Stimulus relations in comparative primate perspective.
This paper documents a series of findings with apes germane to the study of stimulus relations. These studies were not designed to examine stimulus equivalence as defined by Sidman and Tailby (1982). Yet the complex language processes discussed here can only be interpreted as demonstrating defining elements of equivalence (e.g., Savage-Rumbaugh, Rumbaugh, Smith, & Lawson, 1980). Studies like these may ultimately reveal important answers about species, developmental, and procedural variables that contribute to stimulus equivalence and related behavior (cf. D'Amato, Salmon, Loukas, & Tomie, 1985; Dixon, Dixon, & Spradlin, 1983; Sidman, Rauzin, Lazar, Cunningham, Tailby, & Carrigan, 1982).
Species Differences in Discrimination Learning of Primates
Primates are generally recognized for highly developed learning abilities, large elaborated brains, and stereoscopic color vision. Although primate species are related by common ancestry, comparative research on their learning indicates that they may differ in abilities to learn stimulus relations (Rumbaugh & Pate, 1984). Suggestive evidence has been found by Rumbaugh and Pate (1984) in a simultaneous-discrimination task (cf. "learning set" studies of Harlow, 1949). During trials, S+ and S- stimuli were presented to a subject with selections of the S+ producing reinforcement. After reaching a criterion of correct responding, the subject experienced a "transfer test" in which the S+ became the S- and the S- became the S+. This procedure of training a discrimination to criterion and then conducting a transfer test was repeated with new stimulus pairs until the reversal performance stabilized.
Tests with 11 species of primates show that as one moves from prosimians to monkeys through apes, reversals in discriminated responding occur more rapidly. Chimpanzees, orangutans, and gorillas show the most rapid reversals. Prosimians and monkeys showed minimal tendencies to reverse. The only exception is the rhesus macaque, a species of monkey, who perform like the chimpanzees.
This finding is significant for three reasons. First, the transfer index shows a between-species correlation of 0.72 with encephalization (Fobes & King, 1982) suggesting that learning sets can be a behavioral consequence of encephalization. Second, as called for by Sidman (1986), such procedures might provide pure measures of "behavior development," less likely to be contaminated by previous learning than traditional IQ tests. Third, and of greatest relevance here, it shows that asymptotic discrimination reversal performance may be species-delimited. That is, some organisms may learn about particular stimuli imbedded in contingencies; others learn about contingencies in which stimuli occur as well as the stimuli, resulting in extremely rapid learning about the functions of those stimuli (cf. Harrison & Green, 1990).
Language Learning ln Chimpanzees
Efforts in the language training of chimpanzees have furnished several examples of generalized learning performances involving stimulus relations. Chimpanzees show that more can be learned from matching-to-sample than direct stimulus relations; performances discussed in this section provide several illustrations.
Error-cued problem solving. During their language instruction, Sherman and Austin (Pan troglodytes) acquired a repertoire of conditional discriminations in which they were required to "request" various visually presented foods (henceforth, "visual foods"). Given the presentation of a sample visual food, selection of the corresponding lexigram from an array of lexigrams produced a taste of that food (cf. Skinner's 1957 concept of manding, where the response is controlled by deprivation). As new visual food|right arrow~lexigram combinations were presented to Sherman and Austin, they mastered them in fewer trials. One reason for the improvement is suggested by the outcome of an attempt to teach them to "name" visual foods that followed.
Using the same food-lexigram pairs, a sample visual food was presented and their selection of the previously learned lexigram produced a taste of a different food. Whereas in the initial procedure the food that was named corresponded to the food that was received, in the current procedure the named and received foods were different. Under these conditions their performance rapidly disintegrated; they seemed to select lexigrams at random (cf. McIlvane, Dube, Kledaras, de Rose, & Stoddard, 1992). A likely explanation is that, in previous training, "errors," signaled by failures to obtain the displayed food, set the occasion for "searching" the keyboard for the correct lexigram. This "error-cued" searching is formally similar to the reversal-learning performance discussed above (Rumbaugh & Pate, 1984) and may be the behavioral mechanism underlying learning set (Harlow, 1949). Error-cued searching, analogous to an optimal "learning set" performance, would have necessarily assisted in learning new relations more rapidly.
