar word was used to label the target
picture, children now selected the two objects that were related taxo-
nomically. Children could not have been translating in this study because
they did not know what these unfamiliar objects were and had no familiar
labels for them. Nevertheless, the results from this study replicated the
results from the studies that used familiar objects. Again, the presence
of an unfamiliar meaningless word caused children to shift from selecting
objects that are thematically related to selecting objects that are taxo-
nomically related. This suggests that children have placed an abstract
constraint on what words can mean that is not mediated by the meaning
of known terms.
By constraining the meaning of a term to categorical relations, children
are able to rule out a huge number of other potential meanings for any
given term. For example, suppose an adult points to a cup and says
“cup.” With no constraints on possible meanings, a child would have to
consider that the table might also be a “cup” because the cup was on
the table, or that coffee is also called “cup” because the cup was filled
with coffee, or that mother might be a “cup” because mother was lifting
the cup. All of these relational meanings would be eliminated from con-
sideration by the constraint that nouns refer to object categories. By
limiting the number and kind of hypotheses that children need to consider,
this constraint tremendously simplifies the problem of language learning.
Origins of Sensitivity to Constraints on Word Meaning
These findings raise the question of how children come to constrain
their hypotheses about what a word can mean. What leads children to
assume that a word is likely to refer to objects that are similar in some
way rather than to objects that participate in the same event or context?
There are at least two possibilities. One is that sensitivity to the constraint
is innate -from the start, children assume words will refer to categories
24 MARKMAN AND HUTCHINSON

of similar objects. Having such implicit knowledge would provide chil-

dren with an entry into the formidable problem of learning language.
Children would at least be able to readily acquire count nouns, and once
they had a reasonable vocabulary of category terms, they could then
begin to comprehend other linguistic forms. In fact, the huge majority of
children’s first words are count nouns (Clark, 1983; Nelson, 1973; Hut-
tenlocher, 1974).
Another possibility is that the constraint is induced from early language
experience. Having learned many count nouns, almost all of which refer
to objects that are taxonomically related, children may come to expect
this to be true of subsequent terms they learn. If so, then this induction
must take place fairly rapidly at an early point in language acquisition,
since we found that even 2-year-olds believe that count nouns are more
likely to refer to objects that belong to the same category than to objects
that are thematically related.
It is not clear whether or not very young language learners limit the
constraint to count nouns. Particularly if children have some innate
knowledge of the constraint, they may at first overextend it, indiscrimi-
nately believing that any word they hear must refer to a taxonomic cat-
egory. Only somewhat later might they become sensitive to form class
and expect count nouns to be more likely than other classes of words to
refer to categorical relations.
The Role of Language in Aiding Concept Acquisition
Children’s sensitivity to this constraint raises the possibility that lan-
guage may help children acquire new categories. In contrast, it is often
argued that words must map onto concepts that have already been
worked out nonlinguistically (Clark, 1973; Huttenlocher, 1974; Macna-
mara, 1972; Nelson, 1974; Wittgenstein, 1953, 1958). In this view, lan-
guage plays little role in concept learning. But this view may underesti-
mate the importance of language. Young children may create concepts to
fit new words, guided by abstract constraints on word meaning. This
alternative view is a mild form of linguistic determinism (Whorf, 1956),
in that language is believed to shape thought. It is quite different, how-
ever, from Whorf’s conception that each language imposes a particular
world view on its speakers and that cognition is determined and limited
by the specific language one speaks. First, all languages are likely to
share similar constraints on possible meanings for count nouns. Thus the
hypothesis is that, regardless of native language, children look for cate-
gories of similar objects when they hear new nouns. Second, although
nouns help focus children’s attention on categorical relations, we are not
arguing that children would be incapable of forming categories without
exposure to language.
 CONSTRAINTS ON WORD MEANING 25

