RELATIONAL FRAME THEORY AND STIMULUS EQUIVALENCE ARE FUNDAMENTALLY
  DIFFERENT: A REPLY TO SAUNDERS' COMMENTARY

Reading an anonymous, negative review often generates at least a few
moans of frustration from the submitting author(s). Although most
negative reviews do provide information and advice that may well be
useful, they usually also contain comments that the author(s) would like
to debate with the reviewer. Such debate, however, is clearly difficult
if the reviewer wishes to remain anonymous. Fortunately, in this
particular case, our negative reviewer (Saunders, 1996) has "gone
public," for which we are most grateful. We hope his decision to do so,
and the subsequent exchange of views, will provide for the reader even a
small part of the learning experience and the sheer enjoyment that we
gleaned from the exercise. To help the reader follow the debate, (1) our
reply will work through what we feel are the key points raised by
Saunders in the same order in which they appeared in his commentary, and
(2) we will identify the relevant parts of his commentary by section and
paragraph number.

First Section (no heading)

Paragraph 1: Saunders states: "stimulus equivalence and relational frame
theory continue to be perceived as antagonistic positions . . . I am of
the firm opinion that synthesis is possible."

Should we take this to mean that Saunders himself intends to develop yet
another account of stimulus equivalence that combines the mathematical
treatment of behavioral equivalence with relational frame theory (RFT),
or is he simply making a prediction that the two accounts will
eventually merge into one? As we shall see, this is an important
question for some of the points that follow in Saunders' commentary.
Furthermore, contrary to Saunders, we are of the firm opinion that RFT
and "Sidman Equivalence" are fundamentally different, and thus a
synthesis between the two is highly unlikely, if not impossible. We
shall return to this issue at various points in our reply.

Paragraph 2: Saunders correctly states that we embraced RFT because it
attempts "to place equivalence in the broader context of all types of
relations:'

Although we did not state it explicitly, we have additional reasons for
adopting the relational frame account. These include, for example: (i)
its focus on the subject's history of reinforcement as part of the
explanation for arbitrarily applicable relational responding, (ii) the
fact that the transformation of function is a defining feature of a
relational frame, and (iii) the fact that RFT draws heavily on the
concept of the operant (we shall return to each of these issues at
appropriate points in our reply).

Relational Frame Theory

Paragraph 1: Towards the end of this paragraph Saunders states: "The
frame is the particular relation between the third and fourth term
implied by the current response to them, and which frame is operative in
a given instance is determined by the particular fifth term that is
present."

Although it may sometimes be useful to describe arbitrarily applicable
relational responding using the concept of five-term contingency control
(e.g., when describing a particular experimental procedure), it is very
important to recognize that a relational frame is in fact conceptualized
as a three-term contingency. The relational frame, as an analytic unit,
is therefore fundamentally different from the concept of a five-term
contingency, as defined by Saunders. For RFT, the contextual cue is the
third term, the relational response (e.g., responding to stimulus B in
terms of stimulus A and responding to A in terms of B) is the second
term, and the history of differential reinforcement correlated with the
contextual cue is the first term in the contingency. From this
perspective, responding to B given A and to A given B, for example, is
considered to be a single response unit controlled by the relevant
contextual cue (or cues) by virtue of its previous correlation with
differential reinforcement. In effect, the RFT approach invokes a purely
functional concept of an operant, and the term "overarching operant
class" (e.g., Barnes, 1994, 1996; Hayes, 1992) is used to remind the
reader of this fact. Hence, the RFT approach does not appear to involve
breaking with any important conceptual or technical practices within
behavior analysis (compare with Catania, 1992, pp. 148150 on the issue
of learning set).

Interestingly, other researchers have also developed a three-term
contingency account of stimulus equivalence (Stromer, McIlvane, & Serna,
1993) by conceptualizing the sample and comparison stimuli as separable
compounds (i.e., the sample and comparison stimuli function together as
the third term in the contingency). In much the same way that relational
frame theory invokes the term overarching to distinguish a purely
functional from a topographical concept of the operant, Stromer et al.
invoke the term separable to distinguish their definition from the
traditional concept of the compound stimulus. Thus, in our opinion, if a
synthesis is possible, the most likely partner for relational frame
theory will be the separable compound account of stimulus equivalence,
and not the fundamentally different five-term contingency analysis
offered by Saunders.

Paragraph 2: Saunders states: "Thus, relational frames may be said to
refer to instances of 'instructed' relational responding. The
instructional or contextual cues need not be linguistic symbols as long
as the cue is in an equivalence class with or substitutable for the
pertinent linguistic symbols, and vice versa."

We were somewhat puzzled by Saunders' use of the term "instructed" to
refer to the function of the contextual cues. Perhaps instructed (with
its linguistic connotations) is used to highlight his suggestion that
the cue must be either a linguistic symbol, or in an equivalence class
with (or substitutable for) such a symbol. In other words, is Saunders
suggesting that the function of a contextual cue is dependent upon some
special property of a linguistic symbol? If he is, this is a
misrepresentation of RFT. Relational frame theorists define a linguistic
symbol as a stimulus or event that participates in a relational
frame,[1] and thus to avoid circularity the term linguistic cannot be
used as a technical explanation for arbitrarily applicable relational
responding. From the RFT perspective, therefore, it should be possible
to take nonlinguistic symbols, such as red and green cards, and
establish them as contextual cues for responding in accordance with the
arbitrarily applicable relations of larger and smaller, for example,
without the cues participating in equivalence relations with any other
pertinent linguistic symbols (i.e., the subject comes to respond in
accordance with larger and smaller relations only in the presence of red
and green cards, respectively). To achieve this end, however, the
subject would likely require: (a) pretraining in the relevant
nonarbitrarily applicable relations (e.g., reinforcement contingent upon
picking the larger of two objects when presented with the red card, and
reinforcement contingent upon picking the smaller of two objects when
presented with the green card), and (b) a history of explicit
reinforcement for responding in accordance with the arbitrarily
applicable relations of larger and smaller [e.g., reinforcement
contingent upon picking the arbitrary stimulus X instead of the
arbitrary stimulus Y when presented with the red card (i.e., X is
arbitrarily defined as larger than Y) and reinforcement contingent upon
picking Y instead of X when presented with the green card (i.e., Y is
smaller than X)]. At the present time, of course, the exact nature and
extent of the training necessary for the red and green cards to function
as contextual cues for arbitrarily applicable relational responding must
remain an empirical issue (see Barnes, 1996).

