JOURNAL OF APPLIED BEHAVIOR ANALYSIS 1997, 30, 69–91 NUMBER 1 (SPRING 1997)

TEACHING CHILDREN TO SPELL:

DECREASING CONSONANT CLUSTER ERRORS BY
ELIMINATING SELECTIVE STIMULUS CONTROL
BEVERLY BIRNIE-SELWYN AND BERNARD GUERIN
UNIVERSITY OF WAIKATO

Six elementary-aged children were taught to spell words containing initial consonant
clusters (CCs). They were trained to select printed words in response to the corresponding
spoken words using computerized matching-to-sample procedures. After each training
session, they were tested for spelling with a constructed-response transfer test. Based on
previous selective stimulus control research, we hypothesized that only the first letter of
an initial CC might control spelling when CC spelling errors are made. Thus, a critical-
difference matching-to-sample training condition that required the children to respond
to both letters of the CC to be correct was compared to a multiple-difference training
condition that required the children to respond to only one letter of the pair. Results
showed that children made fewer errors during the multiple-difference training condition
than during the critical-difference training condition. On the constructed-response trans-
fer tests, however, more overall errors and CC errors were made in the multiple-difference
condition than in the critical-difference condition, and the words trained in the multiple-
difference condition required more training sessions to reach criterion. All children im-
proved their spelling of novel CC words by the completion of training. If normal class-
room or home reading was to be supplemented by computer tasks of the kind used here,
some spelling problems could be circumvented without costly intervention by a teacher
or a special trainer.
DESCRIPTORS: spelling, children, computers, education, stimulus overselectivity

Most research on spelling, and how to (1990–1991) have developed useful com-
teach spelling, has been done by educational puter technologies for teaching spelling,
researchers based in cognitive psychology whereas other applied behavior analysts have
(Foorman, Francis, Novy, & Liberman, developed interventions that reinforce cor-
1991; Griffith, 1991; Lewkowicz, 1980; rect spelling (Gettinger, 1985; Greenwood et
Treiman & Zukowski, 1988, 1990). Within al., 1987; Neef, Iwata, & Page, 1980; Ollen-
behavior analysis, Lee and colleagues (Lee & dick, Matson, Esveldt-Dawson, & Shapiro,
Pegler, 1982; Lee & Sanderson, 1987) have 1980) and have looked at variables that con-
developed a conceptual basis for spelling that trol practice effects (Cuvo, Ashley, Marso,
emerges from reading and suggested that re- Zhang, & Fry, 1995; Gettinger, 1993).
searchers look more closely at the stimulus Dube et al. (1991) taught 2 young men
control involved in specific types of spelling with intellectual disabilities to spell using a
errors. Dube, McDonald, McIlvane, and computer. Previously trained, arbitrarily
Mackay (1991), Stevens, Blackhurst, and matched words and pictures were used as vi-
Slaton (1991), and Moxley and Warash sual stimuli. Participants were shown the
stimuli and were required to select letters
We thank Vicki Lee and Mary Foster for helpful
comments about this manuscript, and Bill Dube, Rob- presented at the bottom of the screen with
ert Stromer, and Harry Mackay for useful discussions a touch screen to construct the matching
during a visit to the Shriver Center. word. The selection pool contained 10 let-
Correspondence should be addressed to Bernard ters, some of which, if selected in the right
Guerin, Department of Psychology, University of Wai-
kato, Private Bag, Hamilton, New Zealand (E-mail: order, would spell the word corresponding
bguerin@waikato.ac.nz). to the picture. Dube et al. found that the