The small amount of research that bears on this milder form of linguistic
determinism suggests that children can use abstract knowledge of the
semantic correlates of form class to help them discover the concept to
which a word refers. Brown (1957) found that 3- to Syear-old children
interpreted an unfamiliar count noun (“a dax”) as referring to a new
concrete object, whereas they interpreted an unfamiliar mass noun
(“some dax”) as referring to a novel undifferentiated mass. In a study
by Katz, Baker, and Macnamara (1974), children as young as 1’12 years
old interpreted an unfamiliar proper noun (“Dax”) as referring to an
individual. At the same time, these young children understood an unfa-
miliar count noun (“a dax”) as referring to a category of similar objects.
To return to our findings, hearing a noun caused children to shift their
attention from thematic to taxonomic organization. These results lead us
to speculate that linguistic input may serve more generally to shape the
conceptual structure of the child in the direction of greater taxonomic
organization. A word may draw members of a category together for a
child, highlighting their common category membership. Language may
thus play a direct role in making categorical relations a salient and highly
structured mode of organization.
Possible Reasons for Taxonomic Organization of Language
The question arises as to why language is organized this way. Why
don’t words refer typically to objects that are thematically related? As
we earlier pointed out, thematic relations between objects certainly are
important for adults as well as for children. In naturally occurring situ-
ations, objects are not found organized by category, but rather are
embedded in spatial, temporal, and causal contexts. Such relational struc-
tures as events and themes are a common way of organizing information
to make sense of what we encounter (cf. Mandler, 1979; Markman, 1981).
Given that these thematic event-like organizations are a natural way of
construing the world, why should languages force a taxonomic or cate-
gorical structure rather than capturing this thematic bias? Why don’t we
have single words for a boy and his bike, a baby and its bottle, a spider
and its web? One reason may be that if nouns referred exclusively to
relations such as a baby and its bottle or a boy and his bike, there would
be no easy way to express hierarchical taxonomic relations. Because a
taxonomy groups objects into categories nested within broader catego-
ries, it allows deductive inferences to be made that go beyond the Iirst-
hand knowledge one has about a specific object. If one knows, for ex-
ample, that a particular object is an animal, one can be fairly sure that it
takes in food, moves about, reproduces, and has internal organs. In con-
trast, knowing that something is a “dax,” where “dax” could be a boy
or his bike, tells one very little else about it. One reason why nouns tend
26 MARKMAN AND HUTCHINSON

not to refer to thematically related objects, then, may be because of the

advantages of hierarchical organization.
Another more important reason may be that if a language had single
nouns refer exclusively to pairs of thematically related objects, it would
be at great cost. The enormous expressive power of language would be
lost. The expressive power of language derives from its ability to convey
new relations through combinations of words. There are a potentially
infinite number of thematic relations that one might want to express. The
many thematic relations can easily be described through combinations of
words-e.g., sentences and phrases. If single words referred only to
thematic relations, however, there would be an extraordinary prolifera-
tion of words, probably more than humans could learn. One would need
separate words for a baby and its bottle, a baby and its crib, a baby and
its mother, a baby and its diaper, etc. Thus, the combinatorial power of
language would be wasted. This, then, may be the major reason why
nouns refer primarily to taxonomic categories rather than to thematically
related objects.
REFERENCES
Anglin, J. Word, object, and conceptual development. New York: Norton, 1977.
Brown, R. Linguistic determinism and the part of speech. The Journal of Abnormal and
Social Psychology, 1957, 55, 1-5.
Carey, S. Are children fundamentally different kinds of thinkers and learners than adults?
In S. Chipman, J. Segal, & R. Glaser (Eds.), Thinking and learning skills. Hillside, NJ:
Erlbaum, in press. Vol. 2.
Clark, E. V. What’s in a word? On the child’s acquisition of semantics in his first language.
In T. E. Moore (Ed.), Cognitive development and the acquisition of language. New
York: Academic Press, 1973.
Clark, E. V. Meanings and Concepts. In J. H. Flavell & E. M. Markman (Eds.), Cognitive
Development, Vol. 3 of P H. Mussen (General Ed.), Handbook of child psychology.
New York: Wiley, 1983.
Denney, N. Evidence for developmental changes in categorization criteria for children and
adults. Human Development, 1974, 17, 41-53.
Denney, N., & Ziobrowski, M. Developmental changes in clustering criteria. Journal of
Experimental Child Psychology, 1972, 13, 275-282.
Gelman, R., & Baillargeon, R. A review of some Piagetian concepts. In J. H. Flavell &
E. M. Markman (Eds.), Cognitive development, Vol. 3 of P H. Mussen (General Ed.),
Handbook of child psychology. New York: Wiley, 1983.
Gibson, E. J., & Spelke, E. S. The development of perception. In J. H. Flavell & E. M.
Markman (Eds.), Cognitive development, Vol. 3 of P. H. Mussen (General Ed.), Hand-
book of Child Psychology. New York: Wiley, 1983.
Huttenlocher, J. The origins of language comprehension. In R. L. Solso (Ed.), Theories in
cognitive psychology: The Loyola symposium. Potomac, MD: Erlbaum, 1974.
Inhelder, B., & Piaget, J. The early growth of logic in the child. New York: Norton, 1964.
Katz, N., Baker, E., & Macnamara, J. What’s a name? On the child’s acquisition of proper
and common nouns. Child Development, 1974, 45, 469-473.
Macnamara, J. Cognitive basis of language learning in infants. Psychological Review, 1972,
79. 1-13.
 CONSTRAINTS ON WORD MEANING 27

