Osgood, C. Meaning Cannot Be RM
Osgood, C. Meaning Cannot Be RM
THEORETICAL NOTES*
Meaning Cannot Be r ? m
CHARLES E. OSGOOD
In the January 1965 issue of this JOVRZCAL,Pro- But what psychologist has ever made such a claim ?
lessor Jerry Fodor presented a critique of mediation Skinner would agree, I am sure, that the overt oc-
learning theory titled "Could Meaning be an rm ?'' currence of a linguistic (or any other) response de-
He identified his adversary as Hobart Mowrer; pends upon the patterned context of total, momen-
however, the symbol r m is characteristic of my dis- tary stimulation, some elements of which may be
cussions of mediation theory, although the r t re- facilitative and others inhibitory with respect to the
ferred to in the body of the critique is Mowrer's response in question. Of course, the two-stage model
symbol (cf., his APA presidential address, 1954). makes the functional independence of linguistic re-
Perhaps this choice was dictated by the fact that sponses from particular environmental stimuli even
Mowrer's presentation was the first serious attempt more obvious.
in modern behaviorism to deal with the dynamics (2) Learning Theory Cannot Explain the Func-
of sentences. Fodor wrote the original draft of tional Equivalence o] Linguistic Responses. "Just
"Could Meaning be an rm?" during the year that as two physiologically distinct actions may both be
he was a visiting professor in the Institute of Com- instances of a bar-press response, so two phonem-
munications Research at Illinois and was sittir~g in ically distinct utterances may be functionally equiv-
on my graduate seminar in Psycholinguistics--if alent for a given speaker or in a given, language
one can imagine Fodor merely "sitting in" on any- (Fodor, 1965, p. 74)." The terms bachelor and un-
thing! In any case, the notion of representational married man are offered as an example of synonym-
mediation was and continues to be central to my ity. Conversely, responses having the same overt
theorizing about language behavior, and it is there- form may belong to different semantic classes, e.g.,
fore appropriate for me to write a reply to Fodor's homonyms such as light (in a light weight vs. a
critique. light color). If this is meant as a criticism of be-
havior theory's descriptive adequacy, then it is well
Fodor's Criticism o] S - R T h e o r y in General taken--we could well follow the linguist's lead in
searching for the central (mediating) events which
Fodor begins with a critique of single-stage learn- render physically different stimuli or responses func-
ing theory that essentially reiterates points made tionally equivalent. But, there is no inherent in-
earlier by Chomsky (1959) in a review of Skinner's adequacy in behavior theory here: two signs are
Verbal Behavior (1957). The three main points functionally equivalent if they regularly and re-
made need to be reconsidered briefly here because liably produce the same response (overt or media-
they form the framework within which two-stage, tional) and two responses are functionally equiva-
mediation theory will later be evaluated and they lent if they regularly and rar~tomly are produced
are not entirely valid. by the same stimulus (a situation referred to by
(1) Linguistic Responses are no~ Obviously De- linguists as "free variation"). Two words would
pendent upon Specifiable Stimuli. The probability have the same meaning, or two word strings be
of emitting the response book may increase with the paraphrases, if and only if they produced precisely
palpable presence of BOOK objects in the imme- the same simultaneous pattern of rm'S. However,
diate environment, but such presence is neither a this is a situation rarely found, and Fodor himself
necessary nor certainly a sufficient condition. Agreed. points out that the characterization of synonymity
is a problem "for which no solution is at present
* In order to permit discussion of Professor known (1965, p. 74)." Certainly the effect upon
Fodor's paper without undue de~.ay, these notes were one's unmarried daughter of telling her so-and-so is
given early, publication. Editor a bachelor is going to be a bit different than telling
402
ANALYSIS OF MEANING 403
her so-and-so is an unmarried man--the latter al- fleet my assumption that rm'S do retain the response-
most amounts to issuing a hunting license! like properties of the overt behaviors in which they
(3) Linguistic Responses do not Admit o] the presumably have their origin (RT) and that they
Usual Measures o] Response Strer~gth. Certainly it function in theory as events whose occurrence cart
is true that stress (amplitude of response) does not become dependent on observable stimuli (signs).
