VOL. 69, No.

3 MAY 1962

PSYCHOLOGICAL REVIEW
ON THE STRUCTURE OF ASSOCIATIVE MEANING 1
JAMES DEESE
Johns Hopkins University

Ever since Hobbes wrote "From St. about associative learning, though the
Andrew the mind runneth to St. Peter primary "laws" have always been of
and from St. Peter to Stone," psycKolo- greatest theoretical importance.
gists and philosophers concerned with Since Ebbinghaus, the major experi-
the nature of the mind have been trying mental treatment of association has
to describe the nature of the associa- been with artificial material, mostly
tions that are so evident in human nonsense syllables. Consequently the
verbal behavior. Locke tried to picture relations of contiguity, similarity, etc.,
the structure of the human mind by have been arranged by experimenters,
describing relations between associa- not obtained from the natural relations
tions. He attributed associations to the among linguistic elements.
more or less accidental contingencies of Somewhat earlier than Ebbinghaus,
perceptual qualities in the real world. however, Sir Francis Galton inaugu-
By so doing, Locke became responsible rated the study of naturally occurring
for a major tradition in the theory of associations. For the most part, the
both thought and perception. tradition begun by Galton has remained
Almost all of the theoretical attempts psychometric rather than experimental.
to deal with association stem from Recently, these two traditions have been
Locke. For experimental psychologists combined. Nevertheless, most of the
interested in learning, however, the attempts to deal with the structure or
most important theoretical contribu- organization of naturally occurring as-
tions come from the so-called secondary sociations have been empirical rather
laws of association, introduced by than theoretical, in the tradition of
Thomas Brown and David Hartley. Locke.
Associations themselves are supposed Most of the empirical studies of the
to arise by contiguity, similarity, etc., organization of naturally occurring as-
but they occur in the strengths and dis- sociations have been studies of classifi-
tributions that they do because of their cation of associations. Associations are
frequency, vividness, and so on. It is classified by logical relations or by the
experimental treatment of the sec- dictionary meanings of words which
ondary principles that has led to most occur as stimuli or as responses in free
of the stable empirical generalizations association tests.
1
The work reported in this paper was sup-
A good summary of most of the at-
ported by funds from the National Science tempts to discover the structure of asso-
Foundation, Grant 13055. ciations by classification can be found
161
162 JAMES DEESE

in the first edition of Woodworth's Ex- sociations according to other senses of

perimental Psychology (1938). Wood- meaning but to discover the relation-
worth both reviews a half century of ships associations have with one an-
German, American, and English at- other. That is the purpose of the
tempts at classification and presents present paper.
a classification scheme of his own. Aside from the difficulty summarized
Woodworth's own classification is more above, all previous attempts at classi-
obviously rooted in the association fre- fication, including the interesting one
quencies themselves than are the earlier proposed by Karwoski and Berthold
ones. It is also unique in that it is a (1945), have suffered from another de-
two dimensional classification. Wood- fect. All schemes proposed thus far
worth classifies both by meaning, in the have attempted to deal simultaneously
traditional sense, and by meaningful- with the stimuli and the responses to
ness (in the sense used by Noble and them. That is to say, they attempt to
others). classify the relation between stimuli and
As with the earlier classifications, responses, even though the individual
Woodworth's is a combination of dic- members vary. Thus, the result fre-
tionary and grammar book rules, fil- quently is an attempt to equate such
tered through Woodworth's own verbal pairs as SLOW-Fast and BLUE-FeWow.2
behavior and the verbal behavior of the Apart from the implicit assumptions
investigators whose work he summa- made by collapsing disparate stimulus
rizes. The trouble with it and with all and response terms together, the possi-
classifications to date is that they are bility is ignored that certain reciprocal
not completely rooted in the association pairs, for example, VfORK-Play, PLAY-
process itself. They attempt to impose Work, bear no simple consistent rela-
upon the associations the logical rela- tion to one another.
tions found in grammars and the rela- For this reason, it seems more sen-
tions between words stated in diction- sible to simplify the problem by dealing
aries and thesaureses as well as some with the distribution of responses with
of the relations among objects per- stimuli held constant or with the distri-
ceived in the natural world. Perhaps bution of stimuli for any given response.
the major result of such classifications The distributions to different stimuli or
is to convince us that while associations the stimuli for pairs of responses may
are partly related to the meanings of then be compared. This comparison
words in all of the above senses, associ- should provide the first step in under-
ations are meaningful in precisely none standing how associations are related
of the above senses. to one another,
The meaningfulness of words refers
to organized relations among the words RELATIONSHIPS AMONG ASSOCIATES
and among words and objects in the For the past several years, the pres-
natural world. Associations are re- ent author has been studying the extent
lated to one another in some organized to which the relations among associa-
fashion and also related in some way to tions determine such aspects of immedi-
perceived properties in the natural ate free recall as clustering, importations
world. The only conclusion is that into recall, etc. (Deese, 19S9a, 19S9b,
there is some sense in which one may 1960). This work had its origin in
speak of "associative meaning" and that 2
In this paper, free association stimuli will
the way to discover the associative be given in small capitals and responses in
meaning of words is not to classify as- italics.
 ON THE STRUCTURE OF ASSOCIATIVE MEANING 163

