REPRESENTATIONAL MEDIATION THEORY

It must be understood that learning theory is not synonymous with, but rather is a necessary part of, general behavior theory.¹ General behavior theory would include many innately determined properties and functions of the nervous system, including wired-in sensory and motor projection systems, the existence of unlearned reflexes and instincts, and the possession of certain innately determined cognitive abilities by any given species. In other words, behavior theory does not necessarily assume that the ‘ mind ’ is a tabula rasa at birth. Among the innately given of human cognitive abilities would undoubtedly be: (i) the gestalt-like tendencies to establish perceptual entities in terms of such principles as common fate, qualitative sensory similarity, sensory proximity, and contour completeness of continuity; (2) the tendency to organize behavior, both perceptuo-motor and linguistic, hierarchically in terms of levels of units-within-units-within-units; (3) the tendency to organize or differentiate the units within each level componentially, so that a large number of alternatives can be differentiated in terms of a relatively small number of elements or features. Although it is true that these tendencies find expression in, i.e., are universal to, all human languages, this is not because they are linguistic in nature but rather because linguistic systems develop within the constraints (or potentialities) of human cognitive abilities more generally.

Representational mediation theory is the only learning approach that has seriously attempted to incorporate the symbolic processes in general and meaning in particular within an S-R associationistic model. It is traceable directly to the germinal work of Clark L. Hull (1930, 1943), although he utilized it primarily in drive and reinforcement connections (the fractional anticipatory goal reaction, r_g). It was extended as a general rather than occasional explanatory device in the writings of people like Miller and Dollard (1941), Mowrer (i960), Goss (1961), the Kendlers (1962) and many others, including the writer (Osgood, 1953 and subsequently). In its present form it has come to be known as two-stage learning theory or neo-behaviorism (and r_g has become r_m, or representational mediation process).²

My own use of representational mediation theory as the basis for a behavioristic theory of meaning has made explicit the origins of mediation (symbolic) processes in non-linguistic, perceptuo-motor behaviors (cf. my 1957 paper on perception and language as cognitive processes). Significates (referents or things signified) are simply those patterns of stimulation, including previously learned signs, which regularly and reliably produce distinctive patterns of behavior. Thus, early in infant life, presence of a nipple in the mouth reliably produces a complex pattern of sucking movements, milk ingestion, reduction of hunger pangs, autonomic changes characteristic of gratifying states of affairs, relaxation of voluntary musculature, and so forth. Sight of the BOTTLE (a perceptual sign) regularly antedates presence of nipple-in- mouth, and this is a necessary and sufficient condition for sign learning. The primary postulate of this theory of meaning is that a stimulus pattern which is not the same physical event as the thing signified will become a sign of that significate when it becomes conditioned to a representational mediation process, this process (a) being some distinctive part of the total behavior produced by the significate and (b) serving the mediate overt behaviors which are appropriate to (‘take account of’) the significate. Thus the perceived entity, BOTTLE, becomes a perceptual sign of MILK FOOD-OBJECT and, appropriately, the infant will now salivate and make anticipatory sucking movements to the sight of BOTTLE.

Obviously, this is only the crudest beginning of a semantic system, but it is a beginning nevertheless, and by the time LAD is launching himself into linguistic sign learning most of the entities in his familiar environment (utensils, food objects, pets, toys, faces that smile and faces that frown) have been meaningfully differentiated in terms of distinctive mediators (representations, cognitions). In this pre-linguistic process, not only are particular representations established for perceptual signs, which can then serve as ‘prefabricated’ mediators in learning correlated linguistic signs, but even more importantly many of the distinctive features of meaning (components of r_m in behavior theory terms) are being differentiated.

Although this is only the roughest sketch of the neo-behaviorist theory of meaning, it should at least be evident that, in principle, it both provides for the meaningfulness of non-linguistic, perceptual signs and indicates that perceptual and linguistic signs share the same semantic system. Debate over the adequacy of representational mediation theory as the foundation for a theory of meaning (Fodor, this volume, 1966; Osgood, 1966, 1970) has focused on two major issues - the nature of mediating processes and their source.

