Review

Author(s): Moshe Anisfeld

Review by: Moshe Anisfeld
Source: The American Journal of Psychology, Vol. 80, No. 4 (Dec., 1967), pp. 646-649
Published by: University of Illinois Press
Stable URL: http://www.jstor.org/stable/1421205
Accessed: 18-02-2016 16:15 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact [email protected].

University of Illinois Press is collaborating with JSTOR to digitize, preserve and extend access to The American Journal of
Psychology.

http://www.jstor.org

This content downloaded from 165.123.34.86 on Thu, 18 Feb 2016 16:15:26 UTC
All use subject to JSTOR Terms and Conditions

BOOK REVIEWS
Edited by T. A.

RYAN,

Cornell University

The Structure of Associations in Language and Thought. By JAMESDEESE. Baltimore, Md., The Johns Hopkins Press, 1965. Pp. xiii, 216. $6.50.
The aims of this book differ from those of two other major approaches to the study
of free word associations. One line of research, popular in the early days of empirical
study in this area, was concerned with a characterization of the relation between
stimulus and response, in the hope (largely unfulfilled) that an ad hoc classification
of the Ss' responses in relation to the stimuli that occasioned them would reveal the
mental processes that gave rise to these responses. In contemporary psychology, the
attempt to explain free association responses has been replaced by the use of free
association data to explain other phenomena in the verbal learning laboratory.
Deese's goal is more ambitious than either of these two. He uses the free association technique to uncover psychologically significant relations in the (English) lexicon. The responses obtained in the test are not the focus of primary attention but,
rather, serve merely as instruments for manifesting inherent relations among the
stimuli. Such an objective, ably pursued by one of the leaders in the field, deserves
the serious scrutiny of persons interested in the psychology of language and especially
in its semantic aspects.
How can free association results shed light on the nature of semantic relations in
the language? Deese's answer is essentially that this can be achieved through a proper
analysis of the free association data. Giving a set of stimulus-words to a group of
Ss and asking them to respond to each with the first word other than the stimulus
that comes to their minds, one obtains a frequency-distribution of responses. Such
distributions can be analyzed in a variety of ways, yielding ten or more different, but
closely related, indices of word relatedness. Deese's measure, termed index of commonality (IC), bears a close affinity to these measures. For each pair of stimuluswords, it records the proportion of the number of actual common responses to the
maximum number of possible common responses. In calculating the IC, it is assumed
that a stimulus-word automatically elicits itself as a "representational response."
Thus, although only a single response is obtained overtly, two responses are counted.
As an example, consider the stimuli MOTH and INSECT. Of the 50 Ss who responded to these words (among others), 9 gave fly to both words, 2 gave moth to
INSECT and 1 gave insect to MOTH. Counting the representational responses (50
moth to MOTH and 50 insect to INSECT), the number of common responses is 12.
The maximum could have been 100, yielding a ratio of 0.12. IC values are calculated
in this way between each stimulus and each other stimulus in a given set. IC matrices
are then subjected to a factor analysis to extract common dimensions. Each factor
identifies a group of words which belong together on the basis of their IC values.
The loadings of every word on all factors fully describe the relations of this word
with all other words in the matrix. The common variance of each word is thus
broken down into factorial components.
The conception of words as composed of smaller entities is not new. It can be
traced back at least to John Locke's 17th century analysis of complex ideas into their
646

This content downloaded from 165.123.34.86 on Thu, 18 Feb 2016 16:15:26 UTC
All use subject to JSTOR Terms and Conditions

BOOK REVIEWS

647

constituent simple ideas. Contemporary linguists of different theoretical persuasions