covary with the frequency of usage of word forms. Thus all of the conceptual baggage of single-stage
On the other hand, large samples of verbal behavior theory (notions like habit-strength, generalization,
do display a positive relation between frequency of inhibition, and so forth) can be utilized for [S] -> r m
response (usage) and such strength indices as ta- decoding relations. This implies nothing whatsoever
chistoscopic threshold and emission probability (the about the physical nature or locus of rms--whether
well-known "Spew Hypothesis"). But psychologists they are peripheral or, as I suspect for the adult
do not agree on what external indices represent the human, entirely central ("corticalized") events. It
strength of a response in general, to say nothing of is further assumed that, again functionally like other
verbal responses, and various measures of response responses which produce sensory feedback, rm'S will
strength do not correlate highly with each other, have their automatic stimulus consequences, sym-
even in the animal laboratory. Indeed, stress is a bolized by sm in the model. Thus the same concep-
variable in linguistic responses that is obviously de- tual baggage can be utilized for Srn "-> R x encoding
pendent upon central mediating processes, no mat- relations, where R x may or may not be linguistic.
ter how they may be ultimately characterized. In In other words, in two-stage theory representational
any case, it is not clear to me why this is an issue mediation processes have response functions in de-
for language per]ormance; it could be an issue for coding ("understanding") and stimulus functions in
language learning, but here again the functional in- encoding ("creating") language, regardless of what
dependence of decoding and encoding habits pro- may ultimately turn out to be their physical nature.
vided by a two-stage model reduces the rigidity of After granting that the mediation model does ap-
relations between overt S's and R's. pear to free behavior theory from the fatal flaw of
a single-stage model--namely, that responses to signs
Fodor's Specific Criticism o] Mediation must be the same as responses to their significates
Theory irt order for the signs to have meaning--Fodor (p.
77) concludes as follows: "Nevertheless, the theory
Before developing his critique of mediation theory, of meaning implicit in the quotations from Mowrer
Fodor correctly notes the following: (a) that r m is cited above is thoroughly unsatisfactory. Nor is it
]unctionally characterized in two-stage theory; (b) obvious that this inadequacy is specific to the ver-
that r m is not the same thing as R T (total behavior sion of the mediation theory that Mowrer espouses."
to the significate, S), but rather r m is a proper part His critique involves three major issues: (1) Media-
o] R T. These points are critical in the argument, tion theory can be shown to differ from single-stage
both Fodor's (1965) and mine here. It is precisely theory in only one respect--the observability of the
because rrn is part of a particular R T that it can, in responses in question--and hence is subject to the
theory, have its symbolic or representing property, same limitations. (2) Mediation theory, therefore,
and thus allow [S] (a sign) to "stand for" S, the along with single-stage theory, implies that occur-
particular significate which originally elicits R T, The rence of significate (S) or sign (IS]) will be a
basic model is shown here as Fig. 1; it is more sufficient condition for occurrence of their associated
explicit than those used by Mowrer and reproduced linguistic response ( R x ) - - a prediction obviously
by Fodor. contrary to fact. (3) Mediation theory in principle
provides no account of certain well-known charac-
S ~ RT teristics of sign learning and sign usage and is thus
f insufficient in any case. Certain criticisms of
f
f Mowrer's conception of the sentence as a condition-
1 t
J ing device, with which criticisms I generally agree
t
f
f (Osgood, 1963), are not pertinent to general r m
~] = rm'%---~'S m ~R x
theory and need not be discussed.
(1) r m Differs ]rom R T Only in Terms of Ob-
Fro. 1. General paradigm for the development servability. Fodor's argument proceeds as follows:
of representational mediation processes. Beyond stating that r m is a proper part of RT,
mediation theory requires an additional postulate--
As to the characterization of rm, the use of r namely, that each r m which unambiguously mediates
(rather than a more neutral x, let us say) does re- the meaning of a sign must bear a one-to-one,
404 OSGOOD
unique relation to its R T. The consequence of re- a one to one relation between r's and R's, we insure
jecting this postulate--i.e., of assuming that one and that the former lack the "surplus meaning" charac-
the same r m could be part of many I ~ produced teristic of terms designating bona fide theoretical
by many different S - - w o u l d be that signs would entities (Fodor, 1965, p. 81)." As will become clear,
necessarily refer ambiguously to many different sig- I must accept the one-to-one relation between r m
nificates. The same argument applies to ambiguous and RT, but not that between r m and R x.