With regard to the nature of representational mediation processes in behavior, I have stated repeatedly that r_m and its automatic consequence, s_m, are defined functionally in S-R terms in order to incorporate mediation processes within the larger body of learning theory and facilitate transfer of such single-stage principles as habit-strength, generalization and inhibition to this two-stage model. Representing the mediation process with the symbol M, we can say that M has response-like functions as a dependent event (decoding, understanding) and stimulus-like functions as an antecedent event (encoding, creating). In theory r_m’s have the status of hypothetical constructs (with potential existential properties) rather than of intervening variables (convenient summarizing fictions). I assume that, certainly in the adult human language user, these mediating, symbolic events have become purely cortical processes - processes whose neurological nature and locus will not be known for a very long time.

With regard to the source of representational mediation processes, it is true that the theory postulates - for the historical origins of meanings - derivation of r_m’s (mediating reactions to signs) from R_T’s (overt reactions to the things signified), and Fodor has posed (this volume) what he believes to be a dilemma: either r_m’s must stand in unique one-to-one relations with their R_T’s (in which case mediation theory is reducable to single-stage learning theory and subject to Chomsky’s (1959) strictures) or r_m’s must stand in one-to-many relations to R_T’s (in which case the theory fails in principle to account for unambiguous reference). I have accepted the first horn of this dilemma, but argue that it is not really a ‘horn’ at all. Let me summarize my arguments briefly.

(1) Since r_m’s are hypothetical constructs, they bear a part-to-whole relation to R_T’s only in the sense of being ‘derived from’ and being ‘distinctively representational of’ R_T’s. They are not ‘parts of’ in the literal sense of being a sub-set of the overt R’s making up R_T.

(2) Since r_m’s evoked by previously learned signs can also serve as R_T’s in the establishment of subsequent signs, there is no requirement in theory that every sign be based upon significate-produced behaviors. Thus listeners and readers can construct increasingly elaborate systems of meanings for abstractions (gravity, justice), for things not experienced directly (zebras, Viet Cong) and even things non-existent (unicorns and elves), from the context of other words which already have meanings. I have referred to this process as assign learning. Needless to say, a very large proportion of the lexicons of adults are assigns. The distinctive semantic features (components of r_m) established for primary perceptual and linguistic signs are capable of complex, novel and, indeed, ‘emergent’ interactions - a crucial point to which I shall return momentarily.

(3) Representational mediators are assumed to be componential in character. A relatively small number of independent, bipolar reaction components, by virtue of their combination in diverse simultaneous patterns, can serve to differentiate a very large number of distinct r_m’s, each related to its R_T uniquely - but uniquely as a whole, not in terms of unique components. This is what I have referred to as the ‘emic’ principle of neo-behaviorism (cf. Osgood, 1970). Like the phoneme (or, more properly, the sememe) in linguistic theory, the r_m in neo-behaviorism (a) renders functionally equivalent classes of different behavioral events, either signs having the same significance for the receiver of behaviors expressing the same intention for the source, (b) is an abstract entity, unobservable itself but necessary for interpretation of what is observed, and (c) is resolvable into a ‘ simultaneous bundle ’ of distinctive features or components which serve to differentiate among classes of meanings (significances or intentions). In exactly the same sense that one cannot substitute some particular phone (which one?) for a phoneme in linguistic theory, one cannot substitute any particular overt R (which one?) for an r_m in behavior theory. Thus representational mediation theory does not reduce to single-stage S-R theory, as Fodor claims.

It is important to point out that r_m’s are not mere replicas of their R_T’s; rather, they are representations of those features of R_T’s which have made a difference in appropriateness of behaving toward the things signified by signs and have therefore been differentially reinforced - again in close analogy to the distinctive features of phonemes. Thus the locomotor movements by which an organism moves from place A to place B (walking, running, swimming, etc.) may be non-distinctive with respect to avoiding signs of danger at A or approaching signs of safety at B, but the affective features of the diverse ways of avoiding vs. approaching will be common to the different R_T’s and hence be differentially reinforced as components of the meanings of the signs of A vs. B.