Of course, when verbally sophisticated subjects are exposed to the
nonarbitrarily applicable relational pretraining employed by Steele and
Hayes (1991) and Roche and Barnes (1996), it seems likely that the
contextual cues will come to participate in equivalence relations with
previously established contextual cues or linguistic symbols (e.g.,
"same," "opposite," "larger," "smaller," etc.). In effect, the
preexperimentally established functions of these linguistic symbols,
including their control over arbitrarily applicable relational
responding, will transfer to the experimental contextual cues.
Consequently, a history of explicit reinforcement for responding in
accordance with the arbitrarily applicable relations defined by the
experimental cues is not necessary. In fact, this is an explicit
assumption for RFT; that a subject's preexperimental exposure to the
contingencies operating in his or her verbal community will play a
significant role in producing arbitrarily applicable relational
responding. When working with a verbally sophisticated subject,
therefore, the nonarbitrarily applicable relational pretraining is
designed to bring the subject's preexperimental history to bear on the
experimental task in a relatively precise way. In fact, if the
relational pretraining did not readily produce arbitrarily applicable
relational responding in verbally sophisticated adults, RFT would be in
very serious trouble as an account of human verbal behavior.

A Review of Roche and Barnes

Replication of Steele and Hayes (1991). Paragraph 2: Saunders states:
"The elaborate establishing procedures for the contextual cues in both
experiments served only to establish new symbols as members of three
existing classes containing other symbols for the respective relations."

This is certainly very likely, and as stated previously is entirely
consistent with RFT.

Saunders also states in this paragraph; "All of the foregoing in both
experiments [Steele & Hayes, 1991 and Roche & Barnes, 1996] could have
been replaced by verbally instructing the subjects regarding the meaning
of the new symbols or merely using the English words to label the frames
to be applied in the subsequent tests. Using the English words should
not have affected whether the relevant relations were or were not applied."

We certainly could have used English words or instructions, and perhaps
this strategy would have produced similar results. In fact, Steele and
Hayes (1991,p. 554) suggested this very possibility with regard to their
own study. Why then, did we use the relatively complex pretraining
procedure in the Roche and Barnes study?

First, the pretraining procedure provides the experimenter with more
precise knowledge about the history of the subject. Although it seems
likely that subjects will have been taught to use words such as "same,"
"different," and "opposite" in similar ways, the possibility that
slightly different histories have become attached to these words is
always possible. The Steele and Hayes pratraining procedure is one way
in which to avoid this problem.

Second, the pratraining procedure could also be used with nonhuman
subjects. Although relational frame theorists are primarily interested
in human verbal behavior, nonhuman research may allow us to analyze
arbitrarily applicable relational responding at a level that would be
very difficult, if not unethical, using only human subjects. Thus, it is
important that the general procedures used in RFT studies can be readily
adapted for use with nonhumans; using "real" words as contextual cues
would clearly make this very difficult.

Third, using the pretraining procedure allowed us to test a key
assumption of RFT; that human language typically involves arbitrarily
applicable relational responding, which is closely related to
nonarbitrarily applicable relational responding (see Hayes, 1994, pp.
24-25 and see paragraph 3 from the Discussion in Roche and Barnes,
1996). In effect, RFT predicts that if a verbally-able subject is
provided with a particular history of nonarbitrarily applicable
relational responding (e.g., the Steele and Hayes pretraining
procedure), it should be possible to use this history to produce
predictable responding to words that occur in the subject's natural
language. We wished to test this prediction, and thus we employed the
Steele and Hayes pretraining procedure. If we had used "real" words as
contextual cues, however, we could not be certain that the words had
acquired their functions, at least in part, from a history of
nonarbitrarily applicable relational responding, and thus we would have
been left with a relatively weak test for RFT.

Tests of sexual categorization. Paragraph 1: Saunders states: "Clearly,
the data show the test performances we could expect from sophisticated
adult subjects."

Funnily enough, when we first began the study we (and others we talked
to) did not expect the very clear results that eventually emerged.
Contrary to Saunders, therefore, we would argue that the Roche and
Barnes data are clear testimony to the value of RFT as a theoretical
account, insofar as it allowed us to achieve prediction and control over
adult human sexual categorization in a laboratory setting.

In the same paragraph, Saunders states: "Thus, responding on a test
trial was not a matter of which choice was correct, but rather which
choice was the best given the available alternatives and the constraints
of the task."

We challenge Saunders to find a situation in which this is not the case
(even social psychologists are aware that responses to questionnaires
may be affected by the type of scale used and by social desirability
factors). In effect, from a behavior analytic perspective the "correct
choice" is always determined by the historical and current contexts, and
to suggest otherwise represents an attack on the epistemology of
behavior analysis.

At the end of the next paragraph Saunders also states: "Further, a
training history in an equivalence study that sets the occasion for the
subject to partition the stimuli into two classes of equal size . . .
may be desirable if such equal composition is what should be engendered
by the training. In the Roche and Barnes study, the pretraining may set
the occasion for this same type of partitioning, a tendency that could
override a response pattern more nearly reflecting the socially derived
set of classes -- a set of unequal composition perhaps."