the studies of associative clustering but both elicited Note, Song, Sound,
(Jenkins & Russell, 1952) and category Noise, Music, and Orchestra in varying
clustering (Bousfield, 1953). The basic frequencies.
method has been to relate an index Thus, the relationship between free
based upon the average frequency with association stimuli should be deter-
which all words presented for recall mined from all of the responses that
tend to elicit one another as free as- stimuli have in common. If the asso-
sociates with the number of words ciative meaning of any stimulus is given
correctly recalled, the number of im- by the distribution of responses to that
portations, etc. This index is a power- stimulus, then two stimuli may be said
ful predictor of a number of character- to have the same associative meaning
istics of free recall. Free recall itself when the distribution of associates to
is a representative verbal process, and them is identical. Two stimuli overlap
thus it is evident that the associative in- or resemble one another in associative
terrelations among words are powerful meaning to the extent that they have
determiners of the pattern and fre- the same distribution of associates.
quency with which words are emitted. One may also deal with the associa-
The interword associative index is tive distribution 3 of the responses them-
computed by tabulating the frequencies selves. The associative distribution of
with which each word in a particular a response is the collection of stimuli to
collection (usually a list of words to be which it occurs. This relation is less
presented for recall) occurs as a re- interesting than that between responses
sponse to all of the other words as free to stimuli for a number of reasons. It
association stimuli. Casual inspection does not, of course, specify an associa-
of the matrix that provides the basis for tive concept as does the distribution of
such a tabulation gives striking con- responses to a stimulus, and it is sub-
firmation of what every investigator ject to serious sampling limitations.
who has studied free association has One is fairly sure to pick up any rela-
felt to be the case, namely that associa- tions of statistical significance among
tions exist in well organized and in stimuli by a modest sample of re-
some instances tightly organized net- sponses. With the distribution of stim-
works. The networks are so apparent uli problem, however, there is no way
that they turn up even in random col- to insure that one has sampled all stim-
lections of words, if the words are high uli that are high frequency producers of
in frequency of usage (Deese, 1960). a particular response short of exhaust-
The networks that appear upon such ing the language. Therefore, except
a tabulation provide only part of the insofar as the distribution of stimuli
picture, however, because the words problem is related to the problem of
that appear as responses are artificially associative meaning, it will be ignored
restricted to words which appear as in the balance of this paper.
stimuli. When examined, the original
association frequencies show that stim- SPECIFICATION OF RELATIONSHIPS
uli which do not elicit one another as The idea of associative meaning pre-
responses sometimes have in common sented above implies that one describes
a great many responses. An example,
8
picked more or less at random, is the The word "distribution" is used here in
pair PIANO and SYMPHONY. In a sam- analogy to its linguistic, not its statistical
sense. That is to say it indicates the number
ple of Johns Hopkins University under- of situations to which a particular word
graduates, neither one elicited the other, occurs as a response.
164 JAMES DEESE

and classifies associations purely in This creates intuitive difficulties if one

terms of the relations among the re- defines the relation of associative mean-
sponses to different stimuli. There are ing between words as the common dis-
several ways in which such relations tribution of responses, because pairs of
may be specified. One may count the words often appear to be intuitively re-
number of associations in common in lated when they have no overt responses
any fixed size sample, or one may in common. Upon close examination
weigh the common associations by their all such intuitive cases turn out to be
frequency. For prediction of effects in cases in which the responses to one
verbal behavior, the latter procedure is word are the stimuli to the other. An
probably more precise, and it is the ap- example in the Hopkins data is the
proach taken in the present paper. SOFT-LOUD pair. Both of these words
Also, one may wish to describe the elicit one another, but they have no
ways in which stimuli resemble one other responses in common. Thus,
another in associative meaning. This they could have no associative meaning
implies the study of a collection of re- in common if the possibility of a word
lations between a number of stimuli. occurring as a response to itself is
Factor analysis suggests itself as a way denied.
of reducing the structure within such a Fortunately, a way out of this para-
collection to manageable proportions. doxical situation already exists in a
In any event, the heuristic value of the technique independently designed for
concept of associative meaning will de- other purposes by Jenkins and Cofer
pend upon ways of specifying the rela- (1957) and by Bousfield, Cohen, and
tions between free association stimuli. Whitmarsh (1958). The technique
The purpose of this section is to pro- consists of assuming that a word elicits
vide a basic method for measuring the itself implicitly as an associative 100%
relationship. of the time. This assumed implicit re-
There are several difficulties that sponse is called by Bousfield, Cohen,
stand between the definition of associa- and Whitmarsh the "representational"
tive meaning and the specification of response. The implication of the name
particular associative meanings by a is that this response identifies or repre-
study of the relationships among free sents the symbolic stimulus presented
associations. The first of those is cre- to the individual.
ated by the technique of collecting free Both Jenkins and Cofer (who do not
associations, which almost completely explicitly identify the representational
eliminates the occurrence of the stimu- response) and Bousfield, Cohen, and
lus word as a response to itself. This Whitmarsh implicitly make the assump-
problem must be dealt with first. tion that the representational response
The first step in an analysis of asso- is always the same word as the stimulus
ciations is to tabulate for each stimulus presented. This assumption is very
word in a collection the responses nearly correct for words of common
which the word has in common with cultural use, but it must surely be
any other stimulus words in the collec- incorrect for unusual words. Many
tion. Since, by instructions as well as Hopkins undergraduates, for example,
by technique a word usually does not respond to the words ABBESS and ABYSS
occur as a response to itself in free as- in precisely the same way, which sug-
sociation, the cells in such a tabulation gests that the representational response
in which the same word as response for one of them is not the dictionary
and stimulus intersect will be empty. equivalent of the word. Nevertheless,
 ON THE STRUCTURE OF ASSOCIATIVE MEANING 165