In what sense can such a representational system be said to have ‘emergent’ properties? The study of cognitive dynamic has a considerable, if often controversial, literature in psychology (cf. Osgood, i960, for a review). This research has been concerned with laws determining changes in meaning of signs when they interact contextually with other signs (attitude change toward sources and topics, interaction between perceptual and linguistic signs, prediction of the meanings of phrases from the meanings of their component words, and so forth). According to representational mediation theory, each feature of meaning corresponds to a single bipolar reaction system (component of r_m) which can only be in one state at any one time. If a component is simultaneously driven in opposed directions (e.g., as in sincerity admires on an Abstract/Concrete feature or as in manipulate generously on an Ego/Alter orientation feature), then - exactly as is the case for reciprocally antagonistic overt response systems - that component ‘freezes’ and what we refer to as awareness of semantic anomaly is experienced. Similarly, driving a component simultaneously in the same direction produces semantic intensification (as in manipulate slyly or in sudden surprise), and if one sign has coding on a feature for which another sign does not, then the meaning of the combination is semantically modified or enriched (as in plead with sincerely on a Moral/Immoral feature).

By virtue of having their own rules of interaction, these components of r_m’s to signs do display certain emergent properties: (1) They enable the listener to react appropriately to (i.e., interpret) novel utterances; although one probably has never encountered the sentence She will make someone a nice husband before, the masculinization of she via interaction with husband serves to appropriately characterize the woman in question. (2) Meanings for new lexical items can be built out of the inter¬ actions of the features of contextual words which already have meanings (assign learning); in concept formation, superordinate terms like fruit will retain those features shared by their subordinate instances (lemon, banana, apple) but become non-differential on conflictual features (e.g., Shape or Sweetness). (3) Interaction among semantic components allows for selection among alternative senses of words (cf. the ‘projection rules’ of Katz and Fodor, 1963); for example, selection of two different senses of the word play in the sentences he wrote a good play vs. he made a good play is based upon congruence with the contrastive features of write and make. (4) Presence or absence of anomaly in this behavioral sense parallels the linguistic notion of ‘selection rules’ (cf. Chomsky, 1965); overriding such rules, when sufficient numbers of shared features are present, creates metaphors like the thunder shouted down the mountainside - where the — Human feature of thunder is countermanded by the other shared features between thunder and + Human shouted (i.e., thunder is personified).

The most deep-set resistance to this neo-behaviorist account of meaning, I think, is to its anchoring of r_m’s to R_T’s - that is, to having semantic distinctions originate in differences in the behaviors to Things rather than simply in differences in the perceptions of Things. Not only does this seem too restrictive as a source of features, but it reduces the gap between rational (talking) man and mechanistic (non-talking) animal quantitatively and certainly eliminates it qualitatively.³ In my experience, this ‘metaphysical repugnance’ persists even when it is understood that this anchoring is historical, applying to the acquisition of primary perceptual and linguistic signs, and not contemporary for the adult speaker.

From a behavioristic point of view, this anchoring of meanings to subsequent as well as antecedent observables serves at least two functions: (1) It eliminates, in principle, the sheer circularity of defining semantic features in terms of perceptible distinctions - which would erase all but terminological differences between mentalistic and behavioristic positions. (2) It places rather severe constraints upon what differences that are perceptibly discriminable will actually come to make a difference in meaning. The human sensory systems are capable of numbers of perceptible discriminations which are literally incredible; any psycholinguistic theory must be able to account for the relatively small number (and nature) of those discriminations which do in fact make a difference in meaning. Without such constraints, representational mediation theory would indeed become a universal Turing Machine - like God’s Will, capable of explaining anything but patently incapable of disproof. As is currently being charged against contemporary generative grammar, it would be too powerful.

¹ In a way, the learning-theory component of general behavior theory is analogous to the syntactic component of a generative grammar; it is a set of ‘ rules’ for generating increasingly complex behaviors from more primitive, innately given, elements and processes.

² Actually, my own version of neo-behaviorism is a three stage (or level) theory, including a level of sensory (s-s) and motor (r-r) integration designed to account for gestalt-like phenomena of perceptual organization (e.g., closure) and motor skill or programming phenomena. See Osgood, ‘A Behavioristic Analysis of Perception and Language as Cognitive Phenomena’, 1957.

³ I am sure that both Chomsky and T, despite our metaphysical differences, wish that people would behave more rationally with respect to other people - but this would be more like most animals!