As behavior analysts, we certainly agree that the small number of words
used in our study, combined with the training and testing procedures,
helped to produce the response patterns obtained from our subjects.
Indeed, how could it be otherwise? We were somewhat worried, however, by
Saunders' suggestion that the subject's response patterns did not
reflect the socially derived classes. Insofar as any subject's social
behavior will change as a function of context, how could a subject's
response pattern ever reflect the socially derived set of classes? We
argue, therefore, that the response pattern obtained for each subject in
the Roche and Barnes study did reflect the socially derived relations
for that subject in that experimental context. Of course, further
experimental analyses could and should be conducted to increase the
precision and scope of our analyses. Further precision would be gained,
for instance, if we could identify a means of changing the subject's
response patterns in the experimental context (e.g., perhaps by
providing additional words to choose from during the categorization
tests). Additional scope would be gained if we could use the empirical
and conceptual analyses generated by the Roche and Barnes study to
obtain prediction and control in different settings. In fact, additional
scope has been obtained in a recent series of experiments in which the
Roche and Barnes analyses were successfully adapted to explore
children's perception of bullying and violent abuse (Vaughan & Barnes,
1994). Furthermore, as an extension of this work we are planning to
develop a relational-frame procedure for "diagnosing" sexual abuse in
young children.

Paragraph 3: Saunders states: "Roche and Barnes suggest that
questionnaires and scales would fail to help identify the processes
underlying sexual categorization that the current experiment purports to
achieve, but it is not evident how the current experiment addresses
underlying processes either."

The Roche and Barnes study helped to identify behavioral processes
insofar as we (a) observed patterns of relational responding for each
subject with regard to the words PENIS, VAGINA, DOMINATE, SUBMIT,
FORGET, and AMNESIA and (b) successfully expanded these patterns in
accordance with sameness, difference, and opposition using the novel
nonsense syllables X1, X2, and X3, in the absence of explicit
reinforcement during the relational tests, and without relying upon the
physical properties of the stimuli (e.g., the stimulus X1 was not
designed to look like a penis). In effect, the study demonstrated that
categorizing behavior in human adults may be usefully conceptualized as
a process of arbitrarily applicable relational responding.[2] Of course
the study has its limitations, not least of which is the fact that the
preexperimental histories of our subjects had presumably established a
multitude of relational frames. A more complete study of the behavioral
process of relational framing, therefore, would require working with
young children, or perhaps even nonhuman animals, so that the relational
frames could be "built from the ground up" (see final paragraph of the
Discussion in Roche & Barnes, 1996).

At the bottom of paragraph 3 Saunders states: "Presumptively, one might
conclude that responding 'accurately' to the contextual stimuli is
dependent on the preexisting 'equivalence' or other association of Penis
and Dominate and of Vagina and Submit. If they are not so classified or
similarly classified, control by SAME and OPPOSITE cannot be consistent.
This is a strong indication that, in this experiment, equivalence was a
prerequisite to responding to same, opposite, different (i.e.,
relational framing), rather than merely an included concept."

We agree with Saunders on this point; an explicit assumption in the
Roche and Barnes study was that preexperimentally, socially established
equivalence relations were required if the predicted patterns of
relational responding were to emerge during the sexual categorization
tests (e.g., see penultimate paragraph of the Introduction from Roche &
Barnes, 1996). We should also point out, that from the RFT perspective
equivalence is considered to be "the most common and fundamental type of
verbal construction." (Hayes, 1991, p. 32). As outlined by Barnes
(1994,pp. 102-103); "Hayes views equivalence as the most important
relational frame because it is central to the occurrence of referential
relations in natural language . . . and is therefore necessary before
other derived relations can be readily learned. It would be quite
difficult, for example, to teach a child that "hot" is the opposite of
"cold," if the words "hot" and "cold" did not participate in equivalence
relations with the physical events of hot and cold respectively (see
Hayes & Hayes, 1989, pp. 174-175)." Thus, although equivalence is
considered to be one of a number of relational frames, in practice, it
is often more than "merely an included concept;'

Expansion of sexual categories: At the end of the only paragraph in this
section, Saunders states: "The authors [Roche & Barnes] state that the
potential relationships established by these procedures were derived
through the relational frames of coordination, difference, and
opposition, respectively. More simply, perhaps, the subjects were
"instructed" via match-to-sample procedures that S1, X1, and Dominate
are substitutable and all else follows because Dominate is more related
to Penis than Vagina, in this context."

We have already discussed our difficulty with Saunders' use of the term
"instructed." We were also somewhat perplexed by his use of the phrases
"all else follows" and "more related." In one sense of course, it is a
great deal simpler to use these phrases rather than the vocabulary of
RFT. Nevertheless, "all else follows" and "more related" are not
technical terms in behavior analysis, and thus they cannot be used to
describe, at a technical level, the behavioral patterns that were
obtained in the Roche and Barnes study (e.g., when Saunders suggests
that Dominate is "more related" to Penis than Vagina, what exactly are
the defining behavioral properties of this "more related" relation?).
The following analogy may help clarify the point we are making. Although
it is a great deal simpler to say that a rat presses the lever because
it wants the food (instead of using the language of operant classes),
just like "all else follows" and "more related," "wanting food" is not a
technical term in behavior analysis.

Extension of Steele and Hayes (1991), Paragraphs 1 and 2: Towards the
end of the first paragraph Saunders states: "when a novel alternative is
offered and the subject is told to consider its selection, some subjects
are likely to perceive that there must be some occasions when it is to
be used, particularly when the subjects are sophisticated, verbal
individuals. Thus, the specific reasons why subjects responded to the
question mark can only be speculated." Furthermore, after outlining the
data obtained from Subject 5, Saunders states: "These predictions are
precisely confirmed by the data, but the explanations are nonetheless
speculative." (Saunders makes a similar point at the end of the section
entitled "Stimulus Equivalence and the Frame of Coordination").

The results from the Roche and Barnes study (in particular the pattern
of question mark choices) indicate three distinct patterns of relational
responding (see fifth paragraph of the Discussion in Roche and Barnes).
Insofar as RFT provided a means of predicting and controlling these
response patterns in the laboratory setting, we fail to see how the
current RFT explanation is so speculative. Of course, human sexual
categorization could, and indeed should be subjected to further
experimental analysis. As suggested earlier, further analysis may add
both precision and scope to the RFT explanation. Indeed, one of the
purposes behind submitting the Roche and Barnes study was to provide
other researchers with the opportunity to conduct further analyses along
similar lines (we presume this is normally one of the goals behind any
empirical report: to allow others to replicate, to extend, or to refute
the initial work). Saunders, of course, would likely argue that our
decision to submit the current data was premature. We, in contrast, felt
that the results were sufficiently interesting to warrant publication.