Learning by exclusion. Related to this finding was evidence of learning by exclusion (McIlvane & Stoddard, 1981). Having learned many visual food|right arrow~lexigram relations, the chimps were tested to see if "assigning" names to foods was a generalized class of behavior (Savage-Rumbaugh, 1986). Prior to a test, four to seven unassigned lexigrams were added to the keyboard. None of the chimps used the new lexigrams during testing of previous relations, verifying that new lexigrams would not be selected simply because of novelty. Then, given the display of a novel food, the chimps adopted the selection of an unassigned lexigram and maintained it in their language thereafter (Savage-Rumbaugh, 1986; cf. Harrison & Green, 1990; Saunders, Saunders, Kirby, & Spradlin, 1988). This behavior also is analogous to unreinforced conditional selection of comparison stimuli as reported by Saunders et al. (1988). Sherman and Austin learned five new terms in this manner.
Learning by observation. Sherman and Austin also learned visual food|right arrow~lexigram relations by observation. If Sherman would assign a new lexigram to a new food, Austin would also adopt the lexigram; if Austin adopted a lexigram, Sherman would conform to Austin's choice. This kind of imitation involving the learning of conditional discriminations is by no means simple. It resembles Pavlovian sensory-preconditioning because sample-comparison pairs were presented independently of the observers' behavior. The role of operant contingencies is also conspicuous, however, because the reinforced performance of one chimp was the discriminandum for the correct performance of the other.
Matching to sample and grammatical frames. Error-cued searching, unreinforced conditional selection of comparisons, and learning by observation, illustrate unprogrammed learning strategies that facilitate the acquisition of word meanings. Some of these performances suggest that matching to sample may acquire an autoclitic function like that of assertion (i.e., as in Skinner's, 1957, autoclitic). For example, in learning by exclusion, the matching-to-sample task defines the roles of sample and comparison stimuli, setting the occasion for a subject to respond to a new stimulus relation. Grammatical "frames" like those of assertion in natural language similarly establish relations between language terms and their referents. A striking example is provided by typical 2-year-old children who reliably recognize an agent-recipient relation in an utterance containing a novel verb (Hirsh-Pasek & Golinkoff, 1991). The expansion of a speaker's lexicon can be greatly facilitated in this way by allowing stimulus relations to be learned without direct reinforcement (cf. Baer, Peterson, & Sherman, 1967; Hirsh-Pasek & Golinkoff, 1991; McIlvane & Stoddard, 1981).
Procedural Constraints on Emergent Relations
On several occasions the chimpanzees' relational performances showed clear dependency on procedural factors. Stimulus relations learned in one procedure did not influence performance in a new test situation. Having learned a set of stimulus relations in one procedure, their learning would not surface in a different testing arrangement. Only after some additional training was provided in the new procedure did the existing relations emerge. The following examples illustrate how relational responding may be constrained by differences in training and testing regimens.
Paired associates and conditional discrimination. Prior to matching-to-sample exposure, Sherman and Kenton (Pan troglodytes) participated in a study examining whether visual food|right arrow~lexigram relations could be learned on a paired-associate basis (Savage-Rumbaugh, 1986). In that experiment, the chimps had four food/drink lexigrams available for which choices produced corresponding foods. Both chimps discriminated between the four lexigrams as seen in the positive correlation between lexigram preferences and food preferences determined prior to sessions. This behavior fits well with Skinner's (1957) concept of the mand, in which a verbal response is directly under the control of a state of deprivation. In a subsequent test for tacting, however, given the presentation of a visual food sample, they could not correctly indicate the lexigram that they had previously used to produce the sample food type. Instead, they persisted in selecting lexigrams for more preferred foods over hundreds of trials.
Evidence that Sherman and Kenton had learned the food-lexigram relations emerged only later after they were exposed to conditional discrimination training. Using eight new visual food|right arrow~lexigram pairs, the chimps were presented with a visual food (or drink) sample and required to produce a lexigram sequence, "Please machine give (pour) ...," plus a food name to obtain that food. Now their lexigram board had the symbols required by the sentence plus several other food symbols, thus establishing conditional selection of lexigrams. Once these were learned, the three lexigrams employed in the former condition were reintroduced with novel distractor lexigrams. Both chimps made several errors before emitting the correct lexigram sequence. In every case, however, they performed errorlessly after the first correct choice was reinforced.