this assumption works with only trivial the distributions of common responses
exceptions for moderately sophisticated may be inverse to one another, the ob-
individuals and words of ordinary fre- tained product-moment correlations of
quency of usage. It is, therefore, the frequencies will often be negative.
adopted here. Thus, a table of product-moment corre-
With this assumption it is possible to lations obtained from Table 1 would
prepare a table like Table 1. This table yield a few zero correlations, a few high
shows the frequency of responses com- positive correlations and a large num-
mon to any two stimuli, with the fre- ber of low negative correlations.
quency of the stimulus word as a Since it is not the relation between
response to itself entered with 100% the forms of the distributions of re-
frequency. Since the N for this table sponses but the relations between the
is 50, the value entered in the stimulus- frequencies themselves that is inter-
response intersection cells is 50. Note esting, a direct measure of overlapping
that it is not necessary for all the frequency is appropriate. To measure
stimulus-response intersections to ap- the commonness of associative meaning,
pear. If a particular stimulus word we may take the ratio of the sum of the
never appears as a response to the other overlapping frequencies to any pair of
stimuli, it will not appear as a common stimuli to the maximum possible sum.
response. Such instances are rare, Thus, if we turn to the first two col-
however, when the collection of stimuli umns in Table 1, which are the responses
is made on some basis that predicts re- to MOTH and INSECT respectively,
lated associative meaning. we see that the response frequencies in
The second problem is to find an common sum to 12. The total possible
appropriate measure of the relation be- frequency in common is 100; this yields
tween the responses common to differ- a relative common frequency of .12.
ent stimuli. This is the problem of This number is entered into the ap-
finding a measure of the relations be- propriate cell in Table 2.
tween the columns in Table 1. The A simple interpretation of such a
most obvious solution is to use product- measure demands some assumptions
moment correlations as the measure of which are not strictly correct. In other
relations between columns. There are, words, for most problems, this measure
however, a number of objections to can be regarded as an approximation.
such a solution. It supposes that the distribution of re-
The most important objection, de- sponses taken one per individual per
termined in part by the fact that the stimulus word from a homogenous
technique of free association limits population is characteristic of the dis-
each subject to one response, is that tribution of associations at different
the product-moment correlation would times in any one representative indi-
yield a measure of the correlation of vidual from that population. Thus, the
distributions of associations rather than assumption is made that the distribution
the extent to which frequencies are in of responses obtained from a sample of
common. Often, responses that are individuals is the same as the dis-
high in frequency to one stimulus will tribution of associations in any one
be low in frequency to another. This individual.
happens very often because of the This is an assumption shared by most
typically steep rank-frequency distribu- students of free associations in popula-
tion of free association responses to any tion samples (and one only occasionally
given stimulus. Because the forms of explicitly stated). An experimental
166 JAMES DEESE

comparison of repeated testing vs. counting instances of commonness of

single testing is a badly needed addition association.
to the association literature, and it may This measure of the relation in asso-
well lead to a more satisfactory way of ciative meaning between two words can

TABLE 1
RAW FREQUENCIES OF ASSOCIATES IN COMMON TO 19 WORDS
BASED ON RESPONSES OF 50 SUBJECTS

Stimulus words8
Responses
1 2 3 4 s 6 7 8 9 10 11 12 13 14 15 16 17 18 19

1 Moth 50 2 1 8 7
2 Insect 1 50 3 3 6
3 Wings 2 50 4 5
4 Bird 25 50 4 1 2 9 2 4
5 Fly 10 9 12 15 50 2 1 1 4
6 Yellow 50 2 1 4 3
7 Flower 2 50 1 2 1 10 1 2 2
8 Bug 24 4 50 5 1
9 Cocoon 50 2
10 Color 5 50 6
11 Blue 1 1 2 2 8 50 1 40
12 Bees 1 2 2 50
13 Summer 2 1 1 50 1 1
14 Sunshine 1 50 12
15 Garden 6 50
16 Sky 1 6 1 50
17 Nature 50 1
18 Spring 3 50
19 Butterfly 1 8 50
20 Light 4 1 4 1 1
21 Pretty 3 2
22 Ant 3 1 5
23 Bright 1 4
24 Airplane 4 1
25 Feather 2 3
26 Flight 1 2
27 Tree 2 1 1 6
28 Plane 2 5
29 Red 6 1 16 13
30 White 1 5 2
31 Greew 5 2 4 1 3 2
32 Sun 2
33 Seei/e 1 1
34 S£«fer 1 1
35 Gold 1 1 1
36 5facA 1 1 8 2
37 Winter 17 4
38 Warm 3 8 1
39 Pfarat 2 5 1
40 Gray 1 1 2
41 Brown 1 1
42 Vacation 2 1

• The numbers of the stimulus words correspond to the first 19 response words.
 ON THE STRUCTURE OF ASSOCIATIVE MEANING 167
TABLE 2
OVERLAP COEFFICIENTS FOR COMMON ASSOCIATES BETWEEN THE 19 WORDS IN TABLE 1
(Decimals Omitted)