Mandler, J. M. Categorical and schematic organization in memory. In C. R. Puff (Ed.),

Memory organizarion and structure. New York: Academic Press, 1979.
Mandler, J. M., Scribner, S., Cole, M., & DeForest, M. Cross-cultural invariance in story
recall. Child Development, 1980, 51, 19-26.
Markman, E. M. Two different principles of conceptual organization. In M. E. Lamb &
A. L. Brown (Eds.), Advances in developmental psychology. Hillside, NJ: Erlbaum,
1981. Vol. 1.
Markman, E. M., & Callanan, M. A. An analysis of hierarchical classification. In R. Stem-
berg (Ed.), Advances in the psychology of human intelligence. Hillside, NJ: Erlbaum,
1983. Vol. 2.
Men+, C. B., & Rosch, E. Categorization of natural objects. In M. R. Rosenzweig &
L. W. Porter (Eds.), Annual review of psychology. Palo Alto, CA: Annual Reviews,
1981. Vol. 32.
Nelson, K. Structure and strategy in learning to talk. Monographs of the Society for Re-
search in Child Development, 1973, 38, (Serial No. 149).
Nelson, K. Concept, word and sentence: Interrelations in acquisition and development.
Psychological Review, 1974, 81, 267-285.
Nelson, K. The syntagmatic-paradigmatic shift revisited: A review of research and theory.
Psychological Bulletin, 1977, 84, 93- 116.
Quine, W. V. 0. Word and object. Cambridge, MA: MIT Press, l%O.
Rosch, E. H. Principles of categorization. In E. H. Rosch & B. B. Lloyd (Eds.), Cognition
and categorization. Hillsdale, NJ: Erlbaum, 1978.
Rosch, E. H., & Mervis, C. B. Family resemblances: Studies in the internal structure of
categories. Cognitive Psychology, 1975, 7, 573-605.
Rosch, E. H., Mervis, C. B., Gray, W., Johnson, D., & Boyes-Braem, P. Basic objects in
natural categories. Cognitive Psychology, 1976, 3, 382-439.
Smiley, S. S., & Brown, A. L. Conceptual preference for thematic or taxonomic relations:
A nonmonotonic age trend from preschool to old age. Journal of Experimental Child
Psychology, 1979, 28, 249-257.
Whorf, B. L. Language, thought and reality. Cambridge, MA.: MIT Press, 1956.
Wittgenstein, L. Philosophical investigations. New York: Macmillan, 1953.
Wittgenstein, L. The blue and brown books. New York: Harper, 1958.

REFERENCE NOTE
1. Horton, M. S. Category familiarity and taxonomic organization in young children.
Unpublished doctoral dissertation, Stanford University, 1982.
(Accepted August 16, 1983)