It is unlikely that a single reinforcer was sufficient to establish the conditional relations just described since many errors had accompanied learning of the eight intervening visual food|right arrow~lexigram relations. Instead, conditional discrimination training seems to have produced a context for relational responding to previously learned manding relations; this context now set the occasion for tacting (cf. Bush, Sidman, & de Rose, 1989; Savage-Rumbaugh, Sevcik, & Hopkins 1988).
Transforming mands into tacts. Similar context-sensitive responding was displayed by Sherman and Austin during initial attempts to teach tacting with matching to sample (Savage-Rumbaugh, 1986). As recounted previously, their lexigram selections were severely disrupted when, after learning to request visual foods, naming of those foods was reinforced with other foods. The transition to naming was instead successfully achieved using a fading procedure with only three items: During fading, the sample visual food was slowly cut back while there were other foods, praise, and physical activity for correct responses. Sherman and Austin rapidly learned to tact the three foods with this procedure. Finally, they were tested with 21 other visual food samples which they had previously only requested. They responded correctly to all on the first trial without errors, indicating that visual food|right arrow~lexigram matching was no longer influenced by the reinforcer type. In this case, the conditional discrimination training on the three initial relations apparently resulted in a new context for responding, one in which the selection of a lexigram was controlled only by an antecedent and not a consequence (cf. McIlvane et al., 1992).
Equivalence Between Concepts, Names, and Their Referents
The foods and drinks employed in the preceding training were subsequently incorporated into a study on concept learning and mediated transfer (Savage-Rumbaugh et al., 1980). As a requisite to this study, Sherman and Austin had also learned to name 11 tools (including such items as lever, string, and wrench). The chimps then began a sequence of procedures to generate the "food" and "tool" categories as outlined in Table 1, Section A. First sorting three foods and three tools into two bins was reinforced. The experimenter would put a food into one bin and a tool into the other and then required the apes to sort the four remaining items correspondingly. Once they sorted reliably, they were required to put an item into a bin and use a "food" or "tool" lexigram to label it, and finally they learned to label items as "foods" and "tools" without the bins. The chimps were then presented with five foods and five tools for which they had learned lexigram names but had never before categorized as '"foods" and "tools." Austin correctly categorized all and Sherman missed only one, demonstrating abstraction of the "food" and "tool" category labels.
The chimps then received additional training in labeling foods and tools as shown in Table 1, Section B, but this time in the form of labeling photographs. At first, photos of the three foods and three tools used in the initial categorization training were taped to the corresponding objects and the chimps were asked to label them as "foods" and "tools." Once they labeled the pairs accurately, they learned to label the photographs alone. Then they were presented with nine photographs of items for which they had learned lexigram names but never labeled as "foods" and "tools" (four tools and five foods). Sherman labeled all correctly. Austin, however, called all of them tools. With some help he learned to rotate the pictures to reduce surface reflections masking the images and he then also labeled them correctly.
In the final condition summarized in Table 1, Section C, the three food lexigrams and three tool lexigrams used in training were taped onto photos of corresponding objects and the chimps were asked to label them as "foods" and "tools." When they categorized these pairs reliably, the photos were removed and they learned to label the lexigrams alone. Then they were presented with 10 food and 7 tool lexigrams that they had never before labeled as "foods" and "tools." Sherman was correct on 15 of 16 correct (one tool lexigram unfamiliar to Sherman was withheld) and Austin was correct on 17 of 17.
To summarize the study, once the foods and tools could be individually named, sorted by concept class, and named by concept class, the names of foods and tools could then also be named by concept class. As seen in Figure 1, the outcome constitutes a test for equivalence, with only the exception that symmetry between lexigrams and concept lexigrams was not tested.