Stimulus words
Stimulus
words
1 2 3 4 S 6 7. 8 9 10 11 12 13 14 15 16 17 18 19
1 MOTH 100 12 12 12 11 02 00 05 11 00 00 02 02 05 01 01 01 01 15
2 INSECT too 09 09 17 01 01 33 10 01 01 03 00 00 00 00 01 00 12
3 WIN .3 100 44 19 00 00 03 02 00 00 10 00 00 00 00 03 00 13
4 BIRD 100 21 01 00 03 02 01 01 10 00 01 00 01 05 00 12
5 FLY 100 01 01 08 06 01 02 Ofi 00 03 00 02 04 00 11
6 YELLOW 100 07 00 00 17 23 02 02 07 05 02 04 03 05
7 FLOWER 100 02 00 03 07 02 01 06 18 02 06 02 06
8 BUG 100 07 00 00 05 00 00 00 00 02 00 04
9 COCOON 100 00 00 04 01 01 01 00 02 00 22
10 COLOR 100 32 00 00 02 00 08 00 00 00
11 BLUE 100 01 02 04 04 46 03 02 02
12 BEES 100 01 02 03 00 04 02 07
13 SUMMER 100 05 02 00 01 10 00
14 SUNSHINE 100 02 03 02 15 04
IS GARDEN 100 00 04 02 02
16 SKY 100 00 01 00
17 NATURE 100 02 03
18 SPRING 100 02
19 BUTTERFLY 100

vary from .00 to 1.00. The latter value of associative meaning, the proper value
would happen, however, only if the two to enter into these cells is 1.0, since any
stimuli produce completely reciprocal word always has the same distribution
responding. Thus, complete identity in of response frequencies as itself. This
associative meaning can occur by this is like the assumption introduced by
measure only if two stimulus words Osgood and Suci (1952) into their
always elicit each other and no other factorial portrayal of the semantic
words. This is a consequence of the differential.
assumption of the frequency of the rep-
resentational response. Thus, if a pair ANALYSIS OF ASSOCIATIVE MEANING
of stimuli had distributions of responses Before examining some of the impli-
exactly alike except that they never cations of this notion of associative
elicited each other, the maximum rela- meaning, it would be helpful to look at
tive frequency in common would be .5. some sample data showing some of the
Such a result appears at first to be a relations that can exist between the
very undesirable property of the meas- associative meanings of different words.
ure of relation in associative meaning. It would be possible, of course, to
However, it turns out, as will be apply the present analysis to random
pointed out later, that this property is collections of words, but because the
extremely important in generating some overlap in associative meaning between
useful consequences of the concept of words taken at random is not high,
associative meaning. such an application would not be re-
The relative overlap coefficients for warding at this stage. In general, an
all pairs of stimuli within a set produces analysis of associative meaning would
a matrix of the sort found in Table 2. be conducted on a set of stimulus words
The only problem remaining is to fill selected because we would be interested
in the principal diagonal, or the com- in discovering their relationships and
munalities in an ordinary table of corre- because we would suspect, from prior
lations. In keeping with the assump- information, that they were associa-
tions made above about the assessment tively related.
168 JAMES DEESE
The data presented here illustrating sample data is on the factor presenta-
some of the possible relations in asso- tion, a word needs to be said about
ciative meaning are not psychologically factor analysis in this context. Factors,
interesting, with one exception. They of course, depend upon the variables
were, however, selected in such a way used to obtain them. In testing theory
as to insure some degree of related this is a sometimes overlooked restric-
associative meaning; thus they illus- tion if for no other reason than that test
trate fairly well some of the analytic theorists are often convinced that the
possibilities. batteries they factor are in some sense
The sets of stimulus words in Tables "representative" of all the set of abili-
3, 4, and 5 have internal structure for ties they wish to factor. In the analysis
the simple reason that the stimulus of words, however, this is not the case,
words themselves are all responses to so we are forced to an awareness of the
one of the stimulus words by the limitations of factor extraction. The
Minnesota norms (Russell & Jenkins, words that go into any matrix are only
1954). For example, the words in a small part of those that might possibly
Table 3 include BUTTERFLY and words have been used. Thus, the results of
which are responses to BUTTERFLY in the factor analysis are properties of the
the Minnesota norms. limited set, and we cannot talk about
These data are based upon samples the factor loadings for any one word,
of 50 subjects each, which are a bare for example, as if these exhausted all of
minimum for any kind of reliable an- the associative structure of that word.
alysis and too small for anything but As it turns out, this is a very useful
the roughest kind of prediction. They property of the analysis of associative
serve, however, to illustrate how sets of relations among words, but it tends to
words are related to one another and make trivial collections of words, such
what the factor structure of associations as those in Tables 3, 4, and 5 less inter-
is like. The subjects from which these esting in a factor analysis than would
associations were obtained are under- be collections of words with some
graduate students (male) at the Johns suspected important structure to be
Hopkins University. uncovered.
The basic data from which the fac- The first thing to notice about the
tors in Table 3 are derived are in Table factor loadings is the convenient simple
2. These are relative common frequen- structure in the factors. In Table 3,
cies of responses to 19 words in the approximately half of the words have
BUTTERFLY set. These relative fre- nearly zero loadings on Factor I, while
quencies were factored by the centroid the other half of the words have posi-
method, and after a very few rotations tive loadings of about the same magni-
(pairs of factors were rotated through tude. Those words on Factor I that
approximately 45 degrees) the factor have approximately zero loadings have
loadings presented in Table 3 were ob- positive loadings on Factor II, and the
tained. The factor loadings in Table 4, words with positive loadings on I have
based upon stimulus words obtained nearly zero loadings on II. Thus, the
from MUSIC, and Table 5, based upon first two factors effectively sort the
stimulus words obtained from SLOW stimulus words into two classes.
were likewise arrived at by centroid This kind of initial classification is a
analysis and very simple rotation. property of the kind of distribution of
Since the major emphasis in these common frequencies found in Table 2,
 ON THE STRUCTURE OF ASSOCIATIVE MEANING 169

