Beruflich Dokumente
Kultur Dokumente
The basic phenomenon now called stimulus essary on the grounds that the particular re-
equivalence has had several incarnations in sponses (i.e., selecting stimuli) are only
modern psychology. In 1971 Sidman resur- differentiated with reference to the stimuli in-
rected the area by linking it to a powerful volved and thus add little to the explanation
procedure: matching to sample. Since then, be- of the derived stimulus relations themselves
havior-analytic research has focused on the (Sidman, 1986; Sidman, Cresson, & Willson-
limits of the equivalence phenomenon (e.g., Morris, 1974; Sidman & Tailby, 1982). Re-
populations, stimulus modality, stimulus ar- cently, response mediation models have
rangements), refinement of its measurement reemerged in the form of naming-based inter-
(e.g., Sidman & Tailby, 1982), and descrip- pretations (e.g., McIntire, Cleary, & Thomp-
tions of conditions under which it may arise son, 1987). The effects of naming, however,
(e.g., Sidman, 1986). Behavior analysts have also require a theoretical explanation. The
devoted little attention, however, to a theoret- proposed alternatives either do not yet have
ical account of equivalence. At present, stim- experimental support, as in McIntire et al.'s
ulus equivalence is merely the description of homogeneous chain model (Hayes, 1989; K.
a behavioral outcome-the process involved is Saunders, 1989), or themselves assume name-
unknown. object derived relations in order to explain sim-
Early researchers (e.g., Sidman & Cresson, ilar derived relations between samples and
1973; Spradlin, Cotter, & Baxley, 1973) ten- comparisons.
tatively discussed their work in terms of pos- A recent account views equivalence as the
sible response mediation, much along the lines result of relational responding arbitrarily ap-
of stimulus-response learning theory. Re- plied to the matching-to-sample situation
sponse mediation was later rejected as unnec- (Hayes, 1991; Hayes & Hayes, 1989). Or-
ganisms from insects to primates can learn to
respond to nonarbitrary relations among stim-
Correspondence and reprint requests may be addressed uli (e.g., larger than, darker than; see Reese,
to Steven C. Hayes, Department of Psychology, University 1968). These relations can be brought under
of Nevada-Reno, Reno, Nevada 89557-0062. contextual control and can generalize to new
519
520 DAVID STEELE and STEVEN C. HAYES
sets of formally related stimuli (Lowenkron, That is, in a context that brings a particular
1989). Verbal humans, however, seem also to kind of relational responding to bear (Crel.),
apply these kinds of relations when there are the designation of that relation between Events
contextual cues to do so without significant A and B through direct training entails (en-
regard for the form of the items being related. tailment is symbolized by I) a derived re-
The reader told that A = B > C will be able lation between B and A (rely) that may or may
to specify the relation between A and C (>) not be the same as the designated relation. In
or C and A (<) and so on, but the letters A, these terms, symmetry is a special case of mu-
B, and C had nothing to do with the particular tual entailment.
relations being applied. Such arbitrarily ap- Similarly, transitivity is not applicable to all
plicable relations have been termed "relational kinds of arbitrarily applicable relational re-
frames" (Hayes & Hayes, 1989). Relational sponding (Russell, 1919, 1937). Hayes and
frame theory holds that arbitrarily applicable Hayes (1989) suggested the term combinatorial
relational responding has an extensive history, entailment, defined as follows:
largely in the context of language training, that Crelx and Crel {A relxB and B relyC A
can be brought into the experimental situation .relpC and C relqA}.
by virtue of contextual cues to do so. In this
view, stimulus equivalence as commonly seen That is, given contexts (Crel. and Crely) that
may be the result of an application of a learned specify mutual relations (relx and rely) among
frame of "coordination" (sameness) to the three or more items, relations are entailed (relp
stimuli in arbitrary matching-to-sample pro- and relq) between the stimuli based on the
cedures. combinations of these mutual relations. In these
A wide variety of relational frames are pos- terms, transitivity and the equivalence relation
sible, and the nature of the derived perfor- described by Fields, Verhave, and Fath (1984)
mances they comprise v1aries widely. For ex- are special cases of combinatorial entailment.
ample, the abstract relation of oppositeness has Reflexivity, in Sidman's sense, can always
the property that an opposite of an opposite is involve recognizing stimuli as themselves based
the same, an opposite of an opposite of an upon formal properties of the stimuli involved.
opposite is an opposite, and so on. Whereas Identity matching based on form is, by defi-
stimulus equivalence may be an outcome of a nition, not an arbitrary relation. Relations in
history establishing the frame of sameness or the abstract, however, can either be reflexive
coordination, frames of opposition or of dis- or irreflexive. For example, the arbitrary re-
tinction would give rise to very different kinds lation of sameness is reflexive, but oppositeness
of relational networks. Conditional discrimi- is not (A cannot be the opposite of A). We will
nation training could thus give rise to no de- use the term relational reflexivity/irrefiexivity
rived performances, derived equivalence, or to refer to the relation of a stimulus to itself
other derived relations, depending on the na- in a given relational context. This is viewed
ture of the relations brought to bear on these as a special case of mutual entailment.
discriminations via the contextual cues in- The present study sought to establish three
volved. different types of relational responding (same,
The relational frame concept requires a dis- different, and opposite) and apply them to an
tinct nomenclature, part of which will be re- arbitrary matching-to-sample context. De-
viewed because it is necessary to a description rived performances were then examined to see
of the present experiments and their results. if they could be understood as instances of
Symmetry, although appropriate for equiva- mutual and combinatorial entailment.
lence, is not an appropriate general term for
all arbitrarily applicable relations because
many are not strictly symmetrical (Russell, EXPERIMENT 1
1919, 1937). For example, if A is better than The strategy in this experiment was to pre-
B, then B is worse than A. Hayes (1991; Hayes train arbitrary contextual cues to control same,
& Hayes, 1989) suggested the term mutual opposite, or different responding with non-
entailment, defined as follows: arbitrary stimulus sets. Four subjects were ex-
Crel.{A rel.B I B relyA}. posed to same and opposite pretraining, and
3 were exposed to same and different pretrain-
ARBITRARILY APPLICABLE RELATIONAL RESPONDING 521
0
of C2, B2 and C2 are the same, not opposite. _, _
A
Thus, we wanted to assess whether subjects
trained to pick B2 and C2 (given Al) only in Al ------------------------
the presence of a pretrained cue for an "op- s
posite" relation would now not pick B2 given
C2 in the presence of that cue, but instead ' i
-------C -----------------;----------------------------- C2
would do so in the presence of a pretrained I *,
0
cue for a "same" relation. Similarly, stimuli --------------------------------------- Di D2
related across three stages of the opposite re-
lation are opposites: If Al is the opposite of
B2, Al is the opposite of C2, and C2 is the 0
opposite of D1, then B2 and D1 are opposites. Fig. 1. Basic network of relations trained and tested
Thus, we assessed whether subjects trained to in Experiment 1. Solid arrows indicate trained discrimi-
pick these stimuli in the presence of a pre- nations, and dashed arrows indicate assessment of relations
by probe items. Letters S or 0 indicate relational stimulus
trained cue for an "opposite" relation would SAME or OPPOSITE. Same/different subjects were
now pick B2 given Dl in the presence of that trained in the same fashion but DIFFERENT was used
cue, but not in the presence of pretrained cue in place of OPPOSITE. Additional relations were trained
for a "same" relation. and tested with some subjects (see tables and figures).