Training along route to the final outcome carefully maintained and tested the integrity of the stimulus relations. For example, once some foods and tools could be sorted. the chimps were tested with new foods and tools. This ensured that sorting was based on the abstract generalizable properties of the items; the performance was not based on direct training of a sorting response for each item. Such training also established stimulus contexts dictating how lexigrams, objects, and concept names were to be manipulated. In the last test, for example, Sherman and Austin first learned to categorize three lexigrams as foods and tools before they were tested on the remaining lexigrams. That training was important because it then set the occasion for categorizing the remaining lexigrams, and that categorization could be based only on concept class membership, not on direct training (cf. Fields, Reeve, Adams, & Verhave, 1991).
Observational Learning of Stimulus Relations and the Emergence of Symmetry
The training experiences and outcomes with Sherman and Austin contrast sharply with those of two pygmy chimpanzees (Pan paniscus), Kanzi and Mulika (Savage-Rumbaugh, Sevcik, Brakke, Rumbaugh, & Greenfield, 1990; Savage-Rumbaugh, Sevcik, Rumbaugh, & Rubert, 1985). Kanzi's first exposure to the lexigram-based language system was as an observer of Matata, his adoptive mother's, training. His mother, a wild-caught female, spent 2 years on procedures similar to those employed with Sherman and Austin, and only learned six visual food|right arrow~lexigram relations. During that time, Kanzi showed interest in Matata's symbols but did not display evidence of having learned aspects of their use. At the end of 2 years, he began to touch the "chase" symbol, which would occasionally result in a game of chase with a caregiver.
At about 2-1/2 years of age Kanzi was briefly separated from his mother. Following the separation, he began using his mother's lexigram board to request a number of foods and activities (Savage-Rumbaugh et al., 1990). On seeing a particular food, for example, he would request it by pointing to its corresponding lexigram. Later on formal tests where Kanzi was presented with an object and asked to touch the corresponding lexigram, and in test for symmetry in which he was shown a lexigram and asked to point to an object, he scored 90% or better with the symbols that he used.
Similar but more dramatic learning was shown by Mulika, Kanzi's younger sister. She was separated from her mother Matata at 4 months of age. Mulika's first lexigram use was restricted to the "milk" symbol, which she would frequently combine with gestures to indicate another request. For example, she would point to "milk" and then point to an apple. Over the next few months, however, her vocabulary increased greatly in quantity and her generalized use of milk decreased. Kanzi and Mulika's lexigram use was unlike that of their mother Matata, Sherman, and Austin, in that it was acquired through the observation of others; at no time were formal matching-to-sample procedures arranged for them.
In time another important difference with Sherman and Austin became apparent. Sherman and Austin did not learn to comprehend spoken English despite the fact that most of their previous training was accompanied by speech (Savage-Rumbaugh et al., 1985). They could not, for example, point to a picture given a spoken English word, even though they could do so given a lexigram. In contrast it became apparent to those worki ng with Kanzi and Mulika that they could comprehend speech (Savage-Rumbaugh, McDonald, Sevcik, Hopkins, & Rubert, 1986). At 69 months of age, Kanzi could comprehend 150 English words. Upon hearing a word, he could select the corresponding lexigram or point to a picture of the referent. At 30 months of age, Mulika could comprehend 70 spoken English words in the same test even though she only used six symbols.
The question of whether chimpanzees like Kanzi show formal equivalence between spoken English, lexigrams, and referents, as defined by Sidman and Tailby (1982), is yet to be answered. What is known is that novel testing contexts occasioned the emergence of components of equivalence like symmetry, and that equivalences do exist between terms in their present language (cf. Savage-Rumbaugh et al., 1986).
The difference in learning by Kanzi and Mulika in comparison to Matata, Sherman, and Austin seems to be, at least in part, one of early experience (Savage-Rumbaugh et al., 1990). Exposure to English, for example, began for Kanzi at 6 months and for Mulika at 4 months. Matata, a wild-caught female, did not begin her language training or systematic exposure to spoken English until she was reproductively mature, Sherman began at 2-1/2 years of age, and Austin at 1-1/2 years. Moreover, Matata's great difficulty in learning lexigram-referent relations suggests that the difference is not only one of species.