and since this distribution is typical of impose a logical structure on the asso-
all of the 15 or so matrices thus far ciations, and one thing that is clear is
examined by the author, it will probably that associative meaning is not logical.
occur in all centroid analyses of words. Nevertheless, it is clear that the first
The same initial classification occurs two factors have split the stimulus
in Tables 4 and 5, and it also occurs in words into the most obvious and basic
Table 6 for which the stimulus words structure organizing them.
were not selected for associative rela- Factor III in Table 3 shows zero
tions but because they centered about a loadings on the nonanimate words
particular attitude or value system. again (except for NATURE, which ap-
An inspection of the loadings of Fac- parently comes close to straddling the
tor I in Table 3 shows that it occurs in basic division of words in Table 3),
words having to do with animate crea- and it splits the animate words into a
tion (BEES, FLY, BUG, WING, BIRD, etc.) bipolar factor. The positive loadings
whereas the words with loadings on are on WING, BIRDS, BEES, FLY and the
Factor II are those words in the set negative loadings on BUG, COCOON,
that do not have to do with animate MOTH, BUTTERFLY (which probably oc-
things (SKY, YELLOW, NATURE, etc.). curs here because of its strong relation
It would be a mistake, however, to fall to BUG and COCOON) . Factor IV makes
into the practice of naming factors de- a bipolar split of the nonanimate words.
rived from associative distributions, for SUMMER, SUNSHINE, GARDEN, FLOWER,
inevitably we fall into the attempt to and SPRING go together while BLUE,
SKY, YELLOW, and COLOR go together.
TABLE 3 It is instructive to follow the factor
ROTATED CENTROID FACTOR LOADINGS OF profiles for pairs of words within a set.
STIMULUS OVERLAP COEFFICIENTS This allows us one of several ways of
PRESENTED IN TABLE 2 preparing an associative thesaurus, in
(Decimals Omitted)
this case a pair thesaurus. Thus, we
Factors may say that BLUE and YELLOW are
Words both alike within this set in that they
I II III IV V VI share common loadings with SKY, SUM-
MER, SUNSHINE, COLOR, etc. On one
Moth 44 03 -27 -01 -03 -32
Insect SO 01 -33 01 -34 11
factor, however, they diverge; this pro-
Wing 52 01 45 01 29 -07 vides the critical difference between
Bird 52 02 46 01 29 -07 BLUE and YELLOW within this set of
Fly 48 03 32 01 -28 -03 words. BLUE goes with SKY, BUTTER-
Yellow 01 44 -03 34 -32 -02 FLY, WING, and BIRD, while YELLOW
Flower 01 39 -03 -32 03 44
Bug 41 01 -34 00 -14 37 goes with INSECT, FLY, BUG, and BEES.
Cocoon 40 01 -35 00 25 02 This, of course, does not exhaust the
Color -02 42 -04 44 04 -04 contrasts in associative meaning be-
Blue -02 57 -04 52 23 -04 tween BLUE and YELLOW, but it does
Bees 36 04 34 -02 -30 00
Summer -01 31 -03 -34 -02 -34 suggest one way in which they differ.
Sunshine 02 37 -04 -33 -03 -35 Table 4 presents the first six factors
Garden 00 35 -02 -34 -03 44 from a centroid analysis of the matrix
Sky -01 41 -03 43 38 -07 of stimulus words which are themselves
Nature 04 31 29 -02 01 34
Spring -01 35 -03 -37 -02 -36 responses to MUSIC. Again, the first
Butterfly 48 06 -29 -01 26 01 two factors yield a basic split. Factor I
has loadings on TONE, INSTRUMENT,
170 JAMES DEESE

TABLE 4 an overlap index of .71 for the Hopkins

ROTATED CENTROID FACTOR LOADINGS OF population, the highest for any of the
STIMULUS OVERLAP COEFFICIENTS FOR words in Tables 3, 4, and 5. For the
17 WORDS DERIVED FROM MUSIC six factors extracted, the factor profiles
(Decimals Omitted) for these two words are practically
identical. This means that there is not
Factors enough contrast provided within the
Words
I II III IV V VI
matrix of words to separate these words
within the arbitrary limit of six factors.
Tone 19 43 -03 40 02 53 This is in contrast with the pair SOFT-
Instrument 43 -01 33 01 -31 01 LOUD, which separates almost immedi-
Symphony 49 06 34 01 43 00
Sing 41 -01 -53 02 -36 -01 ately in the MUSIC table. The presence
Note 43 01 -33 03 37 -01 of NOISE, HEAR, HORN, etc., managed to
Song 55 -01 -47 00 -31 01 pull SOFT and LOUD apart by providing
Sound 04 59 00 -26 41 07 some associative meaning overlapping
Piano 52 02 37 01 -24 00
Noise 06 59 01 06 02 31 with LOUD but not with SOFT.
Band 51 08 38 -04 -24 -03 Because of the close relation be-
Horn 37 -09 00 -32 00 -43 tween SLOW and FAST, it may be very
Loud 00 48 -01 -42 -01 -29 difficult to provide sets which will con-
Hear 06 60 -01 49 01 -18
Opera 51 03 -33 01 32 00 trast these in associative meaning. They
Ear 00 41 -01 52 -02 -36 are, as are most of the high frequency
Music 58 08 25 01 37 02 opposites, adjectives, and they are de-
Soft 07 23 -02 -01 00 07 scriptive of a common state modified
only in a kind of secondary way. That