Transformation of function via relational frames. Paragraph 1: Saunders
states: "The data do not show that the preexperimentally established
social functions (of Penis, Dominate, etc.) can be 'transformed in
accordance with the three relational frames of sameness, opposition, and
difference.' "

We are not certain why Saunders disagreed with us here, but the most
likely reason is that he and the current authors are using
'transformation of function' in different ways. Because Saunders does
not specify what, if anything, constitutes a transformation of function
from his perspective, it is difficult to take this issue any further.
Nevertheless, in the interests of clarity we will explain in detail
exactly what we mean by 'transformation of function.'

From the RFT perspective all instances of arbitrarily applicable
relational responding involve to some extent a transformation of
function (sometimes referred to as 'transfer of function'; see Dymond &
Barnes, 1995). Take, for example, the relatively simple case of
symmetry. During the conditional discrimination training choosing B1 in
the presence of A1 is differentially reinforced, and then during the
test choosing A1 in the presence of B1 emerges without explicit
reinforcement. From the RFT perspective, the explicitly reinforced
function of A1 (as a sample stimulus with regard to B1) and the
explicitly reinforced function of B1 (as a comparison stimulus with
regard to A1) have been transformed in accordance with the mutually
entailed relation of symmetry (i.e., the sample function of A1 has been
transformed into a comparison function and the comparison function of B1
has been transformed into a sample function). In effect, the
transformation of function is a defining characteristic of a relational
frame.

The reader should also note that RFT normally makes a distinction
between entailment relations (e.g., symmetry, equivalence, opposition,
more-than, etc.) and a transformation of functions because functions may
transform in a wide variety of patterns, and it is scientifically useful
to discriminate these patterns from each other in a relatively
consistent way. These various patterns of a transformation of function
are usually categorized as instances of the mutually and combinatorially
entailed relations of coordination, opposition, comparison, and so on.
In effect, the observed pattern of a transformation of functions defines
the entailed relations, and thus entailed relations do not exist as a
behavioral event until a specific transformation of function has
occurred (e.g., a transformation of sample and comparison functions on a
"standard" equivalence test).

Having briefly outlined the RFT view of a transformation of function we
will now describe, for illustrative purposes, two ways in which the data
from the Roche and Barnes study demonstrated a transformation of the
socially established functions for Penis, Dominate, Vagina, and Submit.
First, training subjects to choose X1 and Dominate when presented with
S1 in the presence of SAME transformed the social functions of Dominate
and Penis in accordance with sameness. In effect, X1 was added to the
socially established equivalence relation between Dominate and Penis,
and this transformation of functions was shown when Penis controlled
choosing Xl in the presence of SAME. Second, it follows that training
subjects to choose X3 and Submit when presented with S1 in the presence
of OPPOSITE transformed some of the socially established functions of
Submit, Vagina, Dominate, and Penis in accordance with sameness and
opposite. In effect, X3 was added to the socially established
equivalence relation between Submit and Vagina, and X1 was added to the
socially established opposite relation between Penis/Dominate and
Vagina/Submit; these transformations of functions were shown when Vagina
controlled choosing X3 in the presence of SAME and controlled choosing
X1 in the presence of OPPOSITE.

It can be seen from the foregoing that the term "transformation of
function" is generic to the concept of the relational frame, and that
relational terms such as symmetry, equivalence, opposite, and so forth
are used to discriminate the various patterns of a transformation of
function that are observed by the behavioral scientist.

Also contained in the first paragraph under the heading, "Transformation
of function via relational frames" Saunders states: "The match-to-sample
training with S1 and Dominate and X1 would have added X1 and S1 to the
class with Penis and Dominate. It is difficult to see how S1 could be a
member of the class containing X2 or the class containing X3, as
suggested by Roche and Barnes. It is equally difficult to understand how
the contextual stimuli control three distinct classes of stimuli,
respectively."

We too had great difficulty understanding the foregoing interpretation
in terms of contextually controlled equivalence classes. Nevertheless,
just such an interpretation was offered by someone who read a much
earlier version of the manuscript, and thus we felt it necessary to
explain clearly how this account was untenable. In fact, much of the
Discussion from the final version of the Roche and Barnes paper was
written in response to earlier reviews that contained, we felt, badly
formulated or incomplete explanations of our data in terms of
equivalence classes alone. In effect, the equivalence account was not
ours, it was someone else's.

Also contained in the same paragraph is the statement "it seems more
accurate to suggest that the contextual stimuli should be conceptualized
as having the function of altering the conditional control of the sample
stimuli. That is, the function of a fifth term in a match-to-sample
contingency is to specify which of two or more functions of a fourth
term (the sample stimulus) is operative on a given trial."

We agree in broad terms with the first sentence, except that we use the
term "transforming" instead of "altering," and we say that the functions
of the sample and comparison stimuli are mutually transformed in the
presence of the contextual cue (i.e., for RFT, the mutual transformation
of functions is a single response unit). With regard to the second
sentence, however, we take issue with Saunders' use of "fifth term" and
"fourth term" to describe the contextual cue and sample stimuli,
respectively. As suggested earlier, the language of four- and five-term
contingencies may be useful in describing experimental procedures, but
from the RFT perspective arbitrarily applicable relational responding is
conceptualized as occurring within a three-term contingency.

Paragraph 2: Saunders states "the relational frame explanation is no
different than the more established explanation: That the responding
reflects contextual control of responding to members of mutually
exclusive equivalence classes."

The first question we asked having read this was: Which established
explanation? Even if we assume that Saunders is referring to the
mathematical treatment offered by Murray Sidman, to which version of
this account is he referring; that offered in Sidman (1986,1990) or in
Sidman (1994)? The more recent version openly revises the earlier work
in a number of very important areas. For example, equivalence is no
longer seen as emerging only at the level of the four-term contingency,
and functional and equivalence classes are now viewed as synonymous
rather than as separate but interacting processes. Sidman also extends
the concept of equivalence in a way that may alarm at least some
researchers, particularly those who have not been involved in the
equivalence research program. For example, the concept of stimulus
equivalence is used to account for respondent conditioning and
conditioned reinforcement, and Sidman even suggests that his treatment
removes the need to distinguish between stimuli and responses. Clearly,
therefore, we have two versions of Sidman's treatment of stimulus
equivalence from which to choose, and Saunders does not clearly indicate
which one he considers to be the established version. This issue also
raises certain questions about Saunders' statement at the beginning of
his commentary that a synthesis between RFT and Sidman equivalence is
possible (i.e., which version of Sidman's account does Saunders' think
might be combined with RFT?).