69
70 BEVERLY BIRNIE-SELWYN and BERNARD GUERIN

use of this method decreased the number of (1991) raw data revealed some CC errors,
spelling errors made by the 2 participants. although they were not all simple CC errors.
Stromer and colleagues (Stromer & Mackay, For example, 1 participant constructed HN
1992a, 1992b; Stromer, Mackay, & Stod- for HAND, FA for FLAG, and WH for
dard, 1992), using the same computer tech- WATCH, and a second participant con-
nology with disabled and nondisabled par- structed H for SHOE and FA for FLAG.
ticipants, found that novel constructed-re- The present analysis follows up the sug-
sponse spellings emerged after matching-to- gestion of Dube et al. (1991) that proce-
sample techniques had established equivalence dures that are designed to overcome stimulus
classes among pictures, printed words, and selectivity could be used to decrease CC er-
spoken words. This occurred when the stu- rors (Allen & Fuqua, 1985; Reynolds, New-
dent was able to perform a delayed match- som, & Lovaas, 1974; Schover & Newsom,
ing-to-sample task in which the printed 1976; Schreibman, Koegel, & Craig, 1977).
words were removed and the student con- In examining selective stimulus control, par-
structed the word in the absence of the stim- ticipants are taught to respond to multicom-
uli. Like Lee and Sanderson (1987), Stromer ponent stimuli and then are tested for re-
and Mackay (1992b) also noted in their dis- sponding to the components separately. Se-
cussion that more attention should be given lective stimulus control occurs when partic-
to training discrimination of the individual ipants respond to only some components of
letters in a word during reading and that the multicomponent stimuli. For CC errors,
such training might improve later spelling. this means that the sound of the word
One of the more common types of spell- SHOW controls only the printed letters S,
ing errors made by children is the misspell- O, and W. In particular, the SH sound con-
ing of consonant clusters (CCs) (Bruck & trols only the printed letter S.
Treiman, 1990; Treiman, 1993). Words with Allen and Fuqua (1985) used three dif-
CCs have two or more consonants in se- ferent conditions in a discrimination train-
quence, such as suCH or THat. Read (1975) ing task to decrease selective stimulus con-
found that children often misspelled conso- trol over a compound stimulus of geometric
nant clusters that occurred at the end of shapes. In the multiple-difference condition,
words, with the most common mistake be- the negative discriminative stimulus (S2)
ing the omission of the first consonant in differed from the original positive discrimi-
the cluster (e.g., HAD instead of HAND). native stimulus (S1 ) in multiple ways
CC errors have also been found to occur at (shape, orientation, number of elements,
the beginning of words, with the second etc.). In the minimal-difference condition,
consonant in such clusters being omitted the S2 differed minimally from the S1 in
(e.g., SAY instead of STAY) (Bruck & Trei- no more than two ways. In the critical-dif-
man, 1990; Miller & Limber, 1985; Trei- ference condition, the S2 differed from the
man, 1991). A review of the spelling skills S1 in only one way. Results from this study
of 67 first-grade children by Treiman (1991) showed that more errors were made by the
found that 42% of initial CC errors were children in the multiple-difference condition
accounted for by the omission of the second than in both the minimal- or critical-differ-
consonant in the cluster. ence conditions. Fewer trials were needed in
In the study by Dube et al. (1991), the the critical-difference conditions than in the
emphasis was on correct spelling using a other conditions to meet their 90% correct
constructed matching-to-sample task rather criterion. The results suggested that critical-
than on CC errors. However, Dube et al.’s difference training was an effective proce-
 STIMULUS CONTROL OF SPELLING ERRORS 71