SYMPHONY, SING, NOTE, SONG, PIANO, TABLE 5

BAND, HORN, OPERA, and MUSIC, while ROTATED CENTROID LOADING OF STIMULUS
Factor II has loadings on SOUND, OVERLAP COEFFICIENTS OF 16 WORDS
NOISE, LOUD, EAR, and HEAR (this is a DERIVED FROM SLOW
less clean split than the previous table (Decimals Omitted)
because of the small loading of TONE
on Factor I). As with the previous Factors
table, Factors III through VI are bi- Words
I II in IV V VI
polar. Again, interesting contrasts may
be seen in factor profiles. OPERA and Slow 67 00 39 -01 20 -01
SYMPHONY, for example, have parallel Walk 41 02 -55 00 -38 03
profiles except on Factor III where they Speed 49 09 -33 07 47 -04
Quick 61 00 25 -02 -07 -28
are bipolar opposites. Here, OPERA Lazy 04 35 05 -38 -05 -36
goes with SONG, NOTE, and SING, while Drive 10 39 -33 02 00 04
SYMPHONY goes with INSTRUMENT, Skid OS 40 -01 36 35 -01
PIANO, BAND, and MUSIC. Run 54 01 -52 -01 31 -01
Work 03 33 36 -01 05 -36
Table 5 exhibits exactly the same sort Fast 73 01 38 -02 20 -02
of pattern with the words centering Down 01 31 -01 -36 -07 -37
around SLOW and so will not be Stop 01 37 05 37 -09 36
commented on further. This table Snail 43 02 37 02 -10 33
Sign 00 36 07 07 02 -44
illustrates an effect to be found in asso- Poke 02 34 05 -37 -05 35
ciations in adults, the close relations of Traffic 04 36 06 33 -04 -36
opposites. The pair SLOW-FAST yield
 ON THE STRUCTURE OF ASSOCIATIVE MEANING 171
is to say, both words imply motion; Completed data are available at pres-
they both describe the rapidity of mo- ent only on high scorers (the distribu-
tion, and they contrast only in the kind tion of scores in the Hopkins population
of rapidity. The way in which SOFT is skewed). Thus, the factors in Table
and LOUD are alike, however, provides 6 are based upon SO Hopkins under-
only part of the distribution for SOFT, graduates who scored at or above one
since it exists also in a close relation probable error in the published norms
with the word HARD. HARD and LOUD for men on religious value. The first
are related through SOFT but otherwise thing to notice about Table 6 is that it
independent. is very much the same as Tables 3, 4,
Table 6 illustrates the factor analysis and 5. The initial classification in this
of a set of stimulus words which are case sorts the words into a group in-
psychologically more interesting. These cluding DIVINE, SPIRIT, PRAYER, WOR-
words are taken from a larger study, to SHIP, REVERENCE, RELIGION, FAITH,
be presented elsewhere, which is con- and DEVOTION and into a group includ-
cerned with the factor structure of asso- ing HOPE, SOUL, LOVE, INSPIRE, SERVICE,
ciative meaning of critical words for SERMON, and CLERGYMAN. One is al-
subjects scoring high and subjects scor- most tempted to say that the classifica-
ing low in each of the value scales from tion divides the words into an institu-
the Allport, Vernon, and Lindzey tional, theological class and a human
(1960) study of values inventory. The individual class.
critical words for each value scale were
obtained from the actual items scored IMPLICATIONS OF ASSOCIATIVE
for that value scale; thus the words in MEANING
Table 6 are from the Religious Value By now, some readers will have
items. recognized the notions presented here
TABLE 6 as extensions of those stated by Bous-
ROTATED CENTROID LOADINGS OF STIMULUS field, Cohen, and Whitmarsh (1958),
OVERLAP COEFFICIENTS FOR IS WORDS and since these authors regard their
CENTERING IN RELIGIOUS VALUE work as an extension of Noble's
(1952), by implication the present
Factors work is also.
Words Thus far we have asserted that the
I II in IV V VI distribution of responses to any free
Sermon 03 43 38 -10 04 33 association stimulus 4 forms the associ-
Clergyman 11 42 05 -47 00 26 ative meaning of the stimulus word and
Religion 54 03 -02 -34 -03 -36 thus an associative concept named by
Service -01 40 -02 -43 37 -16 that word. If the associative meaning
Worship 57 04 33 -05 -30 -02
Reverence 55 04 36 02 27 -03 of a word (as opposed to any particu-
Prayer 54 03 34 -03 -01 -34 lar stimulus-response pair) is to have
Soul 06 38 -02 34 03 39 any use, it must predict something
Spirit 39 00 -37 05 39 05 about verbal behavior. In this section
Divine 50 00 05 35 34 04
52 08 -36 -04 -35 4
Faith 01 Free association stimuli are not limited
Inspire 03 40 -01 33 -01 -41 to words, of course, and the ideas presented
Devotion 49 10 -33 02 00 34 here can be in theory, extended to nonverbal
Love 03 38 00 35 36 05 stimuli, but the lack of a clear criterion for
Hope 06 37 -41 -01 -04 -45 the representational response may produce
difficulties in practice.
172 JAMES DEESE