I I
[
SAME OPPOSITE (EXP. 1) OPPOSITE (EXP. 2) Al
DIFFERENT
r-i
Bi B2 C1 C2
Dl (EXP. 1) D2 (EXP. 1) xi X2
B3 (EXP.2) C3 (EXP. 2)
wT7
Ni N2 N3
trials with each kind of relational stimulus). reinforced comparison, or the "correct" one in
If any errors were made in relational respond- probe trials, is italicized. For example, the no-
ing to the novel stimuli, responses to this set tation O[Al]B1-B2-B3 indicates that in the
of stimuli were trained to criterion and another presence of the OPPOSITE stimulus, select-
set of novel stimuli was presented for six trials. ing B3 given Al was reinforced or correct. At
If all responses to a novel set of stimuli were times there is no need to describe the unrein-
correct, additional novel sets were presented forced (or "incorrect") comparisons and only
without feedback. The criterion for successful the reinforced or predicted comparison is given.
pretraining was errorless performance on all All subjects received arbitrary matching-to-
trials during the presentation of three consec- sample training: 4 with SAME and OPPO-
utive novel sets of stimuli. If a subject made SITE relational stimuli, 3 with SAME and
any incorrect responses with a set of stimuli, DIFFERENT (using the network of relations
responses to those stimuli were trained with shown in Figure 1 except that some subjects
feedback and an additional set of novel stimuli had DIFFERENT instead of OPPOSITE),
was presented. and 2 control subjects who received no pre-
Pretrainingfor same/different control. Three training but were otherwise treated identically
subjects received same/different pretraining. to the same/opposite subjects. In all training
This was identical to the same/opposite train- blocks each problem was presented for 10 tri-
ing above except that the DIFFERENT re- als, randomly intermixed with the other prob-
lational stimulus was used instead of OP- lems. The size of the training block thus de-
POSITE and only two comparisons were pended on the number of specific problems
presented, one of which was identical to the involved.
sample. There was no particular physical di- The basic training and testing sequences are
mension along which the two comparisons dif- presented diagrammatically in Figure 3. Ex-
fered (e.g., the two comparisons might be a amination of Figure 3 is essential to an un-
rectangle and a circle). In the presence of the derstanding of this complex experiment. The
SAME stimulus, the selection of the compar- basic flow of events was as follows. First, A-B
ison that was identical to the sample was re- and Y-X relations were trained. Y-X trials
inforced. In the presence of the DIFFERENT were included so that the X stimuli would
stimulus, selection of the comparison that was provide a pool of incorrect comparisons that
not identical to the sample was reinforced. Two have a history of reinforced selection and could
comparisons were used with these subjects be- be used in subsequent probes for mutual en-
cause if three comparisons were used, there tailment and combinatorial entailment. Probes
would be two correct answers in the presence then assessed whether the subjects showed mu-
of the DIFFERENT stimulus. This could have tual entailment (e.g., S[B1]A1-X2) and rela-
distracted the subjects from the specific nature tional reflexivity/irreflexivity (e.g., O[Al]Al-
of the relation being trained. Training was N2). Note that a novel stimulus, N2, was used
conducted in blocks of 20 trials- 10 trials with in this test to avoid complicating the relational
the SAME stimulus and 10 with the DIF- network. Then combinatorial entailment (e.g.,
FERENT stimulus. Each block used one set O[Bl]B2-X1) of the trained relations was as-
of two arbitrary stimuli (e.g., a rectangle and sessed. Following each additional training set
a circle), and each stimulus in the set was used (A-C and C-D relations) all of the trained
as the sample an equal number of times. relations were reviewed concurrently with
Arbitrary matching-to-sample training. In the feedback given on each trial, and probes as-
tables and in the text for all experiments, ar- sessed mutual entailment and combinatorial
bitrary matching-to-sample problems and entailment of the trained relations without
probes are described using the same conven- feedback. To advance to the next phase of the
tions. The relational stimulus is given first, study, a subject had to achieve 90% accuracy
using the letters S, 0, and D to represent the for the block of trials and no lower than 80%
SAME, OPPOSITE, and DIFFERENT accuracy on any given problem. Failure to
stimuli. The next letter/number combination achieve criterion resulted in a return to the
in brackets is the sample, and the next set of same training block.
number/letter combinations separated by The number of different types of training
dashes is the set of comparison stimuli. The and testing trials was kept to the minimum
ARBITRARILY APPLICABLE RELATIONAL RESPONDING 525
PRETRAINING Probes
(an example: see text) Mutual entailment
0 S 0 S
Dl D2
Cl C2 C1 C2
(correct) (correct)
Combinatorial entailment
0 S 0 S
EXPERIMENT Dl DI Al Al
Train A-B and Y-X relations Bi B2 B1 B2 Dl D2 Dl D2
0 S 0 S
Al Al Yl Yl Fig. 3. (Continued)
B1 B2 B1 B2 Xl X2 Xl X2
needed to support or disconfirm the presence
Probes of a derived relational network. A network
Relatiional based on a lean set of examples was experi-
Mutual en tailment reflexivity/i rreflexivity mentally advantageous in part because the task
0 S S o was quite complex and we did not wish to
B2 B1 Al Al overload the subjects. The primary concern,
Al X2 Al X2 Al NI Al N2 however, was that derived relations can emerge
in increasingly complex and difficult-to-pre-
Combinatorial entailment dict ways (e.g., through combinations of ex-
clusion and equivalence) in complex relational
0 0 networks of the kind trained in this experi-
Bl B2 ment. The results from uncluttered networks
Xl B2 X2 Bi are thus more open to the detection of sources
of control other than those intended. In the
Train A-C relations training and testing sequence as described, it
(and review previously trained relatiions) may not be immediately obvious why specific
0 S problems were constituted as they were (e.g.,
Al Al why certain comparison stimuli were used),
C1 C2 Cl C2 why only a certain subset of tests was done,
or why tests were conducted in particular se-
Probes quences. The number of considerations in-
Mutual entailment volved were very numerous and thus expla-
0 S nations for every decision are not included in
C2 Cl
this report, but the driving consideration was
the development of a network in which the
Al X2 Al X2 sources of control over responding went be-
yond equivalence and simple forms of exclu-
Combinatorial entailment sion, or other sources such as reinforcement
S 0 0 o density or consistent reinforced pairings be-
C2 C1 C1 B1 tween relational stimuli and specific compar-
B1 B2 C2 X2 BI B2 Cl C2 isons. For example, in Experiment 1 testing
for combinatorial entailment dealing solely
Train C-D relations with same (Bl-Cl) was delayed or avoided
(and review previously trained relattions) entirely to reduce the chance that perfor-
0 S
mances derived through relations of different
C2 C2 or opposite involved simple exclusion via a
Dl D2 Dl D2 previously established equivalence relation.
Testing blocks. Two to four types of probes
Fig. 3. The basic training and testing sequence for were presented in the testing trial blocks, ran-
Experiment 1. Specific sequences varied for some subjects. domly alternating with previously trained
526 DAVID STEELE and STEVEN C. HAYES
S (4 - 100%)
Al
S'>
/r~~~~~~~~e%i
IL-- I A=
E(6-96%)l
-9fnfon
X
4 F-lr,%13Z IL
Ws Al
'
S but not 0 (2 - 100%)
. 0(4- 100%) Cl C2
-C --
0 (5-100%)
C2------
A DI D2
0 (6-96%) A
0 (5- 100%)
0 (6 - 96%)
B1 B2 0 (8 -100%)
0 (10-100%)
,,1
Al ------------------------
S (8- 100%) ,, 'S (8 - 100%) 0 (8- 100%) 't,
S (10- 100%) ,, E (10 - 100%) 0 (10 - 100%) \
c2
4 I
i~e,i
?JI''-Io) OtQ-fO)
f%j.y I-~
0/ c"7 7c:
_7_
Dl D2
0 (9 100%)
-
s (9 100%) -
Fig. 4. Testing performance of Subject 1 (pretrained with SAME and OPPOSITE). Dashed lines indicate probes
that were above 80% correct, where "correct" is defined as selecting the indicated stimulus. Wavy lines indicate probes
were below 80% on at least one of the testing blocks recorded in that section. The letters S and 0 indicate the relational
stimulus presented. Numbers in parentheses indicate the specific testing block (these same numbers are used in Table
1 for cross reference) and the percentage correct. See Table 1 for specific comparison stimuli used and training sequences.
problems, all without feedback. In any given ously trained problems presented without
block, all types of probes were presented an feedback and four types of probes, each type
equal number of times (a minimum of eight), of probe would be presented on nine trials (for
and the total number of probe trials and the a total of 36 probe trials) and each trained
number of trials over previously trained prob- problem would be presented without feedback
lems were equal. These constraints dictated on six trials (for a total of 36 trials). Each
the total number of trials in a given block. For block of trials was planned for the smallest
example, if a block was to contain six previ- number of trials that would meet these criteria.