and interests commonly analyze words into their components or features. For instance, boy in the feature analysis is described, in part, as a noun with the specification of common, count, animate, and human; and father in the componential analysis
is characterized as lineal kin, male, and one generation above ego. Psychologists working with the semantic differential also describe words by placing them in reference
to three general factorial axes: evaluation, activity, and potency. Deese's factorial
analysis can be characterized in the following way in relation to these conceptions.
(a) The present classification of each word is empirical, as in the case of the semantic
differential, rather than logical or linguistic, as in the feature and componential
analyses; (b) The classification is continuous rather than discrete, i.e. each word
has different degrees of saturation on each factor; (c) The factors describing a particular set of words are unique to that set and bear no relation to factors obtained for
other sets. In this respect, Deese's conception resembles more closely that of Locke
than the modern approaches mentioned. Deese considers the uniqueness of his factors
as somehow contradictory to the general factors repeatedly found in the work with
the semantic differential and superior to them. It seems obvious, however, that words
can and should be described by reference to both general and specific factors. The
two descriptions complement each other. The description of MUSIC as highly positive on the evaluative factor is perfectly consonant with its description as having a
high loading on a specific Factor I.
For a closer evaluation of Deese's analysis, we will turn now to an examination
of its face validity, predictive validity, and construct validity. Some factorial structures
make good intuitive sense. For instance, in a study of 17 words preselected for their
relation to MUSIC (Pp. 76-78), Factor I showed relatively high loadings for instrument, symphony, sing, note, song, piano, band, horn, opera, and music, and Factor
II high loadings for tone, sound, noise, loud, hear, ear, and soft. It seems that words
were split into a "musical" cluster (Factor I) and a "nonmusical" cluster (Factor
II). Some of the analyses yield classifications which the reader is willing to accept
as sensible and revealing of lexical relations, although he would not have made
such classifications a priori. The conclusions one can draw from these results about
the general validity of this method for studying semantic relations are, however,
limited by two considerations. First, these results were obtained for sets of words
carefully preselected for their interrelations. An analysis of a random selection of
words is more likely than not to produce a chaotic factorial structure. Secondly, even
in cases of preselected sets, as are all studies reported in the book, some factorial
structures do not separate the words into intuitively acceptable groupings. Face
validity can thus be a double-edged sword and Deese is understandably reluctant to
depend too heavily on it.
As support for the validity of his measures, Deese cites their ability to predict the
results of various verbal learning and behavior tasks. In the absence of systematic
comparative analyses, however, it is not possible to assess the advantage of the
IC measure and the factor analytic method associated with it over the other word
association indices. In fact, in the only comparison reported, a measure of direct
interword associations predicts free recall somewhat better than does IC. Thus, the
evidence from the verbal learning and behavior laboratory cited in this book, although
of considerable interest in itself, can only add support to the general conclusion that
free association data can be used to predict the outcome of a variety of verbal tasks.

This content downloaded from 165.123.34.86 on Thu, 18 Feb 2016 16:15:26 UTC
All use subject to JSTOR Terms and Conditions

648

BOOK REVIEWS

One may also seek the validity of Deese's analysis in its theoretical underpinningThe author feels that his analysis is grounded in the mediational notion of stimulusequivalence which asserts that two stimuli, A and B, are equivalent if they share
a common response, C. The equivalence in the model case is reflected in the greater
ease of learning the association B-D after having learned A-D. In that case, the
notion of equivalence is thus specific and fairly well defined. But what kind of
equivalence can be attributed to two words connected by a high IC value? And in
what sense is the equivalence reflected in one factor different from that reflected in
another factor? In various contexts, Deese interprets the idea of equivalence in terms
of parts of speech. But such equivalences are certainly not characteristic of the factorial structures where words belonging to different parts of speech turn up indiscriminately in the same factors, unless arbitrarily excluded in the selection of stimulus-words.
Such exclusions are made deliberately in the second part of the book (Chapters
5-8), which deals specifically with the relation between word associations and parts
of speech. Using the distinction between paradigmatic responses (same part of speech
as the stimuli) and syntagmatic responses (different part of speech), Deese found
the following distribution of paradigmatic responses: nouns, 79%; verbs, 52%;
adjectives, 50%; and adverbs, 27%. No new or specific proposals are made to account for the syntagmatic responses. Rather, the book concentrates on the analysis
of the paradigmatic responses to nouns and adjectives and bases on this analysis
two "new laws of association": a law of grouping for nouns and a law of contrast
for adjectives. One of the main empirical foundations for the principle of contrast
is a study of associations to 278 common adjectives (Pp. 121-128). Of these, Deese
isolated 39 pairs of antonyms, such as alone-together, happy-sad, hard-soft, and thickthin, which constitute "fundamental contrasts." These pairs have the characteristic of
reciprocal association, i.e. the two members in each pair tend to elicit one another
in a free association test. These pairs are considered fundamental because, it is
claimed, each contrast is unique and not reducible to more general factors. The factorial evidence for this claim is scanty and cannot be evaluated without comparable
analysis for other words. Moreover, the argument Deese advances that there are
sentence-frames in which one contrasting pair could fit well but a seemingly closely
related pair could not fit as well, can be made for every word in the language.
Every word is in some sense unique (it wouldn't exist if it weren't) and best suited
for the expression of a particular idea, but this shouldn't prevent it from sharing
common factors with other words.
Deese considers associative reciprocity as a kind of operational definition of
antonymy. This is rather surprising in view of the fact that of 100 Ss, the number
of those who responded reciprocally to any given pair never exceeded 40 and was
usually much smaller than that. Can one seriously believe that for the S who responds to WHITE with house rather than with black, the two adjectives are not
opposites? Or can it be claimed that inside-outside which have an average reciprocity
score of 40 are two times more opposite than clean-dirty, which have a reciprocity
score of only 18 ? One might as well accept the fact that no operational definition is
available at present, and may never be forthcoming, for antonymy as well as for
other important psychological concepts.
As was already indicated, the 39 antonymous pairs serve to establish the law
of contrast for adjectives. The law asserts that "Elements are associatively related