Furthermore, even if we accept, for the purposes of discussion, that
Sidman (1994) is the "established" account, we are troubled by his
rather terse statement "the fact that a stimulus pair can be brought via
contextual control into such differing relations as same, opposite,
different, and so forth, can be handled by any formulation of
equivalence that recognizes the role of context" (p. 561). Our main
problem with this statement is as follows. Defining any relation between
a stimulus pair in terms of a contextually controlled equivalence class
seems to undermine the functional utility provided by the definition of
equivalence in terms of reflexivity, symmetry, and transitivity. In
effect, according to Sidman and Saunders the three defining properties
of an equivalence class do not seem to apply to contextually controlled
equivalence classes. Consider, for example, a subject who is trained to
choose B in the presence of A in "a larger than" context. Presumably
this subject should not then choose in the same context, A given B
(i.e., A is in fact smaller than B) or A given A or B given B (a
stimulus cannot be larger than itself). In this particular example,
therefore, we are left with neither symmetry nor reflexivity as defining
properties of the subject's pattern of responding to the stimulus pair!
In other words, the term contextually controlled equivalence class is
not restricted to stimuli that participate in arbitrarily applicable
sameness relations (i.e., the RFT definition), but is used to describe
any stimuli that simply "go together" in a particular context. We are
left, therefore, with no specific pattern of relational responding that
can be isolated as an example of a contextually controlled equivalence
class.

Although Sidman and Saunders are free to propose this "going together"
definition, as stated above we are of the firm opinion that it seriously
weakens the functional utility of the concept of an equivalence class.
Perhaps Saunders is aware of this problem, and that is why he suggests
we use the term equivalence class in two distinct ways (see next
section). Specifically, he makes a distinction between "arbitrarily
established equivalence" which is presumably defined by the properties
of reflexivity, symmetry, and transitivity, and "contextually controlled
equivalence" which need not be defined by these properties. In effect,
"arbitrarily established equivalence" maintains functional utility.
Nevertheless, the need to distinguish between two types of equivalence
represents, for us, a serious and unnecessary breach of parsimony, that
will most likely serve only to confuse. In the interests of both
parsimony and clarity, therefore, we prefer to reserve the term
equivalence for those patterns of relational responding that accord with
reflexivity, symmetry, and transitivity, and use different labels to
discriminate those patterns that do not accord with these three
properties. This, of course, is entirely consistent with RFT.

Stimulus Equivalence and the Frame of Coordination

In this section, at the beginning of the next (i.e., Implications for
Roche and Barnes, paragraph 1), and in Footnote 2 Saunders makes a
distinction between "contextually controlled equivalence" and
"arbitrarily established equivalence." We believe that this distinction
has important implications for Saunders' opinion that a synthesis
between RFT and stimulus equivalence is possible. We shall now consider
these implications, and a number of related issues, in some detail.

Saunders states: "Equivalence, arising as a function of the
contingencies of reinforcement among otherwise unrelated third- and
fourth-term stimuli is an outcome not addressed or explained by
relational frame theory."

From the RFT perspective, however, the matching-to-sample format itself,
the verbal instructions provided to the subjects, and a myriad of other
possible stimuli may function as the relevant contextual cues for any
instance of equivalence responding (see Barnes, 1994). In effect,
equivalence is always under some form of contextual control,[3] and thus
for RFT "arbitrarily established equivalence" is in fact contextually
controlled equivalence (i.e., it is an example of the frame of
coordination). Clearly, therefore, Saunders' distinction between two
types of equivalence is inconsistent with RFT.

Saunders also argues that arbitrarily established equivalence is
"derived . . . from reinforcement for arbitrary responding," whereas
equivalence as a case of the frame of coordination is derived "by
reinforcement for perceptually based responding."

From the RFT perspective, however, equivalence responding is normally
based on a history of explicit reinforcement for nonarbitrarily
applicable (i.e., perceptually based) and arbitrarily applicable
sameness responding.[4] Allow us to explain. The nonarbitrarily
applicable sameness responding is normally established when a child
learns to categorize physically similar objects together (e.g., pushing
apples together in one pile and oranges in another; see Stromer &
Mackay, in press, on the issue of feature stimulus classes). This
history may then be brought to bear on a matching-to-sample task if
picking a comparison in the presence of the sample evokes a response
such as "this same as that" (i.e., the matching-to-sample format itself
functions as a contextual cue for reflexivity). Although this type of
reinforcement history may be sufficient to produce reflexivity in a
matching-to-sample context, it seems unlikely that this history alone
would produce responding in accordance with symmetry and transitivity
(i.e., arbitrarily applicable sameness responding).[5] Explicitly
reinforced equivalence responding, such as that provided when learning
the names for objects and events in the world, is most likely needed for
matching-to-sample equivalence[6] (see Barnes, 1994, pp. 95-97; Hayes,
1992). When such a naming history is absent, however, an explicit
history of reinforcement for matching A-B, B-A, A-C, C-A, B-C, and C-B,
for example, may produce equivalence responding with a novel set of
stimuli (see Footnote 2 from Saunders commentary). From the RFT
perspective, the naming and matching histories are functionally similar,
insofar as they both explicitly reinforce responding in accordance with
arbitrarily applicable sameness relations.

The important point to be gleaned from the foregoing is that for RFT the
relational frame of equivalence is defined, in part, by a history of
reinforced nonarbitrarily applicable sameness responding (i.e.,
reflexivity). Thus, what Saunders calls "arbitrarily established
equivalence" is in fact (for RFT) an example of the frame of
coordination because it incorporates a history of perceptually based
responding which manifests itself in the test for reflexivity. Saunders'
distinction between two types of equivalence is again clearly
inconsistent with RFT.