ARBITRARILY APPLICABLE RELATIONAL RESPONDING 527
Table 1
Percentage of correct responses on training problems and probes for Subject 1 (SAME/OP-
POSITE pretrained).
\4 ----------------
I -a --------------------
0 (8 -95%)
--------
S(B2-_%)
(v ~~~~~~~~~S
(lo 0%) -
S (12 - 100%)
( s (9 - 100%) Al
(
C
I
) S (1 - 1 00%) ,,0tA " (9 - 1 00%) ,
C ) ~~~~
S but not O (7 - 1 00%)
'O- 0 (10-100%) C2
0(13 -0%) 0-(12--100%) A
0 (14 - 0%) / S (1 1 - 100%)
0 (17- 100%)
4A r1-----------____.
----- B2 --
_
.~ ~ ,~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~1
S (16 100%)
- S (16- 100%)
r
Al 'N
> O (15 -100%)
l-
S
,S(15 - 1 00%)
C1------------,-,-
v t(17-100%)
-100- C2 -------i t
Table 2
Percentage of correct responses on training problems and probes for Subject 2 (SAME/OP-
POSITE pretrained).
After C-D training, Subject 1 responded cor- lecting B1 given Al only in the presence of
rectly to D-C trials only 75% of the time (Test SAME, and of selecting B2 given Al only in
Block 7 in Figure 4). An error in the computer the presence of OPPOSITE. Yet when sub-
program allowed progress to the next set of jects were given the choice of B 1 or B2 given
probes, and Subject 1 responded correctly to C2, in the presence of SAME they now chose
D-B probes. After a brief review of C-D train- B2 100% of the time, not Bi.
ing, responses to D-C probes for mutual en- After initial training of the A-B relations,
tailment were 100% accurate. The probes for Subject 2 (see Figure 5 and Table 2) failed to
D-B relations were repeated, and responses show mutual entailment in the presence of the
were again 100% accurate. SAME stimulus. Even after repeated review
Thus, responses to a wide variety of com- of the trained A-B relations, mutual entail-
binations of A, B, and C stimuli were consis- ment given SAME was not demonstrated. At
tent with relational control by SAME and OP- this point the procedure was altered (alteration
POSITE contexts. The relations described in of procedure is indicated in the tables with an
the introduction to Experiment 1 provide an asterisk) to provide additional probes for mu-
example. Subjects had a trained history of se- tual entailment of the SAME relations. Orig-
530 DAVID STEELE and STEVEN C. HAYES
0 (10 -100%)
B1 ---------------------------------
0 (8-100%)
B2 -
Al S (10 - 100%)
', S (9 - 1 00Yo) 0,A> (9 - 1 00%)
, -- S but not 0 (7 - 90%)
1C2-
------ C1 o-O---0 (10-100%)
0 (10 - 100%) AI
Al -----------------------------------------
0 (13 - 88%)
(
S (12 -63%)
) S (14 -100%) , S(13-100%) C
) ' C~~2
Fig. 6. Testing performance of Subject 3 (pretrained with SAME and OPPOSITE). Dashed lines indicate probes
that were above 80% correct, where "correct" is defined as selecting the indicated stimulus. Wavy lines indicate probes
were below 80% on at least one of the testing blocks recorded in that section. The letters S and 0 indicate the relational
stimulus presented. Numbers in parentheses indicate the specific testing block (these same numbers are used in Table
3 for cross reference) and the percentage correct. See Table 3 for specific comparison stimuli used and training sequences.
ARBITRARILY APPLICABLE RELATIONAL RESPONDING 531
Table 3
Percent of correct responses on training problems and probes for Subject 3 (SAME/OPPOSITE
pretrained).
inally this probe presented SIBI ]Al -X2. The The roles of specific arbitrary visual stimuli
additional probes used Xl and B2 as incorrect (Figure 2) were randomly reassigned for Sub-
comparisons. This procedure resulted in 100% ject 4's training and probes (e.g., the stimulus
accurate responses to probes for mutual en- that functioned as Al for earlier subjects might
tailment. Following the training of A-C re- now be the B2 stimulus). This was done to
lations, Subject 2 responded accurately to all make sure that some incidental feature of the
probes for combinatorial entailment, except she stimuli had not produced the pattern of control
did not relate C2 and B2 in the presence of observed with the first 3 subjects. Subject 4
SAME. Review of trained relations did not mastered the trained A-B relations quickly and
alter this performance. Again, the procedure then demonstrated B-A mutual entailment and
was altered to provide for additional unrein- combinatorial entailment (see Figure 7 and
forced probes (see Table 2 and Figure 5 for Table 4). After A-C training, Subject 4 failed
details) for combinatorial entailment of the to pick B2 in probe S[C2]Bl-B2 (Test 5 in
SAME relation. The subject then showed 100% Figure 7). A scheduled session break occurred
accurate responding. At this point Subject 2 at that point, so in the next experimental ses-
chose to withdraw from the experiment. sion previously trained relations were re-
Subject 3 (see Figure 6 and Table 3) made viewed. Subject 4 responded correctly to all
90% or more correct responses on all blocks of probes for mutual entailment and combina-
training and probes until the D stimuli were torial entailment of the A, B, and C stimuli.
added. Subject 3's responses to probes for con- Following C-D training, Subject 4 responded
trol by D-B relations were 63% to 88% ac- incorrectly to probes for D-B relations (Test
curate on the first test (12, Figure 6). After 12 in Figure 7). After probes for the inter-
probes for the intermediate A-D relations, D-B mediate A-D relations, his D-B performances
relations were 100% accurate. rose to 88% correct, and after another test of
Subject 4 (Pretrained Same/Opposite)
B1 o------------------------------ B2 Taieed:
0 (3 -100%)BI B
S (1 - 100%) -- - 0 (1 -100%) j
Al S(5-0%) Al
S(4-100%) -- 4 0 (4-100%)
but not 0 (2 -C%) C2
Cl - ~~~0(5 -92%) C
0
DI D2
I A1 s (10- 83%%)
0 (10-92%)
i 0 (8 -100%) -
S(6- 100%) 0 (6 -100%)
Al 83%) -
S (10
S(9 -100%) ~ '~ 91 00%)
S but ot 0 (7-100%)
------ Cl ----------------- ----- --------------C2--------
0 (10-100%)
0 (10-100%)
0 (12 - 13%)
Bi Bl
~~<80% any~Bes
on B2 ~~~~~0(15 -88%)
0 (17 100%)
100%)
(17
S S(D1 0(11-100%) 0(11-100%) D2
DI 0 110 S0I-10) D
Fig. 7. Testing performance of Subject 4 (pretrained with SAME and OPPOSITE). Dashed lines indicate probes
that were above 80% correct, where "correct" is defined as selecting the indicated stimulus. Wavy lines indicate probes
were below 80% on at least one of the testing blocks recorded in that section. The letters S and 0 indicate the relational
stimulus presented. Numbers in parentheses indicate the specific testing block (these same numbers are used in Table
4 for cross reference) and the percentage correct. See Table 4 for specific comparison stimuli used and training sequences.
ARBITRARILY APPLICABLE RELATIONAL RESPONDING 533
Table 5
Percentage of correct responses on training problems and probes for Subject 5 (no pretraining).
Table 5
(Continued)
Table 6 Table 6
Percentage of correct responses on training problems and (Continued)
probes for Subject 6 (no pretraining).