signs (e.g., the homonymous form bat, as in baseball By way of reply, let me point out first that this
bat vs. winged bat); mediation theory tnust assume difference in observability is by no means trivial
distinctive rm'S based on different experiential his- and, indeed, is part of the "surplus meaning" Fodor
tories (RT1 vs. RT2) , the selection among them be- demands of theoretical entities. In the well-docu-
ing based on contextual stimulus conditions (about mented experiments on semantic generalization--
which more anon). when, for example, a novel finger-lifting response
With this I entirely agree. Both in 1957 (in Os- is trained to the word joy and transfers sponta-
good, Suei, and Tannenbaum), with the argument neously in the test situation to the word glee but not
restricted to three gross affective factors, and in to the physically more similar word boy--it is ob-
1963, the argument here being more generalized, I vious that the basis of the transfer is semantic
have proposed that rra is not a single reaction but similarity. But it is equally obvious that nothing
rather a multicomponential affair. In a fashion anal- remotely tL~ (total overt behavior originally occur-
ogous to the notion of a phoneme as a simultaneous ring in glee and joy using situations) occurs between
bundle of distinctive phonetic features (cf., :[akob- glee and the novel finger-lifting response---indeed, no
son and Halle, 1956), the meaning of a sign is con- particular mediating responses of an overt nature
ceived to be a simultaneous bundle of distinctive typically occur. Psychological theory in general, and
semantic features--which I identify with component psycholinguistic theory in particular, must be able
r's of the total r m. Whereas within a given compo- to incorporate such phenomena.
nent reaction system (factor, feature) only one al- Let me point out, second, that Fodor apparently
ternative can occur to a presented sign (e.g., r m misses the primary difference between two-stage
cannot be simultaneously q- and - - in the same com- and single-stage theories--the functional separation
ponent reaction system any more thar~ one can o/ decoding and encoding (S-r and s-R) phases. It
simultaneously clench and open his hand), the pos- is the lack of such functional separation that forces
sibilities for combination across N independent,com- Skinner (1957) into such intuitively unsatisfying
ponents of r m increase with at least 2x (discrete, assumptions as (a) that we must subvocally mimic
all-or-nothing alternatives) or some larger value a speaker in order to comprehend what he says or
(e.g., 7N if one assumes seven discriminable alter- (b) that having learned to name objects (tacting)
natives within each component). In such a system we will not therefore be able to ask for them
there is obviously great economy; with a relatively (manding), since they are independent single-stage
small number of components one achieves capacity operants.
for distinctive representation of an extraordinarily Note also in this connection that in mediation
large number of entities. It is also obvious that r m theory R x does not bear a one-to-one relation to
does not need to include correlates of all parts of r m (the relation can be completely arbitrary as in
RT, but only those parts which make a difference much of the lexicon, and can be modified, as in
in the behavior toward significates and hence in second-language); neither does R x bear a one-to-
meaning. one relation to R T, In the substitution (Pavlovian
But at this point Fodor moves in with what he conditioning) version of single-stage theory, R x
seems to feel is the "clincher." If one accepts the must be the same as R,r (which is patently false);
necessity of a one-to-one relation between rnl and in the Skinnerian (operant conditioning) version, if
RT--which I do, with the proviso that it is still a signs produce something other than Rx, or ante-
part to whole relation--then, he claims, the formal cedent to Rx, then this something would have to be
difference between single-stage and mediation theo- rx (i.e., unobservable subvocal replicas of the overt
ries disappears and we are left with the trivial verbal response), and hence "rm" would have to
(implied) difference of observability. Thus, we have bear a one-to-one relation with R x (which leads
[S] - - - ( - > r m - - - > Sm--)---> R X to conclusions which are patently nonsensical, as
instead of indicated above). In sum, it is the functional sep-
aration of decoding and encoding stages--yet both
IS] --> R T - - - > S T -> R x utilizing the full machinery of single-stage t h e o r y - -
and are subject to all of the criticisms of single-stage that provides the greater flexibility and explanatory
theory. "To put it still differently, once we grant power of the mediation model. The real danger, as
ANALYSIS OF MEANING 405
I see it, is that by proliferating unobservables (al- By way of reply, first with regard to (a) and (b)
beit of the same kind as overt S's and R's) media- above, although r m is a part of RT, r m is not the
tion theory may be able to "explain" too much and same as R T (as Fodor grants earlier), and therefore
hence become untestable. ST (the automatic feedback from R~) will not be
And, third, let me stress again that the notion of identical with sm. Therefore S should have a much
r m as a simultaneous bundle of events in a limited stronger relation with other responses (e.g., ap-
number of component reaction systems possesses proaching and reaching for the A P P L E ) than the
the combinatorial properties which should render linguistic R x (naming it). However, S T must be
distinctive representation compatible with biological to some degree similar to s m along a generalization
economy. The total momentary pattern of compo- continuum (by virtue of the whole/part relation of
nents of r m can be thought of as a kind of c o d e - - R T to rm) , and therefore the probability of R x
a code which represents for the organism those dif- should indeed increase; I submit that the saying of
ferences among Rw's which make a difference with "apple" does increase in the presence of such objects,
respect to responding appropriately to S's, or things that the saying of "turkey" increases when it appears
signified. To maintain this capacity, there is neither on the Thanksgiving table, and so on. The reason
the requirement that each r m be unique from all within behavior theory why R x does not regularly
others in all respects (a single feature difference occur in situations where the presence of s m is
would be sufficient) nor the requirement that all guaranteed will be made clear in a moment.