As an aside, one empirical question that could arise at this point is as
follows. If a subject was provided with a history of reinforcement for
responding in accordance with symmetry and transitivity, but not for
reflexivity, would that subject demonstrate symmetry and transitivity
without reflexivity on a subsequent test with a novel set of stimuli? If
reflexivity failed to emerge, then for RFT the pattern of responding
could not be categorized as equivalence because all three defining
properties would not be present. Thus, another type of relational frame
would be operating. What if reflexivity did emerge, however? From the
RFT perspective, reflexivity might emerge following an explicit history
of reinforcement for symmetry and transitivity alone because such a
reinforcement history likely establishes indirect reflexivity (see
Barnes, 1994). Consider, for example, a subject who is trained to choose
B in the presence of A and to choose A in the presence of B. For this
subject each stimulus comes to actualize the perceptual functions of the
other stimulus. Upon seeing A as a sample, for instance, the subject
will also "see" B. and thus choosing B as a comparison is based to some
degree on indirect reflexivity (i.e., see A arrow right "see" B arrow
right choose B). Similarly, seeing B as a sample will actualize the
perceptual functions of A and thus choosing A also involves indirect
reflexivity (i.e., see Barrow right "see" A arrow right choose A). In
effect, explicitly reinforcing A-B and B-A matching may be functionally
similar to explicitly reinforcing A-A and B-B matching, and thus during
a subsequent test reflexivity might emerge without an explicit history
of reinforcement for nonarbitrarily applicable sameness responding. In
any event, resolving this empirical issue will clearly require some very
precise and delicate analyses by those researchers who specialize in
nonhuman stimulus control.

Saunders' distinction between "arbitrarily established equivalence" and
"contextually controlled equivalence" also leads him to state:
"relational frames based on experience with perceptually based classes
can affect responding to arbitrary classes, but relational frames cannot
arise from experience with arbitrary classes alone."

Although RFT does not recognize Saunders' distinction between two types
of equivalence, this statement touches upon an important issue. As we
pointed out earlier, equivalence is considered to be the most common
type of relational responding (e.g., it is central to the occurrence of
referential relations in natural language, and is therefore necessary
before other types of nonequivalence relations can be easily learned).
Thus, RFT predicts that if specific cues for responding in accordance
with nonequivalence relations are not provided, it is quite likely that
a subject will "default" to the most common type of relational
responding.[7] Furthermore, as suggested previously, other factors such
as the verbal instructions, the matching-to-sample format, and the
equivalence test itself (e.g., reversing the sample and comparison
stimuli) may also function as contextual cues for equivalence (see
Barnes, 1994, for a detailed discussion). In short, the standard
matching-to-sample procedure produces equivalence, rather than any other
nonequivalence relation, because it readily brings to bear the subjects'
history of reinforced nonarbitrarily and arbitrarily applicable sameness
responding.

Interestingly, this view also provides an answer to the second question
asked by Saunders in his second footnote, in which he asks; "given a
novel A1 as sample, and novel B1 and B2 comparisons, where A1, B1, and
B2 share no perceptual characteristics in common, to which comparison
should the subject respond?"

From the RFT perspective, a subject with a history of reinforced
nonarbitrarily and arbitrarily applicable sameness responding will most
likely choose the comparison stimulus that is deemed to be the same as
the sample on some idiosyncratic, nonarbitrary basis (e.g., if abstract
forms are used, perhaps the sample and chosen comparison both remind the
subject of a particular object or event; if nonsense syllables are used,
the subject may choose a particular comparison because its first letter
is closer in the alphabet to the first letter of the sample stimulus)
(see Hayes, 1994, p. 24-25, for a discussion of the nonarbitrary nature
of arbitrarily applicable relational responding).

The foregoing arguments clearly highlight some of the fundamental
differences between RFT and the version of stimulus equivalence proposed
by Saunders. In light of these differences, we fail to see how a future
synthesis of the two accounts will ever be possible.

Implications for Roche and Barnes, paragraph 2: In the latter half of
this paragraph Saunders repeats many of his earlier criticisms of the
Roche and Barnes study (i.e., the number of words used was too small;
paper-and-pencil tests would have been an easier alternative; the test
performances of the subjects reflect classes that were likely different
than those perceived by Roche and Barnes; the study does not distinguish
relational frames from other interpretations). We have addressed each of
these criticisms in detail at various points throughout the reply, so we
will not repeat ourselves here. In the next paragraph in this section,
however, Saunders does offer three additional comments in response to
our suggestion that the Roche and Barnes study provides a means by which
we can model the development of sexual categorization. We will deal with
each of these comments in turn.

First, Saunders states: "If the two important classes of stimuli in
their [Roche and Barnes] study had included military terms . . . would
we have a laboratory model of military rank? Well, yes, but is it an
explanatory model of categorization or merely an instance or example of
categorization?"

Insofar as we controlled the categorization of novel stimuli (X1, X2,
and x3) into already existing social categories using procedures and
concepts consistent with a behavior-analytic account, it is indeed an
explanatory model. Furthermore, recent relational frame research at the
Cork laboratory on children's perception of bullying and violent abuse
(Vaughan & Barnes, 1994) is helping to extend the scope and precision of
the analyses presented in the Roche and Barnes study.

Second, Saunders states: "Would the results have suggested that
relational frame theory provides a behavior-analytic interpretation of
military rank? Well, yes, but not at a level of understanding that we
did not possess before?"

The results were entirely consistent with RFT (an already existing
behavior-analytic account) and thus the data could not provide a new
understanding. Of course, if the study had failed to support RFT, then
perhaps a new understanding might have emerged. In effect, the Roche and
Barnes study represents one way in which to do science (we shall return
to this issue in the final section of our reply).

Third, Saunders states: "Would the results have had an important impact
on discipline in the military which is dependent upon such
categorization. Well, no! That, I believe, is the litmus test of this
type of research."