Trained problem or testing probe Correct
Trained problem or testing probe Correct
13. Probe combinatorial entailment:
Train S[Al]B1-B2 and O[Al]Bl-B2 (20 O[Bl]Xl-B2 100
trials) 95 and O[B2]X2-B1 0
Train S[Yl]X1-X2 and O[Yl]X1-X2 Train A-C relations (20 trials) 95
(20 trials) 95 Review A-B, Y-X, and A-C relations
Train all A-B and Y-X relations (40 tri- (24 trials) 100
als) 100 14. Probe C-A mutual entailment:
1. Probe B-A mutual entailment: O[C2]A1-X2 and S[Cl]A1-X2 100
O[B2]A1-X2 and S[Bl]A1-X2 100 15. Probe combinatorial entailment:
2. Probe reflexivity: S[Al]A 1-NI 0 S[C2]B1-B2 0
and irreflexivity: O[Al ]A1-N2 100 and a S[C1]B1-B2, a S[B1]C1-C2 100
Review A-B and Y-X relations with and a S[B2]Cl-C2 0
feedback (24 trials) 100 Review A-B, Y-X, and A-B relations
3. Probe combinatorial entailment: (24 trials) 100
O[B1]Xl-B2 and O[B2]B1-X2 95 16. Probe combinatorial entailment:
4. Probe B-A mutual entailment: S[C2]Bl-B2 0
O[B2]A1-X2 and S[Bl]A1-X2 100 and O[Cl]C2-X2, O[Cl]Bl-B2 100
5. Probe reflexivity: S[Al]A 1-Nl 0 and O[B1]C1-C2 0
and irreflexivity: O[Al]AI-N2 100 17. Probe combinatorial entailment:
Train A-C relations (20 trials) 100 S[C2]Bl-B2 0
Review A-B, Y-X, and A-C with feed- and O[C 1 ]C2-X2 100
back 100 18. Probe combinatorial entailment:
6. Probe B-A mutual entailment: O[Cl]Bl-B2 and O[B1]C1-C2 100
O[B2]A1-X2 and S[B1]A1-X2 100 a Change of problems normally used in testing.
7. Probe combinatorial entailment:
S[C2]Bl-B2 0
and O[C1]C2-X2 100 but this pattern of responding was observed in
8. Probe combinatorial entailment:
O[C1JB1-B2 and O[Bl]Cl-C2 100 later sessions. Subject 6 responded to these
Review A-B, Y-X, and A-C with feed- probes similarly to the pretrained subjects.
back (24 trials) 100 Combinatorial entailment across stimulus sets.
9. Probe combinatorial entailment: Following A-C training, subjects were given
O[C1]B1-B2 and O[B1]Cl-C2 100 three probes (S[C2]Cl-B2, O[Cl]Bl-B2, and
[Break between sessions] O[B1 ]C 1 -C2) that tested for combinatorial en-
Review A-B and Y-X relations with tailment. The control subjects answered cor-
feedback (24 trials) 100 rectly on the two OPPOSITE problems but
10. Probe B-A mutual entailment:
O[B2]A1-X2 and S[B1]A1-X2 100 even though trained relations were repeatedly
11. Probe reflexivity: S[A1]A1-Nl 0 reviewed and expanded probe sets were given,
and irreflexivity: O[Al]Al-N2 100 they consistently responded to the probe
a Probe reflexivity: S[Al ]A 1-N2 00
and a irreflexivity: O[Al ]Al-N1 S[C2]C1-B2 by selecting the comparison Cl.
Review A-B and Y-X relations with This overall pattern makes sense because each
feedback 100 of these performances in the presence of SAME
12. Probe combinatorial entailment: and OPPOSITE could have been established
O[Bl]Xl-B2 100 in the experimental subjects (with pretraining)
and O[B2]X2-B1 0 via relations in the other context. For example,
Review A-B and Y-X relations with
feedback (24 trials) 100 the sameness between B2 and C2 may have
been established in the experimental subjects
Fig. 8. Testing performance of Subject 5 (no pretraining on relational stimuli). Dashed lines indicate probes that
were above 80% correct, where "correct" is defined as selecting the indicated stimulus. Wavy lines indicate probes
were below 80% on at least one of the testing blocks recorded in that section. The letters S and 0 indicate the relational
stimulus presented. Numbers in parentheses indicate the specific testing block (these same numbers are used in Table
5 for cross reference) and the percentage correct. See Table 5 for specific comparison stimuli used and training sequences.
DAVID STEELE and STEVEN C. HAYES
C2
<80% on any test S (7 - 0%)
S (2 - 0%)
0 (2-100%)
S (5 - 0%) Tralned:
0 (5-100%)
BI B2
0 (8-100%) 0 13-0%)
0 (9- 100%)
--------- B1
0 12-0%)
B2
sy A Al
0 (12 -100%)
0 (I 3 - I100%) - C1 C2
S (10-100%) 0(10-100%)
Al
S (1 4 - 1 00%),,--v* 0w (1 4-1 00%)
i---- C1 C2
0
S(11-0%)
(11-100%)
S (1 1 - 100%) See Table 6
0 (11 0%) See Table 6 i0 (8-100%)
-
(9- 100%)
0
0 (7-100%)
S (1 5 - 0%)
S(15-100%) S (16 - 0%)
S (17 0%)
-
i -----------------
~0 (16 -100%) _
0 (17-100%)
Fig. 9. Testing performance of Subject 6 (no pretraining on relational stimuli). Dashed lines indicate probes that
were above 80% correct, where "correct" is defined as selecting the indicated stimulus. Wavy lines indicate probes
were below 80% on at least one of the testing blocks recorded in that section. The letters S and 0 indicate the relational
stimulus presented. Numbers in parentheses indicate the specific testing block (these same numbers are used in Table
6 for cross reference) and the percentage correct. See Table 6 for specific comparison stimuli used and training sequences.
ARBITRARILY APPLICABLE RELATIONAL RESPONDING 539
D (6 - 8%)
D (7 - 92%)
D (9 - 70%)
D (10 - 90%)
/- /r f-
Bi B2 _--
()
S (8 -83%) S (8 -100%) S (8- 100%) S (8 - 83%)
D (9 - 100%)
ci _ _ D_(10 - 100%l
__ _
-lbo- C2 --------;
'* -------
I-/ \/ J v
D (6 - 8%)
D (7 - 67%)
D (9- 100%)
D (10- 100%)
Fig. 10. Testing performance of Subject 8 (pretrained with SAME and DIFFERENT). Dashed lines indicate
probes that were above 80% correct, where "correct" is defined as selecting the indicated stimulus. Wavy lines indicate
probes were below 80% on at least one of the testing blocks recorded in that section. The letters S and D indicate the
relational stimulus presented. Numbers in parentheses indicate the specific testing block (these same numbers are used
in Table 7 for cross reference) and the percentage correct. See Table 7 for specific comparison stimuli used and training
sequences.
trained relations. It involved finding some de- stimulus, a different detail was used. He ex-
tail of the stimuli that could be related to each plained that the SAME stimulus meant "choose
other. In the presence of one relational stim- the same one," whereas the DIFFERENT
ulus, one detail of the sample and comparison stimulus meant "choose the other one" -re-
stimuli was used; with the other relational ferring to the same or different formal details.
ARBITRARILY APPLICABLE RELATIONAL RESPONDING 541
Table 7 Table 8
Percentage of correct responses on training problems and Percentage of correct responses on training problems and
probes for Subject 8, Experiment 1 (SAME/DIFFER- probes for Subject 9, Experiment 1 (SAME/DIFFER-
ENT pretraining). ENT pretraining).
Trained problem or testing probe Correct Trained problem or testing probe Correct
Train A-B and Y-X relations (40 trials) 88 Train A-B and Y-X relations (40 trials) 100
Train A-B and Y-X relations (40 trials) 98 1. Probe B-A mutual entailment:
1. Probe B-A mutual entailment: D[B2]A1-X2 and S[Bl]A1-X2 98
D[B2]A1-X2 and S[Bl]A1-X2 100 2. Probe reflexivity/irreflexivity 100
2. Probe reflexivity/irreflexivity 95 3. Probe combinatorial entailment:
3. Probe combinatorial entailment: D[Bl]Xl-B2 and D[B2]B1-X2 100
D[Bl]Xl-B2 and D[B2]B1-X2 100 Train A-C relations (20 trials) 100
Train A-C relations (20 trials) 90 Review A-B, Y-X, and A-C relations
Review A-B, Y-X, and A-C relations with feedback 100
with feedback (24 trials) 100 4. Probe C-A mutual entailment:
4. Probe C-A mutual entailment: D[C2]A1-X2 and S[Cl]A1-X2 100
D[C2]A1-X2 and S[Cl]A1-X2 100 5. Probe combinatorial entailment:
5. Probe combinatorial entailment: S[C2]Bl-B2 and D[Cl]C2-X2 100
S[C2]Bl-B2 83 6. Probe combinatorial entailment:
and D[Cl]C2-X2 100 D[CI]Bl-B2, D[Bl]Cl-C2 100
6. Probe combinatorial entailment:
D[C1I]B1-B2, D[Bl]Cl-C2 8
Review A-B, Y-X, and A-C relations
with feedback (24 trials) 100 ence of SAME, showing B-A and C-A mutual
7. Probe combinatorial entailment: entailment and B-C combinatorial entailment.
D[C1]B1-B2 67 Al, B2, and C2 also entered into an equiva-
and D[BI]C1-C2 92 lence class, but in the presence of OPPOSITE.