ways in which l ~ ' s differ from each other be rep- Actually, Fodor fails to make any distinction be-
resented in their rm'S (only those ways which make tween signifieates (S) and perceptual signs of ob-
a difference in meaning). For these reasons, one of jects. Signifieates are those (usually proximal)
Fodor's concluding statements is misleading--where stimuli of objects which evoke RT, e.g., the posture
he says (p. 81) that "the assumption that each of the hand in holding APPLE, the masticating and
mediator belongs to one and only one total response swallowing of APPLE, and so on. Perceptual signs
appears to destroy any formal distinction between are those (usually distal) stimuli, like the sight of
mediation and single-stage theories." the A P P L E object, whose significances are not in-
(2) Occurrence o] Either S or [S] Should be a nate but themselves must be learned, and presum-
Sufficient Condition ]or Occurrence o] the Linguistic ably via the same basic model.
R x with Which They Have Been Associated. The
Second, with regard to (c) above, since presenta-
argument is as follows: (a) Since r m is a, proper
tion of [S] (either perceptual or linguistic) does
part of R T and S evokes RT, therefore S must in-
increase the probability of rm, and since r m is a
crease the probability of r m as well. (b) Since sm
sufficient condition for occurrence of sin, why does
is an automatic consequence of rm,1 and s m evokes the overt labelling response, R x , so often fail to
linguistic Rx, therefore S must also increase the
occur? Note in the first place that sm is associated
probability of emitting linguistic R x. In other
with a divergent hierarchy of possible responses.
words, the presence of significates should increase the
These can be grossly categorized, as suggested in Fig.
probability of overtly naming them. With respect
2, as linguistic response R x (the labeling, reading,
to signs, the argument is even more straightforward:
or mimicking response in question), linguistic re-
(c) Since [S] produces r m and r m is a necessary sponse R y (e.g., saying "May I have one?" to the
and sufficient condition for occurrence of sin, IS] statement I have some apples b y another), non-
should increase the probability of emitting linguistic
linguistic responses R (e.g., reaching when presented
R x. I n other words, presentation of signs of objects
with the perceptual sign of A P P L E ) , and other rm'S
should increase the probability of making the overt
(as when instructed to free associate). Note in the
linguistic responses associated with the signs. If
second place that which response actually occurs is,
mediation theory does, indeed, differ from single-
in theory, the result of a momentary convergent
stage theory only in observability of the mediating
hierarchy of antecedent stimuli. A context which in-
responses, then it would also seem to lead to pre-
cludes such linguistic cues as Where are the apples?
dictions contrary to fact. We should always read
is not likely to converge upon the R x " a p p l e s " - -
out loud, we should always mimic other people, and
unless the listener is mystified by the question, in
(the reductio ad absurdum) once we emitted a
which case a response with rising intonation is very
word we should always get caught in a cycle of
probable. Such contextual factors may be obvious,
repeating it until semantically satiated (since hu-
but they are certainly not trivial in a theory of the
mans hear the words they emit) !