This content downloaded from 165.123.34.86 on Thu, 18 Feb 2016 16:15:26 UTC
All use subject to JSTOR Terms and Conditions

BOOK REVIEWS

649

when they may be contrastedin some uniqueand unambiguousway" (P. 165). One
may first ask: What is the lexical domainto which this law applies?It appearsthat
the English lexicon was searchedthrough appropriatelyand only the discussed39
pairs were found to be both contrastiveand reciprocallyassociated.At best, then,
the law of contrastis nothing more than a summaryof this finding.It appearsthat
the author himself is uncomfortablewith this law for he interpretsit in terms of
contextual equivalence:contrastselicit each other because they are used in the
same linguisticenvironments(e.g. P. 138). In one study (Pp. 139-140) designedto
supportthis interpretation,a single S produced15 sentencesin which he used the
words hot, cold, good, and bad. These adjectiveswere then deleted and a group of
100 Ss was asked to fill in appropriateadjectivesin the blank spaces.It was found
sentencesthan one
that the oppositewas more often substitutedin the reconstructed
of the other two words. For instance,in the sentenceswhere bad had originally
appeared,good was given 4.1% of the caseswhereashot was given only 1.9% and
cold 0.2%. Such studies, however,do not constitutesufficientevidencefor the contextual determinationhypothesis.To verify this hypothesis,one is under obligation
to show that an oppositeis morelikely to replaceits counterpartthanany otherword,
not just an arbitrarilyselected pair. More generallyand importantly,the sentences
an S gives are not usually reproductionsof previouslyexperiencedones, but are
new constructions.Therefore,one cannotassumethat the sentencesan S constructson
the requestof an experimenterrepresenta sample of sentencesexperiencedin the
past by the S or by other speakersin the same community.It is thus unjustified
to hold linguistic context responsiblefor free associativebehavior.It appearsthat
free associationsare not direct reproductionsof past experiencesbut ratherconstitute, as does most behavior,productiveevents relatedto the past in an indirectand
complex manner.At some points, Deese seems to voice a relatedviewpoint but it
is not clearhow it relatesto the contextualequivalencenotion.
The problemsof theoreticalstatus and empiricalverificationraised with respect
to the principleof contrastfor adjectivesapply also to the principleof groupingfor
nouns.In addition,it suffersfrom anotherdifficulty.Deese states (P. 165): "elements
are associativelyrelated when they may be groupedbecausethey can be described
by two or more characteristicsin common."It is apparent,however, that every
word sharessome featureswith a great many other words, and unless one specifies
what featuresare relevantunderwhat conditions,the principleof groupingremains
devoid of explanatoryand predictivevalue. Without such explicit specificationthe
principle of grouping makes little advanceover the early philosophicalnotion of
similarity.
Current investigatorsworking with free associationsgenerally exhibit interests
that go little beyond the empiricaldescriptionof functionalrelations.Against this
background,Deese's work stands out in its broad goals and dedicatedsearchfor
an understandingof the central and most complex problem areas in psychology,
languageand thinking. The very attemptto deal seriouslywith these problemsin
the verballearningframework,whateverits outcome,musthave a profoundinfluence
in focusing the attentionof investigatorsin this area on the ultimategoals towards
which their researchmust be directed.It would not be surprisingif a re-orientation
and clarificationof researchaims led to a gradualreplacementof the free association
task in favor of other more explicitlydirectedand constrainedtasks.
ANISFELD
MOSHE
CornellUniversity

This content downloaded from 165.123.34.86 on Thu, 18 Feb 2016 16:15:26 UTC
All use subject to JSTOR Terms and Conditions