Review A-B, Y-X, and A-C relations
with feedback (24 trials) 100 For example, given O[C2]B1-B2, these sub-
8. Probe combinatorial entailment: jects chose B2. They had learned to pick both
S[C2]Bl-B2 and a S[Bl]C1-C2 100 B2 and C2 given Al in the presence of OP-
and a S[B2]Cl-C2 and a S[C1]B1-B2 84 POSITE, and thus B2 and C2 were in an
9. Probe combinatorial entailment:
D[Bl]Cl-C2 70 equivalence class given OPPOSITE.
and D[ClIC2-X2 and D[C1]B1-B2 100 All of the pretrained subjects except Subject
10. Probe combinatorial entailment: 7 showed patterns of responding that go be-
D[Bl]Cl-C2 90 yond equivalence or conditional equivalence.
and D[C1]C2-X2 and D[C1]BI-B2 100 Probes for mutual entailment showed bidirec-
a
Change of problems normally used in testing. tional stimulus functions, as with the control
subjects but tests for combinatorial entailment
showed different results. Subject 1, for ex-
The relational stimuli apparently exerted con- ample, selected an opposite of an opposite only
textual control-but over formal selection cri- in the presence of SAME and not OPPOSITE
teria, not over arbitrary matching. (the C2-B2 relation-Test 5 in Figure 4); an
opposite of an opposite of an opposite was
DISCUSSION selected only given OPPOSITE and not SAME
The performance of Subjects 5 and 6 (the (the D1-B2 relation-Tests 8 and 10 in Figure
subjects without pretraining establishing dis- 4). An opposite of an opposite of a same was
tinct relational histories for SAME and OP- selected only given SAME (the Di-Bl rela-
POSITE) showed patterns much like those in tion-Tests 8 and 10 in Figure 4).
the equivalence literature. It is well known Some of the subjects required retraining or
that conditional equivalence classes can emerge special patterns of testing, but the final pat-
that arrange subsets of stimuli into classes given terns were quite similar across subjects. Of
second-order conditional stimuli (e.g., Bush, course, networks such as these give rise to a
Sidman, & de Rose, 1989; Wulfert & Hayes, large number of possible alternative interpre-
1988). For Subjects 5 and 6, Al, BI, and Cl tations of the results-a topic that will be ad-
entered into an equivalence class in the pres- dressed later. Of more immediate interest are
542 DAVID STEELE and STEVEN C. HAYES
100%)
F-------------------------------------------------------
Trmined.-
Bi B2 ---_B2 4--
r--r---- BI -- D (3 - I_IW
A00nO
.
S (1 - 98%) D (1 - 98%)
sl Al C
,
*
Al S (5- 100%)
C1 C2 IS (4 - 1 00%) D (4 - 100%)
S butnotD(2-100%)
D (5 -100%)4 --C2--------
-P----------------D (6_-
OO6OZ)
1 T
D (6- 100%)
D (10 - 100%)
EXPERIMENT 2 Subject 9 (Pretrained
D (11 - 1 00%) Same/Different/Opposite)
V * O 1-100%
yT s O 2-100%)
L-Bl B2 B3
S (6- 100%)
+ s (1
I
- 100%)
S (2 - 100%)
0O
(-100%) ,,-
O(2 -100%) --'
S (10 - 100%)
S (11 - 100%) S (4- 100%) Al S (5- 100%) 0 (4- 100%)
S (8 - 100%) * S (9- 100%) 0 (8- 100%)
Uv , S (3- 100%) 0 (3- 100%)Y a
- XS(S (7- 100%) 0 (7 -100%)> T
---,_ C1 C3
0 (3-100%)
'"' 0 (7-100%) Tralned:
Bi B2 B3
,, 0 (5-100%)
! o~~~~~~~~
(9 - 100%) s (1 2 -1I 00%)
Al
0 (6 - 100%) (via comparison with Bi and B2. See text.)
0 (10 -100%) S##D^O
0 (11 -100%) o 12 - 100%) - Cl C2 C3
Fig. 11. Testing performance of Subject 9 (pretrained with SAME and DIFFERENT) in Experiment 1 (top
half) and her performance in Experiment 2 after being pretrained with SAME, DIFFERENT, and OPPOSITE
(bottom half). Dashed lines indicate probes that were above 80% correct, where "correct" is defined as selecting the
ARBITRARILY APPLICABLE RELATIONAL RESPONDING 543
the similar results between subjects who re- predictable patterns of responding among three
ceived same/different and same/opposite pre- separate relations cannot as readily be ex-
training. On the basis of the subjects' histories plained on the basis of only two principles:
with the relational stimuli, they may simply equivalence and nonequivalence. The use of
have learned that in the context of one rela- these three relations also allowed for the dis-
tional stimulus responding on the basis of tinction between mere difference and oppo-
equivalence was reinforced, whereas in the siteness.
context of the other relational stimulus re-
sponding on the basis of nonequivalence was
reinforced. This would show conditional con- EXPERIMENT 2
trol over equivalence per se (not to be confused
with conditional equivalence classes), but it The intent of Experiment 2 was to bring
does not necessitate an appeal to relational three relations-same, opposite, and differ-
frames. If equivalence is a basic behavioral ent-under stimulus control. Both opposite and
process, only conditional control over that pro- different are irreflexive relations, but have dif-
cess is needed as an explanation. ferent implications for a network of relations.
In Experiment 1, however, only two rela- METHOD
tional stimuli and two comparisons were used.
This causes analytic ambiguities in certain key Subjects, Apparatus, and Stimuli
areas. Consider the probe for combinatorial The subjects for this experiment were Sub-
entailment S[B2]Cl-C2. If B2 is the opposite ject 8 and Subject 9 who had received SAME/
of Al and C2 is the opposite of Al, then B2 DIFFERENT pretraining in Experiment 1.
and C2 are the same. Subjects given same/ The apparatus and stimuli were the same as
opposite pretraining would be expected to se- in Experiment 1 except that provision was now
lect C2, which they did. The subjects who made to present three different relational stim-
received same/different pretraining had been uli and two or three comparisons, depending
trained to select B2 and C2 as being different on the specific problem. Three comparisons
from Al. In the abstract, this leaves the re- were needed in certain types of probes to re-
lationship between B2 and C2 undefined; they duce the applicability of simple forms of ex-
are both different from Al, but they could be clusion. Two comparisons were used in some
either the same as or different from one an- training and probe items to distinguish the
other. The fact that there were only two com- different and opposite relations.
parisons made available another source of con-
trol, however, in the actual matching-to-sample Procedure
task. Cl, Al, and Bl had entered into an The general network of trained and tested
equivalence class in the presence of SAME for relations is shown in Figure 13. The plan for
the SAME/DIFFERENT subjects (see Fig- the training and probes is given in Figure 14.
ures 10 and 1 1). B2 was not in that class, and Subject 8 and Subject 9 were first given pre-
thus given B2 as a sample in the presence of training with SAME and OPPOSITE iden-
SAME, subjects could merely exclude Cl and tical to that used with other subjects in Ex-
select C2. These kinds of problems with two- periment l. Subjects had already learned
choice procedures in equivalence research have S[Al]B 1, D[Al]B2, S[Al]C1, and D[Al]C2 in
been previously noted (Sidman, 1987). Experiment l. They were now given the same
To address this analytic ambiguity, a second training as the same/opposite subjects in Ex-
experiment was conducted in which three re- periment I except that the B2 and C2 stimuli
lational stimuli were trained: SAME, OP- in Experiment I (what will be called B3 and
POSITE, and DIFFERENT. Distinct and C3, respectively, in this experiment) were re-
indicated stimulus. Wavy lines indicate probes were below 80% on at least one of the testing blocks recorded in that
section. The letters S, D, and 0 indicate the particular relational stimulus presented. Numbers in parentheses indicate
the specific testing block (these same numbers are used in Tables 8 and 11 for cross reference) and the percentage
correct. See Tables 8 and 11 (for Experiments 1 and 2, respectively) for specific comparison stimuli used and training
sequences.