Research presented here on language abilities of chimpanzees can be interpreted with concepts derived from the study of stimulus relations. In doing so one finds compelling evidence of symmetry and transitivity. En route to those outcomes were findings of complex learning strategies as in learning by exclusion. Most recent findings suggest that early experience greatly influences the quality of learning. The acquisition of language-encompassed relations seen in pygmy chimpanzees born and raised in language-rich environments is impressive. Those findings show that stimulus relations can be learned in young, preverbal chimpanzees without evidence of reinforcement. Future research on the influence of early experience on learning will undoubtedly open new and important avenues in our understanding of behavior.
BAER, D. M., PETERSON, R. F., & SHERMAN, J. A. (1976). The development imitation by reinforcing behavioral similarity to a model. Journal of the Experimental Analysis of Behavior, 10, 405-416.
BRELAND, K., & BRELAND, M. (1961). The misbehavior of organisms. American Psychologist, 16, 681-684.
BUSH, K. M., SIDMAN, M., & DEROSE, T. (1989). Contextual control of emergent equivalence relations. Journal of the Experimental Analysis of Behavior, 51, 29-45.
D'AMATO, M. R., SALMON, D. P., LOUKAS, E. & TOMIE, A. (1985). Symmetry and transitivity in monkeys (Cebus apella) and pigeons (Columba livia). Journal of the Experimental Analysis of Behavior, 44, 35-47.
DIXON, M., & DIXON, L. S., & SPRADLIN, J. E. (1983). Analysis of individual differences of stimulus control among developmental delayed children. In Advances in learning and behavioral disabilities (Vol. 2). New York: JAI Press.
FIELDS, L., REEVE, K. R., ADAMS, B. J., & VERHAVE, T. (1991). Stimulus generalization and equivalence classes: A model for natural categories. Journal of the Experimental Analysis of Behavior, 55, 305-312.
FOBES, J. L., & KING, J. E. (1982). Measuring primate learning abilities. In J. L. Fobes & J. E. King. (Eds.), Primate behavior (pp. 289-321). New York: Academic.
HARLOW, H. F. (1949). The formation of learning sets. Psychological Review, 56, 51-65.
HARRISON, R. J., & GREEN, G. (1990). Development of conditional and equivalence relations without differential consequences. Journal of the Experimental Analysis of Behavior, 54, 225-237.
HIRSH-PASEK, K., & GOLINKOFF, R. M. (1991). Language comprehension: A new look at some old themes. In N. A. Krasnegor, D. M. Rumbaugh, R. L. Schieflbusch, & M. Studdert - Kennedy (Eds.), Biological and behavioral determinants of language development (pp. 301-320). Hillsdale, New Jersey: Lawrence Erlbaum Associates.
MCDOWELL, J. J. (1988). Matching theory in natural human environments. The Behavior Analyst, 11, 95-109.
MCILVANE, W. J., DUBE, W. V., KLEDARAS, J. B., DE ROSE, J. C., & STODDARD, L. T. (1992). Stimulus-reinforcer relations and conditional discrimination. In S. C. Hayes & L. J. Hayes, Understanding verbal relations. Reno, NV: Context Press.
MCILVANE, W. J., & STODDARD, L. T. (1981). Acquisition of matching-to-sample performances in severe retardation: Learning by exclusion. Journal of Mental Deficiency Research, 25, 33-48.
REVUSKY, S. H., & GARCIA, J. (1970). Learned associations over long delays. In G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 4). New York: Academic Press.
RUMBAUGH, D. M., & PATE, J. L. (1984). The evolution of cognition in primates: A comparative perspective. In H. L. Roiblat, T. G. Bever, & H. S. Terrace (Eds.), Animal cognition (pp. 403-420). Hillsdale, NJ: Lawrence Erlbaum Associates.
SAUNDERS, R. R., SAUNDERS, K. J., KIRBY, K. C., & SPRADLIN, J. E. (1988). The merger and development of equivalence classes by unreinforced conditional selection of comparison stimuli. Journal of the Experimental Analysis of Behavior, 50, 145-162.
SAVAGE-RUMBAUGH, E. S. (1986). Ape language: From conditioned response to symbol. New York: Columbia University Press.
