In the first phase of the experiment participants are trained to match (typically three) sample stimuli to the same number of comparison stimuli e.g. A1 to B1, A2 to B2 and A3 to B3. It is presumed that automatic feedback by the computer i.e. ‘Correct’ or ‘Incorrect’ acts as reinforcement for correct responses. Across trials sample stimuli varies (A1, A2 or A3) but the comparison stimuli appears in different locations so that position gives no clue to correct responses. Once performance criterion is achieved – normally correct responses on more than 90% of a batch of trials, the second phase of the experiment begins. In this phase B1, B2 or B3 appear as sample stimuli and the new comparison stimuli, C1, C2 and C3 appear at the bottom of the screen. Correct answers will be C1 given B1, C2 given B2 and C3 given B3. Again there is no physical resemblance or logical or spatial relationships between stimuli, training continues until the criterion level is reached.
At this stage, verbally competent human participants can correctly identify various untrained matching responses. For example, participants will correctly match the appropriate A stimuli with the B stimuli if B appears as a sample and the A’s as comparisons. This is the reverse of the previously trained relationship and is referred to as ‘symmetry’. If the A stimuli are presented as both samples and comparisons, the participant will correctly match each A stimulus with itself. This phenomenon is referred to as ‘reflexivity’ and will also be observed with B and C stimuli. This may seem quite obvious but reflexivity does not normally occur with non-human participants. If the A stimuli appear as samples and the C stimuli as comparisons, verbally competent human participants will correctly match C stimuli to the appropriate to the appropriate A stimuli, this is referred to as ’transivity’. Combined ‘symmetry and transivity’ is observed when C stimuli are presented as samples and A stimuli as comparisons. When all these matching responses have been observed, a stimulus equivalence class is said to have been formed because each member of the class is treated equivalently.
A to B and B to C training is sufficient to produce stimulus equivalence classes in verbally competent human participants but evidence suggests that this is a uniquely human capacity and does not apply to other species. Dugdale and Lowe (1990) found that animals of other species, including chimpanzees, may succeed in reaching criterion performance on matching to sample tasks but they do not show any emerging relationships, they may not, for example, even match stimuli to themselves without explicit training. This work by Dugdale & Lowe (1990) and others produces very revealing findings about the human capacity for constructing equivalence classes and ‘verbal labelling’. Ever since the late 19th century origins of modern psychology, investigators have been preoccupied with determining the difference in learning capacity between species. Probably the most obvious example of this is human’s ability to verbally communicate. Hayes & Hayes (1992) have suggested that it is the process of relational learning that underpins human verbal skills. A key event in their view is ‘bidirectional training’, a common parenting practice. This is best understood by the example of a child being reinforced to identify the family dog by the spoken word ‘dog’ or, by pointing to the dog when asked “Where’s the dog?”. The child would soon form further members of the stimulus equivalence class by learning to identify photographs of the dog and also other dogs it encounters. In this context, all members are treated as being the same even though they are physically dissimilar. This recent work however, does appear to suggest a clear discontinuity between humans and other species, even chimpanzees.
An important feature of equivalence classes is that as the number of members increase, so too does the number of emergent matching responses. For example, training two relationships (A to B and B to C) produces a three member class with a further seven matching relationships emerging (A to A, B to B, C to C, B to A, C to B, A to C, C to A, a total of 3² - 2 = 7). But, by training 4 relationships (5 member classes) we get 21 matching relationships occurring (5² - 4 = 21). In real life, stimulus equivalence classes, that is those that enter into relationships with human behaviour may have very many members with a huge number of emerging matching responses. This fact, along with the account of ‘generalisation’ between physically similar stimuli insinuates the potentially vast capacity of human learning and behaviour. Fields, Adams, Yang, Bufferman and Verhave (1996) demonstrated the transfer of a response from one member of an equivalence class to a wide range of stimuli that were physically similar to another member of that equivalence class. A class was established which consisted of two nonsense syllables, A1 and B1, and one line length, C1. Subsequently, a response trained to A1 was found to also be controlled by not only B1 and C1, but also by lines quite different in length to C1.
Similar mechanisms might be involved that control complex aspects of human behaviour. For example, if we are reinforced to speak to a same sex friend in a certain way, then we might also adopt the same approach when speaking to any new same sex friends we meet without a history of reinforcement for talking to that person in that way. Clearly, this is a very powerful method of extending ones behavioural repertoire. Similarly, this could contribute to less desirable outcomes by attaching fear or anxious behaviour to specific situations that were not previously feared. For example, a child could have a conditioned fear of water from a distressing incident in a swimming pool. It is possible that their fear related behaviour could come under control of all members of this equivalence class such as the beach, the bath or even the smell of chlorine. As we can see both operant responses and classically conditioned responses can be transferred across members of stimulus equivalence classes and those classes can in turn have many members, the phenomena of stimulus equivalence vastly increase the explanatory power of behavioural analysis.