544 DAVID STEELE and STEVEN C. HAYES
Table 9
Percentage of correct responses on training problems and probes for Subject 7 (SAME/DIF-
FERENT pretraining).
placed with new figures because the old ones also given a review of the same/different train-
had already been related as different from Al ing that they had received in Experiment 1.
for these subjects (in the problems D[Al ]B2 As in Experiment 1, probe blocks consisted
and D[Al ]C2). This essentially added the fol- of equal numbers of probe items and previ-
lowing relations to those trained in Experi- ously trained problems presented without re-
ment 1: O[A1 ]B3 and O[A1 ]C3. Subjects were inforcement. Order of the probes was random-
ARBITRARILY APPLICABLE RELATIONAL RESPONDING 545
D (21 - 0%)
B1 B2
S (22 - 0%)A{ S (22 - 50%)
S (23 - 0%)T%:, S (23 - 0%) S (20 - 0%)
S (21 - 0%) )
t) ; < 'Al S (22 - 17%)
S (23 - 0%)
Cl - 1
C1(1 ,%1# ^n
Boo- C2
u (ZU- w*)%
o/ I
S (22 - 0%)
S (23 - 0%)
Fig. 12. Testing performance of Subject 7 (pretrained with SAME and DIFFERENT). Dashed lines indicate
probes that were above 80% correct, where "correct" is defined as selecting the indicated stimulus. Wavy lines indicate
probes were below 80% on at least one of the testing blocks recorded in that section. The letters S and D indicate the
relational stimulus presented. Numbers in parentheses indicate the specific testing block (these same numbers are used
in Table 9 for cross reference) and the percentage correct. See Table 9 for specific comparison stimuli used and training
sequences.
ized with one exception noted below. On each 27 trials in which the trained A-B problems
trial the placement of comparisons (left, center, were presented in extinction. A-C training was
or right) was randomly determined. A-B train- conducted in blocks of 27 trials, and then A-B
ing was given in blocks of 27 trials (nine for and A-C training was reviewed with each
each problem). The probes for mutual entail- problem presented three times. Probes N, 0,
ment and the probe for combinatorial entail- and P, Probes Q and R, and Probes S and T
ment (Probes K, L, and M in Figure 14) were (see Figure 14) were each presented sequen-
presented in blocks of 27 probes combined with tially as blocks.
546 DAVID STEELE and STEVEN C. HAYES
Probe for
combinatorial
C1 C2 C3 Probe for mutual entailment entailment
(Also Trained in Experiment 1)
(K) (L) (M)
S 0 0
B1 B3 B3
A 1 B2 B3 Al B2 Bi B2
Basic Set of Tested Derived Relatlons
Train: (H) (I) U)
S 0 D
Al Al Al
Cl C2 C3 Cl C2 C3 Cl C2
---51 B2 B3
Probe for
combinatorial
Probe for mutual entailment entailment
D S Al sO0:: .0
S --- 0" (N) (0) (P)
S 0 0
C1 C3 C3
j_Cl C2 C3 -
A 1 C2 C3 Al C2 Cl C2
--
0
,_----_____--_____-----_ -----_--------_-----________________--____----------_-_-_-___-___ _ -____ Probe for combinatorial entailment
Fig. 13. Basic network of relations trained and tested (Q) (R) (S) (T)
in Experiment 2. Solid lines indicate trained discrimina- S 0 S 0
tions, and dashed lines indicate assessment of relations by
probe items. Letters S, D, and 0 indicate relational stimuli Bl B1 B3 B3
SAME, DIFFERENT, and OPPOSITE. Training and Cl C2 C3 Cl C2C3 ClC2 C3 Cl C2 C3
testing differed for some subjects.
Probe for combinatorial entailment
(U) (V) (W)
Some of the first 10 types of probes (Probes S D 0
sented with two familiar comparisons (BI and Probe R (O[B 1 ]C 1 -C2-C3). Review of trained
B2) and a novel stimulus (N3) as a compar- relations and further probe trials failed to alter
ison. Selecting Bi in S[Cl]B1-B2-N3 (Probe the pattern. On Probes U, V, and W, Subject
U) can be derived from combinatorial entail- 8 responded at criterion levels on the second
ment of the same relation. In the probe, presentation (see Figure 15b). On the fourth
D[C1 ]B1 -B2-N3 (Probe V), picking B2 comes presentation of these probes, he sometimes se-
from combinatorial entailment of the relations lected comparison N3 when given Probe V
same and different. Subjects were trained that (D[C1]B1-B2-N3). It should be noted that in
B2 is different from Al, and Cl is the same terms of control by arbitrary relations, this is
as Al. Thus B2 is different from Cl. Re- not an incorrect response. If the previous ex-
sponding in the last probe, O[Cl]Bl-B2-N3 posure to N3 established it as the opposite of
(Probe W), assessed whether OPPOSITE Cl, then N3 is also different from Cl (this is
controls the same kind of responding as DIF- one reason that exposure to Probe W was re-
FERENT. If not, and if responding is based served until after Probes U and V). On three
on combinatorial entailment of the relations further exposures to this set of probes, Subject
same and opposite, then neither Bi nor B2 8's responses followed the predicted pattern on
can be a correct choice. B 1 and C1 are the 100% of the trials. Finally, the experimenter
same as Al and B2 is different from Al, but noticed a clue to Subject 8's failure to respond
not opposite of Al, and thus not of Cl. The correctly on Probe R. When the previously
only choice left is the novel stimulus. If OP- trained relations were presented in extinction,
POSITE merely controls nonequivalence, then Subject 8 sometimes made erroneous responses
either B2 or N3 is a possibility. The predicted to the previously trained problem O[AI]C1-
pattern of responding to Probe W (O[Cl]Bl- C2-C3. A set of probes that included O[A1 ]C 1-
B2-N3) depends on control by the extended C2-C3, S[Bl ]A 1-B2-B3, and O[Bl ]C1-C2-C3
network of relations. The subject can select the were added. Immediate increase in correct re-
novel stimulus (N3) by eliminating the other sponding to Probe R was observed, to 100%
comparisons as incorrect. Thus, for Probes U, after two blocks of these probes.
V, and W, the order of presentation was not
randomized; subjects responded to Probes U Subject 9
and V at least three times each before being Subject 9's performance was characterized
exposed to Probe W. by extremely accurate responding (see Table
Additional probes for Subject 9. If subjects 11). Her data are displayed side by side with
did pick N3 in Probe W, then N3 might enter Experiment 1 data (see Figure 11) to enable
into the network of relations as the opposite a fuller grasp of her total performance. After
of Cl and therefore the same as C3. Subject same/opposite pretraining and a review of the
9 was given an additional probe for mutual same/different pretraining, A-B training was
entailment (O[N3]C1-C2-C3) and a probe for accomplished with only one wrong response.
combinatorial entailment (S[N3]C1-C2-C3) A-C training was also accomplished with only
to see if N3 had entered into the network of one wrong response. Responses to all probes
relations via testing alone. were 100% correct. Probes U, V, and W were
given Subject 9 three times even though she
RESULTS AND DISCUSSION was 100% accurate on the first presentation in
Subject 8 order to assess her consistency of responding.
After same/opposite pretraining and a re- Subject 9 was given the two additional probes
view of same/different pretraining, Subject 8 to see if N3 entered into the network of re-
(see Table 10) mastered the A-B relations in lations, and again all responses were consistent
one block of 27 trials. There are some initial with the predicted pattern.
problems with probes for mutual entailment
and combinatorial entailment, but after four
blocks of probes responding was at criterion GENERAL DISCUSSION
levels (see Figure 15a). Previous work has shown that nonarbitrary
Training A-C relations required only one stimulus relations can be brought under con-
block of 27 trials, but Subject 8 failed to show textual control and applied to novel sets of
combinatorial entailment of responding with formally related stimuli (Lowenkron, 1989).
548 DA VID STEELE and STEVEN C. HAYES
Table 10
Percentage of correct responses to training and probe trials for Subject 8, Experiment 2 (SAME/
DIFFERENT/OPPOSITE pretraining).