The Roche and Barnes study was clearly an example of basic not applied
research, and thus the litmus test suggested by Saunders may be seen as
somewhat extreme. Nevertheless, it is important to remember that the
current study did provide an important foundation for more recent work
at the Cork laboratory which we hope will provide a procedure for
"diagnosing" sexual abuse in young children. Furthermore, a recent
relational-frame study (Biglan & Hayes, personal communication) found
that when employees were encouraged to categorize a chaotic work
environment as a creative environment their working behavior changed
significantly for the better. In effect, transforming the function of
the word "chaotic" by relating it via equivalence to "creative" had an
important impact in an industrial setting. Presumably, therefore,
similar measures could be taken in a military environment. Perhaps, for
example, discipline could be improved if during military training words
such as "Command" and "Brass" were related via equivalence to words such
as "reliable," "safe," and "experienced." Clearly, a great deal of basic
and applied research needs to be done, in a range of settings, to
determine whether RFT can obtain sufficient scope and precision and thus
be considered a "true" explanation for human behavior. An important part
of this research will obviously involve responding successfully to the
type of challenge offered by Saunders (i.e., to change behavior in
applied settings using the knowledge gained from basic RFT research).

Conclusion

In this section Saunders states: "One way to approach this work
[stimulus equivalence and relational frames] is to design experiments
that are intended to reinforce our own way of thinking about the issues
while enhancing the relative significance of a particular theory. The
Roche and Barnes study is suggestive of that method. Unfortunately, this
method represents for me a trend in equivalence research that does poor
service to the field of equivalence as well as relational frame theory
-- an outcome neither area deserves. A more viable approach I believe is
to arrange for subjects to perform under carefully controlled
conditions, without prior expectations of how they surely must perform,
and conduct an open-minded analysis of the results. From such activity,
the path will be more clear."

There are many ways to do science, and each has its advantages and
drawbacks. One way is to conduct experiments that will either support or
fail to support a particular account or explanation for a given
phenomenon. The main advantage of this approach is that it can help to
develop a single parsimonious account of a wide range of phenomena, and
to help researchers see regularities and relations where none were seen
before (see Skinner, 1957, p. 282). This can have important implications
for applied issues (see Hayes & Wilson, 1993, for an example of how RFT
provides the conceptual underpinning of Acceptance and Commitment
Therapy). The disadvantage to this approach, however, is that it can
help researchers fail to see important regularities and relations that
are either irrelevant to their "pet" theory or directly contradict it.
We are of the firm opinion, however, that this disadvantage can be
completely avoided if theoretical analyses are used to achieve
prediction and control with increasing scope and precision. In effect,
we will not be "blinkered" by a particular theory, if the truth of the
theory is measured by the extent to which it helps us predict and
control behavior with increasing accuracy in increasingly diverse
situations. Once a theory no longer helps us achieve these goals, then
as behavior analysts we should discard the theory. The current authors
approach RFT in exactly this way.

Another way to do science is to conduct experimental analyses in a
supposedly atheoretical manner, and to "conduct an open-minded analysis
of the results." Parenthetically, open-mindedness in science is a myth
-- even in choosing to do a particular study one must adopt many
theoretical and empirical assumptions. The idea of the scientist as an
open-minded or passive observer is truly passe. We presume, therefore,
that Saunders is really advising us to design our studies and analyze
the results without deliberately attempting to support RFT or "Sidman
Equivalence." We would remind Saunders, however, that he is "of the firm
opinion" that a synthesis of these two accounts is possible, and thus we
question his open-mindedhess on this issue (perhaps his agenda is to
combine successfully these accounts, and produce "Saunders' Equivalence!").

Leaving aside the open-mindedness issue, there does appear to be an
approach to research in which no particular formal theory is used to
guide the experimental design and data analysis. The advantage of this
approach is that we are less likely to ignore perhaps important
questions and aspects of our data that are inconsistent with, or
irrelevant to, a particular theory (as argued above, however, this
problem can be avoided if we use a theory to achieve specific pragmatic
goals, and do not focus on trying to prove that the theory is "true" in
some nonpragmatic sense). The down side with this approach is that our
experimental and conceptual analyses may become increasingly disparate
and unconnected, as each researcher identifies and studies relatively
arcane and obscure phenomena with no concerted effort to increase the
scope of the analyses. We believe that the stimulus equivalence research
program may well fall victim to this problem.

Originally, stimulus equivalence was deemed interesting because it
appeared to offer a means of analyzing human language, both empirically
and conceptually. More recently, however, it seems to us that some
researchers have become interested in equivalence per se and have
forgotten why the phenomenon was considered important in the first
place. We are in danger, as a scientific community, of devoting
considerable time and energy to "understanding" the minutia of stimulus
equivalence (in the same way we tried to understand the minutia of
schedules of reinforcement). In twenty years time, therefore, we may
know a great deal about stimulus equivalence, but very little about
human language, or anything else for that matter. Who will benefit from
such a research program? Certainly not the lay community, and in the
long run neither will we. In our opinion, therefore, RFT should be used
as a conceptual and empirical tool to analyze human language, and we
should not devote our time to analyzing stimulus equivalence as a
phenomenon in its own right. Of course, Saunders may not agree with us,
and thus when he is asked to review one of our studies in which we
attempt to analyze human language using RFT, you may well hear a few
moans of frustration emanating from the Cork laboratory!

We thank Richard Saunders for participating in this exchange of views,
and we also express our gratitude to Steve Hayes for providing
invaluable feedback on an earlier version of this reply. Any
inaccuracies or errors, however, remain the responsibility of the
authors. Requests for reprints should be addressed to Dermot Barnes,
Department of Applied Psychology, University College Cork, Cork, Ireland.