we will discuss some of the character- we are unlikely to say "this is a rapid
istics of verbal behavior that may be train" or "this is a sluggish train."
predictable from the associative mean- The second kind of environmental
ings of words and from collections of relation (that of coordination) is illus-
associative meanings. trated by the SKY-BLUE pair. Given the
Associative meaning will not predict frame, "he into the
the tendency of words to elicit one an- sky," a highly likely substitution in the
other ; that is predicted by the free as- second blank is "blue." All words in
sociation frequencies of particular pairs English seem to be able to produce both
or by some general measure of pair re- kinds of environmental relation, but for
lation (such as the interword associa- many words one or the other type
tive index). Associative meaning, in predominates. Thus, while "the blue"
general, should predict the words that may appear as a substitution for "sky,"
will occur in the verbal environment of such useage is poetic and rare.
a particular word. This statement en- One problem is produced by the ap-
tails the assumption that words are parent assymetry of substitution in par-
used, subject to certain other con- ticular frames. It is possible, indeed
straints, in particular environments be- probable, that the frame, "he
cause of the distribution of associations into the blue ," will not pro-
they possess. Thus, the distribution of duce "sky" with the same frequency
associates takes on the character of that "blue" is produced in its comple-
mediation determining the use of par- mentary frame. Particular test frames
ticular words. have an influence, and it is possible
Thus, if a particular word appears in that substitution frequencies may be
some verbal environment, a word close symmetrical when averaged over a rep-
to it in associative meaning should ap- resentative group of test frames. Such
pear in the same environment. Gener- symmetry is a property of the definition
ally, the closer the relation, the higher of associative meaning, though the
the probability of the two words appear- property cannot be properly tested,
ing in the same environment. The since any failure of symmetry may
words may appear in the same environ- easily be attributed to a lack of ade-
ment in two ways: (a) as substitutes for quate sampling.
one another, or (b) as part of one an- Earlier it was pointed out that one of
other's environment. It is not a part of the consequences of the assumption of
associative meaning to predict which of the representational response was that
these two ways will appear; though perfect agreement in associative mean-
such a prediction should arise out of ing between two words would occur
theory about the origins of associations. only if they were perfectly reciprocal,
The first kind of environmental rela- that is to say, if they were the only
tion (that of substitution) is illustrated associates of each other. While this
quite clearly by the highly related seemed at first to be an undesirable
SLOW-FAST pair. The environment that property, it does conform with the
produces Slow, all other things equal, limiting case of substitution. For, if
is very likely to produce Fast also, but two words are perfectly correlated in
not nearly so likely to produce diction- associative meaning, then they ought to
ary or thesaurus equivalents of Fast. be the only substitutes for one another
Thus, we are likely to say "this is a in the substitution relation. Notice
slow train" or "this is a fast train," but that this situation would not theoreti-
 ON THE STRUCTURE OF ASSOCIATIVE MEANING 173