Table 10
(Continued)
0 (9 - 100%)
LEJ1 B3 J (
">s S (9 -100%) 0 (9-89%),, ' )
1,I
-
q
0 (13-44%) S (11 - 89%) S 13 - 89%)
0 (14-89%) S (12 - 89%) Al N>p~
S 14- 89%
S 15- 89%) 0 (11 - 0%)
0 (15 - 100%)
I
IIVK ,
0 (12 - 0%)
0 (10 -100%)
Fig. 15a. Testing performance of Subject 8 (pretrained with SAME, DIFFERENT, and OPPOSITE) in Ex-
periment 2. Dashed lines indicate probes that were above 80% correct, where "correct" is defined as selecting the
indicated stimulus. Wavy lines indicate probes were below 80% on at least one of the testing blocks recorded in that
section. The letters S, 0, and D indicate the particular relational stimulus presented. Numbers in parentheses indicate
the specific testing block (these same numbers are used in Table 10 for cross reference) and the percentage correct.
See Table 10 for specific comparison stimuli used and training sequences.
Fig. 15b. Continuation of testing performance of Subject 8 (pretrained with SAME, DIFFERENT, and OP-
POSITE) in Experiment 2. Dashed lines indicate probes that were above 80% correct, where "correct" is defined as
selecting the indicated stimulus. Wavy lines indicate probes were below 80% on at least one of the testing blocks
recorded in that section. The letters S, 0, and D indicate the particular relational stimulus presented. Numbers in
ARBITRARILY APPLICABLE RELATIONAL RESPONDING 551
S (19-100%) 0(19-100%) ,,
C
(
A
S (16 88%):
" S (25- 100%) 0 (25- 100%),,,-'
A1
C
)
C
-
IS - 1 00%j)"
S (17-100%): s;(18
( (21 - 100%) S (22- 1040%):
S (23-100%)I is (27 - 100%) 1U I 0 (18-11%)
S (24-100%) :s (28 - 100%) 'r 0 (21 -0%)
0 (27 - 0%)
S (20-100%) 0 (20-1 00%)j s 0 (28 - 0%)
-
I,
S (26 - 88%) 0 (26-100%)
:1 C3--
A 0 (20-100%) 0 (21-100%) 0 (22-100%)
* O(26- 100%) O (27- 100%) O (28- 100%)!
Tralned:
0 (16 - 88%) via comparison with Bi and B2. See text. BI B2 B3
0 (17-100%)
0 (23- 100%) O 024 - 100%)
stS Al
---------------------------------------------------________
D (30-100%)- ----------
B2 B3
S (31 - 100%) 0 (31 -100%), -
A >^S (32- 100%)
S (30- 100%) "IA
S (34- 100%) S (33 - 100%) Al S (29 -1 00 1) O (32 - 89%)
S (34- 10093 C6) 0 (33 - 100%)
I---v--C C1 C3_I
Lo. 0 (33-100%) 0 (29 - 89%)
0(34-89%)
0 (30 - 100%) (via comparison with Bi and B2. See text.)
0 (35-89%)
l
---------------------------------------------------------------- N3
parentheses indicate the specific testing block (these same numbers are used in Table 10 for cross reference) and the
percentage correct. See Table 10 for specific comparison stimuli used and training sequences. See also Figure 15a.
552 DAVID STEELE and STEVEN C. HAYES
Table 1 1
Percentage of correct responses on training and probes for Subject 9, Experiment 2 (SAME/
DIFFERENT/OPPOSITE pretraining).
N3 could thus be explained on that basis. But Consider the wide variety of qualitative re-
if such an explanation is adopted, performance lations that can be modeled with digital com-
on Probe M (Figure 14) would have to be puters. At the level of circuitry, all can be
explained, because subjects selected B1 given reduced to combinations of "on" and "off."
O[B3]Bl-B2. Many of the other probes would Equivalence and exclusion have this same on/
also predict different performances than those off quality and presumably could be used by
actually seen if mere S+ and S- control were behavioral theorists to model a wide variety of
the issue. An account in terms of S+ and S- complex cognitive relations, including opposite
control does not fit with all of the testing results and different. In computer modeling of com-
and is inadequate on these grounds. plex relations, however, very many combina-
Equivalence and exclusion. Many of the probe tions of on and off can be necessary, and the
performances can be explained on the basis of same may be true for models of complex re-
equivalence and exclusion. Explaining Ex- lations based on equivalence and exclusion
periment 2 strictly in these terms is difficult alone.
because three distinct patterns of performance Conversely, arbitrarily applicable relational
were shown. This seems at least to require an responding may itself be taken to be the basic
appeal to higher order forms of exclusion, in unit. In that case, both exclusion and equiv-
which stimuli selected by virtue of exclusion alence would be viewed as examples of a
in the presence of DIFFERENT were them- broader behavioral process. The results of the
selves excluded in the presence of OPPOSITE. present experiments fit with this idea, but they
Such an analysis would be complicated but is do not eliminate the alternatives. Selecting
surely not impossible. among these and other alternatives will re-
ARBITRARILY APPLICABLE RELATIONAL RESPONDING 553
quire better behavior-analytic methods, and This finding is distinct from demonstrations of
the present study may be of some use in this conditional equivalence classes (e.g., Wulfert
area. & Hayes, 1988) that involve the composition
of equivalence classes, not the presence or ab-
Implications for Equivalence sence of equivalence.
The present studies have several implica- To the extent that these data strengthen the
tions for the study of equivalence. plausibility of the relational frame account,
Compound stimuli. In their study of condi- there are other implications for issues of con-
tional equivalence classes, Bush et al. (1989) textual control of equivalence. Arbitrarily ap-
pointed out that apparent second-order con- plicable relational responding must be able to
ditional stimuli may have entered into a com- be brought to bear by the context, not solely
pound with the sample and thereby exerted by formal properties of the items being related.
control over conditional discriminations. The What contextual factors might be involved in
procedure in the present study rules out control equivalence research?
of responding by a compound stimulus and Probably the most fundamental language
gives unequivocal evidence for second-order process is that of naming. Of importance to
conditional control. Consider the probe the present argument is the development of
S[C2]B1-B2. Pretrained subjects in Experi- coordination between the productive and re-
ment 1 reliably selected the comparison B2, as ceptive aspects of naming. For example, chil-
predicted by a relational response account. dren are taught to name an object and also to
There were no training items that used C2 as orient toward a named object. Each discrim-
a sample, and all previous probes with C2 as ination may be trained unidirectionally at first,
a sample had been presented with the OP- but the overall performance occurs in consis-
POSITE stimulus as the relational stimulus. tent contexts in which the bidirectional rela-
Further, there were no training items in which tion is applicable. (Parenthetically, coordi-
B2 was the reinforced comparison selection nated name-object and object-name relations
when the SAME stimulus was presented as are not strictly symmetrical because the re-
the relational stimulus. The relational stim- sponses involved differ. This may be resolved
ulus and the sample must have functioned in- if the child has a generalized imitative rep-
dependently to produce the pattern of respond- ertoire that enables the repetition of sounds
ing observed in the present study. This provides that are heard. Thus, the symmetrical version
support for Sidman's development of the four- of productive naming is: hear name-orient
and five-term contingency nomenclature (Sid- toward object [given A then B]; when oriented
man, 1986), but only if these terms are avail- toward object-hear name [then say the name
able to control distinct responses. For example, heard] [given B then A produce A].)
although the SAME, OPPOSITE, and DIF- If a child with an extensive naming history
FERENT stimuli can be thought of as fifth is taught "This is your boat," contextual cues
terms in contingencies, their effects were dis- (such as the word "is," or the naming context
tinct. more generally) reliably predict that if this is
Contextual control over equivalence and non- a boat, a boat is this. Thus, the child may now
equivalence. For pretrained subjects, one re- orient toward the boat when asked "Where is
lational stimulus reliably resulted in the choice your boat?" without direct training to do so
of reflexive or equivalent sample-comparison because contextual cues brought a frame of
selections. The other relational stimuli re- coordination to bear on the trained relation.