   1. If we train a dog to approach the larger of two trees when we
      shout "big," but to approach the smaller tree when we shout
      "small," would these words function as linguistic symbols? From
      the RFT perspective they would not, because they do not
      participate in relational frames.
   2. The term process is used to describe a generic pattern of
      behavioral interactions extended in time and space (e.g., operant
      behavior, respondent behavior, stimulus generalization,
      arbitrarily applicable relational responding). The truth or
      falsity of these functional-analytic processes is determined by
      the extent to which their use promotes prediction and control with
      sufficient scope and precision in both basic and applied settings.
      The term process, as used here, does not therefore refer to a
      cognitive, biological, or hypothetical "process" or "mechanism"
      occurring in a nonbehavioral domain.
   3. In the same way that a rat's free-operant lever pressing is under
      the control of unspecified discriminative stimuli (e.g., the
      sight, smell, and sound of the operant chamber), equivalence
      responding is also under the control of unspecified contextual cues.
   4. From the RFT perspective nonarbitrarily applicable sameness
      responding (or reflexivity) is a background basis for all forms of
      relational responding. For instance, if an event when viewed as a
      sample was seen as different than when it was viewed as a
      comparison, no form of arbitrarily applicable relational
      responding could occur.
   5. A recent study by Schusterman and Kastak (1993) reported the case
      of a sea lion that had successfully been trained to produce
      generalized identity matching, but failed to show equivalence
      responding until the animal had been provided with an explicit
      history of reinforcement for responding in accordance with
      symmetry and transitivity.
   6. RFT recognizes that very limited forms of derived behavior may
      occur without a history of explicitly reinforced equivalence
      responding (see Hayes & Wilson, 1993, p. 286; see also Zentall &
      Urcuoli, 1993, for evidence that suggests derived behavior may
      occur in nonhumans). Nevertheless, these types of behavior are
      normally defined as largely respondent because they do not emerge
      from an appropriate history of arbitrarily applicable relational
      responding (note, this is a purely functional distinction).
      Respondent behavior is, however, considered to be an important
      foundation for relational framing (see Barnes, 1994, for a
      detailed discussion of this issue in terms of indirect reflexivity).
   7. In the first version of the Roche and Barnes (1996) paper that we
      submitted to The Psychological Record, we included data that
      showed that even subjects who had successfully responded in
      accordance with sameness, opposition, and difference relations,
      responded in accordance with equivalence relations alone when no
      contextual cue was presented during novel matching-to-sample tasks
      (i.e., the subjects appeared to "default" to equivalence responding).

References

BARNES, D. (1994). Stimulus equivalence and relational frame theory. The
Psychological Record, 44, 91-124.

BARNES, D. (1996). Naming as a technical term: Sacrificing behavior
analysis at the altar of popularity. Journal of the Experimental
Analysis of Behavior 65, 264-267.

CATANIA, A. C. (1992). Learning (3rd ed.). Englewood Cliffs, NJ:
Prentice-Hall.

DYMOND, S., & BARNES, D. (1995). A transformation of self-discrimination
response functions in accordance with the arbitrarily applicable
relations of sameness, more than, and less than. Journal of the
Experimental Analysis of Behavior, 64, 163-184.

HAYES, S.C. (1991). A relational control theory of stimulus equivalence.
In L J. Hayes & R N. Chase (Eds.), Dialogues on verbal behavior (pp.
19-40). Reno, NV: Context Press.

HAYES, S.C. (1992). Verbal relations, time, and suicide. In S.C. Hayes &
L. J. Hayes (Eds.), Understanding verbal relations (pp. 109-118). Reno,
NV: Context Press.

HAYES, S.C. (1994). Relational frame theory: A functional approach to
verbal events. In S.C. Hayes, L. J. Hayes, M. Sato, & K. Ono (Eds.),
Behavior analysis of language and cognition (pp. 9-30). Reno, NV:
Context Press.

HAYES, S.C., & HAYES, L. J. (1989). The verbal action of the listener as
the basis for rule-governance. In S.C. Hayes (Ed.), Rule-governed
behavior: Cognition. contingencies. and instructional control (pp.
153-190). New York: Plenum Press.

HAYES, S.C., & WILSON, K. (1993). Some applied implications of a
contemporary behavior-analytic account of verbal events. The Behavior
Analyst, 16, 283-301.

ROCHE, B., & BARNES, D. (1996). Arbitrarily applicable relational
responding and sexual categorization: A critical test of the derived
difference relation. The Psychological Record, 46, 451-475.

SAUNDERS, R. R. (1996). From review to commentary on Roche and Barnes:
Toward a better understanding of equivalence in the context of
relational frame theory. The Psychological Record, 46, 477-487.

SCHUSTERMAN, R. J., & KASTAK, D. (1993). A California sea lion (zalophus
californianus) is capable of forming equivalence relations. The
Psychological Record, 43, 823-839.

SIDMAN, M. (1986). Functional analysis of emergent and verbal classes.
In T. Thompson & M.D. Zeiler (Eds.), Analysis and integration of
behavioral units (pp. 213-245). Hillsdale, NJ: Erlbaum.

SIDMAN, M. (1990). Equivalence relations: Where do they come from? In D.
E. Blackman & H. Lejeune (Eds.), Behaviour analysis in theory and
practice: Contributions and controversies (pp. 93-114). Hove, England:
Erlbaum.

SIDMAN, M. (1994). Equivalence relations and behavior: A research story.
Boston, MA: Authors Cooperative, Inc.

SKINNER, B. F. (1957). Verbal behavior. New York: Appleton-Century-Crofts.

STEELE, D., & HAYES, S.C. (1991). Stimulus equivalence and arbitrarily
applicable relational responding. Journal of the Experimental Analysis
of Behavior, 56, 519-555.

STROMER, R., & MACKAY, H. A. (in press). Naming and the formation of
stimulus classes. In T. R. Zentall & P.M. Smeets (Eds.), Stimulus class
formation. Amsterdam: Elsevier.

STROMER, R., MCILVANE, W. J., & SERNA, R. W. (1993). Complex stimulus
control and equivalence. The Psychological Record, 43, 584-598.

VAUGHAN, M., & BARNES, D. (November, 1994). Children's perception of
pain and bullying: A relational-frame interpretation. Paper presented at
the Annual Conference of the Psychological Society of Ireland, Kilarney,
Ireland.

ZENTALL, T. R., & URCUIOLI, P. J. (1993). Emergent relations in the
formation of stimulus classes by pigeons. The Psychological Record,
43,795-810.

~~~~~~~~

By DERMOT BARNES, University College Cork, Ireland and BRYAN ROCHE,
University of Bath, England