cally prevent these words from being which by psychoanalytic theory, for
part of the associative meaning of a example, are supposed to symbolize the
third word. Thus Word A could elicit phallus, the female figure, hostility, etc.,
only B as a free associate and B elicit ought to have their common symbolic
A, which would make them overlap character revealed in the interrelations
completely in associative meaning, but and factor structure of their free asso-
both A and B could be free association ciates. This is not to say that phallic
responses to a third Word C. Obvi- symbol words ought to reveal a com-
ously this situation would not happen mon core of phallic meaning (by dic-
in any real language, but an approxi- tionary criteria) but that they ought to
mation to it frequently occurs. The converge on some common factor. In a
general rule for such a case is that the word, no matter how remote their dic-
reciprocal pair of words are both of tionary meanings and references in the
high frequency of usage, while the third real world, they should reveal a com-
word, which elicits them both but never mon associative meaning. If they do
occurs as a response to either, is a not, it is hard to see how the particu-
word of very low frequency of usage. lar psychoanalytic symbol investigated
An example in the Hopkins data is could have any general validity for the
the overlapped pair WOMAN and GIRL, population tested.
neither of which elicit Virginal, but Currently, the present author is
both of which are elicited as responses studying the structure of associative
by VIRGINAL. meaning within a set of words that
The use of test frames to validate the were selected from Rorschach protocols
notion of associative meaning will be as indicators of aggression. No de-
difficult in practice, since the influence tailed account of these data can be
of grammatical constraints, etc., needs given at present, but a striking qualita-
to be removed. Furthermore, we do tive aspect of them is the extent to
not know enough about the problem of which the overlap in associative mean-
compounding to know whether or not ing among them is mediated by a small
compound effects can be derived from number of response words, many of
the concept of associative meaning it- which appear out of context to be rela-
self. Nevertheless, whatever other in- tively neutral (Stick, for example).
teresting effects may be achieved, the The technique may be used also to
ultimate validation of the concept of study the verbal organization associated
associative meaning must come from its with attitudes. Table 6 presents the
ability to predict substitutions in the factor structure of associations to words
flow of language. centering about religious values for in-
The most important use of the notion dividuals who score high in religious
of associative meaning is to study the value. A complete set of data is not
structure of sets of words having intrin- yet available for low scorers, but it is
sic psychological interest. The number almost certain that the most important
of such sets is very large, and indeed is difference between the high and low
only limited by the ideas psychologists scorers will be in the extent to which
have about the influence of psychologi- the associations to these words are well
cal variables upon the use of words. organized. The factor loadings for low
An obvious application is to any col- scorers will almost certainly be lower;
lection of words as metaphorical sym- indeed, many of the stimulus words
bols. Those words (or concepts) (SERVICE, LOVE, HOPE, SPIRIT) will all
174 JAMES DEESE
but disappear from the factor structure, like BIRDS and flutter in the BLUE SKY.
and it is very likely that the factor Thus, another associative category in-
structure of the remaining words will trudes on the meaning of BUTTERLLY
be very different. and makes it an associative concept not
quite the same as any other. Contrary
ORIGIN OF ASSOCIATIONS AND to zoology, associative BUTTERFLIES are
SUMMARY as closely related to the BIRDS as to the
MOTHS.
The last question we face concerns Large collections of words ought to
the origin of associations. The view enable us to discover the categories of
of Locke that attributed them to the association, if they exist, and how they
accidental contingencies of nature has change with age, if they do. In any
led to a kind of mosaic theory of mind, event, an alternative to the contiguity-
a theory that has scarcely ever had a similarity tradition in association theory
serious rival on its own ground, though can come about only by a study of the
it has often been challenged. associations themselves, and the pur^
No real substitute for the contiguity- pose of this paper is to present a
similarity postulate is offered here. The somewhat different approach to the
highly organized economy of associa- study of such associations. That ap-
tive meaning has impressed the author, proach consists in the assumption that
however, and it was a belief that the the distribution of associates to any
human mind derived associations from word provides the associative meaning
categories of its own that sent him on of that word and the techniques neces-
the search for a technique by which to sary to apply this assumption to data.
study associative meaning. Thus, the
least that can be offered is the sug- REFERENCES
gestion that associations derive in whole ALLPORT, G. W., VERNON, P. E., & LINDZEY,
or part from the structures or cate- G. A study of value. (3rd ed.) Boston:
gories of the human mind. This belief Houghton Mifflin, 1960.
is probably the mover of attempts, such BOUSFIELD, W. A. The occurrence of clus-
tering in the recall of randomly arranged
as Woodworth's, to classify associa- associates. J. gen. Psycho}., 1953, 49, 229-
tions. Such attempts are fruitless, 240.
however, for if there are categories BOUSFIELD, W. A., COHEN, B. H., & WHIT-
in association, they are categories of MARSH, G. A. Verbal generalization: A
theoretical rationale and an experimental
association, not categories of subordi- technique. Technical Report No. 23, 1958,
nation, coordination, etc. This is ad- Contract NONr onr 631(00).
mittedly the belief that motivated the DEESE, J. Influence of inter-item associative
present work. strength upon immediate free recall. Psy-
chol. Rep., 1959, 5, 305-312. (a)
Thus, the implicit distributions of re- DEESE, J. On the prediction of occurrence of
sponses that define associative meaning particular verbal intrusions in immediate
may exhibit patterns of overlap because recall. J. exp. PsychoL, 1959, 58,17-22. (b)
they are derived from simple structures. DEESE, J. Frequency of usage and number of
words in free recall: The role of associa-
An adult may use the word BUTTERFLY tion. Psychol. Rep., 1960, 7, 337-344.
to correspond to a class of living crea- JENKINS, J. J., & RUSSELL, W. A. Associa-
tures that fly and are bugs. This tive clustering during recall. /. abnorm.
doesn't however, exhaust the associa- soc. Psychol., 1952, 47, 818-821.
tive meaning of BUTTERFLY, for most JENKINS, PATRICIA M., & COFER, C. N. An
exploratory study of discrete free associa-
flying bugs are dirty, unpleasant and tions to compound verbal stimuli. Psychol.
YELLOW, while BUTTERFLIES are pretty, Rep., 1957, 3, 599-602.
 ON THE STRUCTURE OF ASSOCIATIVE MEANING 175
KARWOSKI, T. F., & BERTH OLD, F. Psycho- RUSSELL, W. A., & JENKINS, J. J. The com-
logical studies in semantics: II. Reliability plete Minnesota norms for response to 100
of free association tests. 7. soc. Psychol., words from the Kent-Rosanoff Association
1945, 22, 87-102. Test. Technical Report No. 11, 1954, Uni-
NOBLE, C. E. An analysis of meaning. Psy- versity of Minnesota, Contract N8 onr
chol. Rev., 1952, 59, 421-430. 66216. '
OSGOOD, C. E., & Suci, G. J. A measure of WOODWORTH, R. S. Experimental psychol-
relation determined by both mean differ- ogy. New York: Holt, 1938.
ence and profile information. Psychol.
Bull, 1952, 49, 251-262. (Received October 6, 1960)