user of a language. In other words, s m may be a
1 Fodor does not include this particular link in necessary condition for its linguistic Rx, but it cer-
his reasoning, but it is true and is consistent with tainly is not a sufficient condition.
his argument, so I include it. (3) Some Implied Insufficiencies o] r m as a Basis
406 OSGOOD
contrasted with how we may learn from t h e m - - mediators, there are significant differences between
that is, failing to distinguish between performances single- and two-stage theories which Fodor either
and learning--Mowrer clearly opened himself to fails to recognize or misinterprets. The most im-
criticism, which in fact he recognized explicitly in portant of these is the functional independence of
the printed version of his presidential address. I decoding and encoding processes in a two-stage
have contributed to this criticism myself, along psy- theory. It is interesting to note in passing that a
chological lines as well as those pointed up by Fodor. recent (and to me convincing) paper by Jakobovits
However, both in The Measurement of Meaning (1966) argues that single-stage theory can be "re-
(1957) and in my own APA address (1963) I have duced" to two-stage theory--that Skinner is really
tried to show how principles of cognitive interaction a mediation theorist in disguise! However, my argu-
(which in my treatment include rm'S) can apply ments in this paper must not be construed as im-
both to performance (the momentary cognition of plying either that I consider a two-stage model
phrase and sentence meanings) and to learning, or sufficient (elsewhere, e.g., 1957, I have indicated
semantic change (change in r m due to repeated that S-S and R-R integration processes are required
exposure to co-occurrence of word combinations). at least) or that I consider the linguistic phenomena
Thus, in theory, I can interpret the meaning of a under discussion to be simple in nature.
sentence like Tom is a thief on a single presentation
without being necessarily much modified by it (par- REFERENCES
ticularly if I do not believe it!), yet I could have
my semantics permanently distorted by repeated
CHO2aSlC¥, N. Verbal behavior (a review). Lan-
guage, 1959, 35, 26-58.
exposure to Filth Amendment Communist (par-
C~O~SKV, N. Aspects o/ the theory o/ syntax.
ticularly if I believed the implied assertion ])--as
Cambridge, Mass.: M.I.T. Press, 1965.
in fact happened during the heyday of McCarthyism.
FODOR, J. Could meaning be an r m. J. verb. Learn.
Conclusion verb. Behav., 1965, 4, 73-81.
JAXOBOWTS, L. A. Mediation theory and the "single-
As far as the psycholinguist is concerned, Fodor's stage" S-R model: Different? Psychol. Rev.,
critique of rna theory is entirely destructive. If it 1966, in press.
were valid, it would leave behavioristic psychology JAxoBso~r, R. C., AND HALLE, M. Fundamentals o/
(as well as behavioristically oriented psycholinguis- Language. The Hague, Netherlands: Mouton,
tics) with a big, gaping hole right in its middle. 1956.
There are well-documented phenomena (like seman- MOWRER, O. H. The psychologist looks at language.
tic generalization and semantic satiation) which American Psychologist, 1954, 9, 660-694.
cannot be incorporated within single-stage theory OSCOOD, C. E. Method and theory in experimental
in any obvious way or in any nonobvious way so psychology. New York: Oxford U. Press, 1953.
far proposed by its adherents--and as both Chomsky OSGOOD, C. E. Motivational dynamics of language
(1959) and Fodor (1965) have demonstrated, single- behavior. In M. R. Jones (Ed.), Nebraska sym-
stage theories have nothing to say about symbolic posium on motivation. U. Nebraska Press, 1957.
processes. Fodor proposes nothing to fill in this hole OSCOOD, C. E. On understanding and creating sen-
--unless it be an unabridged Webster's Dictionary! tences. American Psychologist, 1963, 18, 735-751.
All of which is not to say that entirely destructive OSGOOD, C. E., Sucl, G. J., AND TANNENBAUI%f, P. H.
criticism must be useless; if valid, it can force com- The measurement oI meaning. Urbana, Illinois:
pletely ne,v directions of theory. Univ. of Illinois Press, 1957.
But I do not believe that his criticisms--including
SXI~rER, B. F. Verbal behavior. New York: Ap-
his attempt to reduce mediation theory to single-
pleton-Century-Crofts, Inc., 1957.
stage theory--are valid. I have tried to demonstrate
that, beyond the matter of observability of the (Received April 7, 1966)