sulted in irreflexive or nonequivalent choices. The naming situation is similar to the
At the least, this shows that forming equiva- matching-to-sample preparation usually used
lence itself can be brought under contextual in experimental studies of equivalence. In nat-
control. In the presence of one relational stim- ural language circumstances children are often
ulus, a given comparison would enter into an asked, for example, which item of several "is
equivalence relationship with a sample. In the called" a sample name-essentially a match-
presence of the other relational stimulus, the ing-to-sample situation. Thus, the matching-
same comparison would be excluded from the to-sample procedure itself may serve as a con-
class of stimuli equivalent to that same sample. textual cue for responding in terms of sameness
554 DAVID STEELE and STEVEN C. HAYES
because of the formal properties of the training the words "same," "different," and "opposite"
and testing situation in combination with the were used as relational stimuli, the results could
child's history with a variety of relational tasks well have been similar. The study of contex-
such as naming. Note that we are not ap- tual control (including verbal control) over re-
pealing to naming as a mediational process. lational responding may provide a fruitful av-
From the point of view of relational frame enue of research for the study of equivalence,
theory, a history of relational responding would exclusion, and other types of relations.
explain the derived relations seen both in nam- Increasingly, behavior analysts are viewing
ing and in equivalence. Such a history in one stimulus equivalence and similar phenomena
area may affect responding in another, how- as important preparations for the investigation
ever, through the development and instantia- of human language and cognition. The present
tion of a common behavioral process. study provides a method for the study of a
To assess the kinds of contexts in which much wider range of stimulus relations that
equivalence will emerge, more research should can be brought to bear on arbitrary stimuli.
be done on the ease with which equivalence is As such, it may be useful for behavior-analytic
shown in a variety of tasks and experimental investigations of complex cognitive and verbal
preparations other than matching to sample, phenomena.
especially with very young children. Work
needs to be done on ways of breaking up and
preventing the formation of equivalence rela- REFERENCES
tions because this may assess whether there Bush, K. M., Sidman, M., & de Rose, T. (1989). Con-
are contexts in which equivalence is unlikely. textual control of emergent equivalence relations. Jour-
To assess the role of history in equivalence, nal of the Experimental Analysis of Behavior, 51, 29-45.
longitudinal studies need to be conducted on Devany, J. M., Hayes, S. C., & Nelson, R. 0. (1986).
the development of equivalence in infants Equivalence class formation in language-able and lan-
guage-disabled children. Journal of the Experimental
younger than 2 years (because children already Analysis of Behavior, 46, 243-257.
show equivalence by then; Devany, Hayes, & Fields, L., Verhave, T., & Fath, S. (1984). Stimulus
Nelson, 1986). equivalence and transitive associations: A methodolog-
The experimental procedure used in many ical analysis. Journal of the Experimental Analysis of
Behavior, 42, 143-157.
studies of equivalence could also have encour- Gatch, M. B., & Osborne, J. G. (1989). Transfer of
aged its formation. Some studies (e.g., Lazar, contextual stimulus function via equivalence class de-
Davis-Lang, & Sanchez, 1984; R. Saunders, velopment. Journal of the Experimental Analysis of Be-
Wachter, & Spradlin, 1988; Sidman et al., havior, 51, 369-378.
1985; Sidman & Tailby, 1982) exposed sub- Harrison, R. J., & Green, G. (1990). Development of
conditional and equivalence relations without differ-
jects to identity matching (though often in un- ential consequences. Journal of the Experimental Anal-
reinforced trials) before beginning the training ysis of Behavior, 54, 225-237.
with the arbitrary stimuli. This may have cued Hayes, S. C. (1989). Nonhumans have not yet shown
the application of the relation of sameness in stimulus equivalence. Journal of the Experimental Anal-
ysis of Behavior, 51, 385-392.
the arbitrary matching-to-sample task that fol- Hayes, S. C. (1991). A relational control theory of stim-
lowed, much as SAME did in the present stud- ulus equivalence. In L. J. Hayes & P. N. Chase (Eds.),
ies. In other studies (Devany et al., 1986; Gatch Dialogues on verbal behavior (pp. 17-40). Reno, NV:
& Osborne, 1989; Spradlin et al., 1973; Weth- Context Press.
Hayes, S. C., & Hayes, L. J. (1989). The verbal action
erby, Karlan, & Spradlin, 1983), subjects were of the listener as a basis for rule governance. In S. C.
instructed to choose the comparison that "went Hayes (Ed.), Rule-governed behavior: Cognition, contin-
with" the sample, perhaps cuing a frame of gencies, and instructional control (pp. 153-190). New
coordination. York: Plenum Press.
Rather than view such factors as problems Kennedy, C. H., & Laitinen, R. (1988). Second-order
conditional control of symmetric and transitive rela-
to be avoided, the present concept suggests that tions: The influence of order effects. Psychological Rec-
they are important variables to be studied. ord, 38, 437-446.
Verbal humans appear to have a wide variety Lazar, R. M., Davis-Lang, D., & Sanchez, L. (1984).
of relational responses under contextual con- The formation of visual stimulus equivalences in chil-
dren. Journal of the Experimental Analysis of Behavior,
trol, and, in the present study, pretraining 41, 251-266.
probably only actualized already learned be- Lowenkron, B. (1989). Instructional control of gener-
havior. If pretraining had been omitted and alized relational matching to sample in children. Jour-
ARBITRARILY APPLICABLE RELATIONAL RESPONDING 555
nal of the Experimental Analysis of Behavior, 52, 293- crossmodal transfer of stimulus equivalences in severe
309. retardation. American Journal of Mental Deficiency, 77,
McIntire, K. D., Cleary, J., & Thompson, T. (1987). 515-523.
Conditional relations by monkeys: Reflexivity, sym- Sidman, M., Cresson, O., Jr., & Willson-Morris, M.
metry, and transitivity. Journal of the Experimental (1974). Acquisition of matching to sample via medi-
Analysis of Behavior, 47, 279-285. ated transfer. Journal of the Experimental Analysis of
Reese, H. W. (1968). The perception of stimulus relations: Behavior, 22, 261-273.
Discrimination learning and transposition. New York: Sidman, M., Kirk, B., & Willson-Morris, M. (1985).
Academic Press. Six-member stimulus classes generated by conditional-
Russell, B. (1919). Introduction to mathematical philos- discrimination procedures. Journal of the Experimental
ophy. London: Allen & Unwin. Analysis of Behavior, 43, 21-42.
Russell, B. (1937). The principles of mathematics (2nd Sidman, M., & Tailby, W. (1982). Conditional discrim-
ed.). London: Allen & Unwin. ination versus matching to sample: An expansion of
Saunders, K. J. (1989). Naming in conditional discrim- the testing paradigm. Journal of the Experimental Anal-
ination and stimulus equivalence. Journal of the Ex- ysis of Behavior, 37, 5-22.
perimental Analysis of Behavior, 51, 379-384. Spradlin, J. E., Cotter, V. W., & Baxley, N. (1973).
Saunders, R. R., Wachter, J., & Spradlin, J. E. (1988). Establishing a conditional discrimination without di-
Establishing auditory control over an eight-member rect training: A study of transfer with retarded ado-
equivalence class via conditional discrimination pro- lescents. American Journal of Mental Deficiency, 77, 556-
cedures. Journal of the Experimental Analysis of Behavior, 566.
49, 95-115. Wetherby, B., Karlan, G. R., & Spradlin, J. E. (1983).
Sidman, M. (1971). Reading and auditory-visual equiv- The development of derived stimulus relations through
alences. Journal of Speech and Hearing Research, 14, 5- training in arbitrary-matching sequences. Journal of the
13. Experimental Analysis of Behavior, 40, 69-78.
Sidman, M. (1986). Functional analysis of emergent Wulfert, E., & Hayes, S. C. (1988). Transfer of a con-
verbal classes. In T. Thompson & M. D. Zeiler (Eds.), ditional ordering response through conditional equiv-
Analysis and integration of behavioral units (pp. 213- alence classes. Journal of the Experimental Analysis of
245). Hillsdale, NJ: Erlbaum. Behavior, 50, 125-144.
Sidman, M. (1987). Two choices are not enough. Be-
havior Analysis, 22, 11-18. Received May 15, 1990
Sidman, M., & Cresson, O., Jr. (1973). Reading and Final acceptance April 30, 1991