SAVAGE-RUMBAUGH, E. S., MCDONALD, K., SEVCIK, R. A., HOPKINS, W., & RUBERT, E. (1986). Spontaneous symbol acquisition and communicative use by a pygmy chimpanzee (Pan paniscus). Journal of Experimental Psychology: General 115, 211-235.
SAVAGE-RUMBAUGH, E. S., RUMBAUGH, D. M., SMITH, S. T., & LAWSON, J. (1980). Reference: The linguistic essential. Science, 210, 922-924.
SAVAGE-RUMBAUGH, E. S., SEVCIK, R. A., BRAKKE, K. E., RUMBAUGH, D. M., & GREENFIELD, P. M. (1990). Symbols: Their communicative use, comprehension, and combination by bonobos (Pan paniscus). In C. Rovee-Collier & L. P. Lipsitt (Eds.), Advances in infancy research (Vol. 6). Norwood, NJ: Ablex.
SAVAGE-RUMBAUGH, E. S., SEVCIK, R. A., & HOPKINS, W. D. (1988). Symbolic cross-modal transfer in two species of chimpanzees. Child Development, 59, 617-625.
SAVAGE-RUMBAUGH, E. S., SEVCIK, R. A., RUMBAUGH, D. M., & RUBERT, E. (1985). The capacity to acquire language: Do species differences have anything to say to us? Philosophical transactions of the Royal Society of London, B308, 177-185.
SIDMAN, M. (1986). The measurement of behavioral development. In N. A. Krasnegor, D. B. Gray, & T. Thompson (Eds.), Advances in behavioral pharmacology. Vol. 5: Developmental behavioral pharmacology (pp. 43-52). Hillsdale, NJ: Erlbaum.
SIDMAN, M., & TAILBY, W. (1982). Conditional discrimination vs. matching to sample: An expansion of the testing paradigm. Journal of the Experimental Analysis of Behavior, 37, 5-22.
SIDMAN, M., RAUZIN, R., LAZAR, R., CUNNINGHAM, S., TAILBY, W., & CARRIGAN, P. (1982). A search for symmetry in the conditional discriminations of rhesus monkeys, baboons, and children. Journal of the Experimental Analysis of Behavior, 37, 23-44.
SKINNER, B. F. (1957). Verbal behavior. New York: Appleton-Century-Crofts.
Sequence of Training and Testing Procedures in Savage-Rumbaugh et al. (1980)
A. Training and Testing of Food/Tool Categories.
1. Learn to sort three foods and three tools into two bins.
2. Test for generalized sorting with five other foods and five other tools.
3. Learn to sort the three foods and three tools into two bins, labeling each item with "food or "tool" lexigrams as they are put into bins.
4. Learn to label the three foods and three tools with "food" and "tool" lexigrams, respectively.
5. Test for appropriate abstraction of food and tool concepts by requiring labeling of five new foods and five new tools with "food" or "tool" lexigrams.
B. Training and Testing of Food/Tool Categories with Photographs.
1. Learn to label three foods and three tools with photographs taped to them with "food" and "tool" lexigrams, respectively.
2. Learn to label the photographs of the three foods and three tools with "food" and tool" lexigrams.
3. Test for correct labeling of five new food and four new tool photographs with "food" and "tool" lexigrams.
C. Training and Testing of Food/Tool Categories with Lexigrams.
1. Learn to label three photographs of foods and three photographs of tools with lexigrams taped to them with "food" or "tool" lexigrams.
2. Learn to label the three food and three tool lexigrams with "food" and "tool" lexigrams.
3. Test for correct labeling of ten new food and seven new tool lexigrams with "food" and "tool" lexigrams.
Note. Prior to this study, Sherman and Austin had learned the lexigram names for all of the foods and tools employed within it.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||Special Issue: Stimulus Equivalence|
|Author:||Cerutti, Daniel T.; Rumbaugh, Duane M.|
|Publication:||The Psychological Record|
|Date:||Sep 22, 1993|
|Previous Article:||Emergent relations in the formation of stimulus classes by pigeons.|
|Next Article:||A California sea lion (Zalophus californianus) is capable of forming equivalence relations.|