dure for decreasing selective stimulus con- procedures would produce differences in
trol. Applying this procedure to spelling the performance on a constructed-response
compound stimulus SH in the word SHOW, transfer test.
the use of PLAY as a comparison in match-
ing-to-sample would be part of a multiple-
METHOD
difference condition, the use of SLOP would
be part of a minimal-difference condition, PARTICIPANTS
whereas the use of SLOW would be part of Six parents volunteered their children as
a critical-difference condition. The present participants in this study after volunteers
study applied only multiple- and critical-dif- were recruited for a computer experiment.
ference methods to CCs. The children were of normal intelligence
Interestingly, selective stimulus control and were not selected because of special
has been found to be widespread in children spelling problems. They were between the
with autism and intellectual handicaps but ages of 4 years 5 months (Fiona) and 7 years
is not usually found in nondisabled children (Colin).
(Allen & Fuqua, 1985; Lovaas, Koegel, &
Schreibman, 1979; Lovaas, Schreibman, MATERIALS AND SETTING
Koegel, & Rehm, 1971). This might be a The experiment was carried out in a stan-
function of the types of stimuli examined, dard office, and the experimenter was pres-
however. Nondisabled children might not ent throughout each training session. Toys
have selectivity problems with the shapes were available during breaks between ses-
and patterns commonly used to research se- sions. The children could ask for a break
lective stimulus control with people with in- from the experiment at any time during the
tellectual disabilities or autism, but they session. An Apple Macintosh IIsi computer
might exhibit selective stimulus control with with a keyboard and mouse was used to
more complex stimuli. CC errors might be present the visual and auditory stimuli. The
one example of selective stimulus control in visual experimental stimuli were displayed
nondisabled children (Lee & Sanderson, on a color screen (20 cm by 15 cm), and
1987; Treiman, 1993). the auditory experimental stimuli were re-
The aim of the current research was to corded in Hypercardt and attached as a re-
reduce the number of initial CC errors made source file to the main program. Five differ-
by nondisabled children by training word ent screen events, intended to give positive
discriminations using multiple- and critical- reinforcement and consisting of differing
difference methods and testing transfer to combinations of flashing colors, moving pic-
constructing the words from individual let- tures, and sound, were constructed.
ters. The critical-difference condition trained The child responded to the experimental
the children to identify both letters of the stimuli by positioning an arrow on the
CCs, whereas in the multiple-difference con- screen with the mouse and then clicking the
dition, the children needed to identify only mouse on the key he or she had chosen. In
the first letter of the word. The matching- the training sessions, the child selected one
to-sample task used to train the words was of three words presented on the screen. In
similar to Dube et al.’s (1991), except that the constructed-response tests, he or she
the sample stimuli were spoken words rather chose letters to spell the stimulus word. The
than pictures. The question was whether the top portion of Figure 1 shows the training
critical-difference and multiple-difference screen for preliminary training with three-
matching-to-sample discrimination training letter words. In the experimental training
72 BEVERLY BIRNIE-SELWYN and BERNARD GUERIN

Table 1
Word Lists for Multiple-Difference and
Critical-Difference Training Sessions

Critical-difference Multiple-difference
training comparison training comparison
Word pool words pool words
num- Correct
ber word 1 2 1 2

1 SNOW SLOW SNAP NICE REST

2 SKIM SWIM SKED MUCH WALK
3 SLAT SPAT SLIP GIRL MINE
4 STAY SPAY STIR FIND TURN
5 SHOW STOW SHED LIVE HAND
6 SCAB STAB SCUM SEED HAIR
7 SMOG SLOG SMEW PINT LEFT
8 WHEN WREN WHAT LONG PART
9 WRIT WHIT WRAP SUCH FELT
10 PLAY PRAY PLOW WENT SOFT
11 BLEW BREW BLOT SANG GENT
12 BRED BLED BRAN TOLD SHIP
13 CHAP CLAP CHEF BEST WELL
14 CLIP CHIP CLUB GONE WORK
15 FROG FLOG FRET READ JUST
16 FLAY FRAY FLIP BOOK TOWN
17 GROW GLOW GRAB BIKE MIST
18 GLAD GRAD GLUM FORK JUMP
19 GNAT GHAT GNAW HERE FISH
20 GWEN GLEN GWAT COST ROAD
21 CROP CHOP CRAB BABY GAVE
22 THEY TREY THIS COOK KING
23 TWIN THIN TWAY FROM HOUR
24 TRIG TWIG TRAP HOME WASH

sessions, these were four-letter words. The

bottom portion of Figure 1 shows the screen
for the constructed-response tests.
Development of Word Lists
Two word lists were compiled, one each
for the multiple-difference and critical-dif-
ference training conditions. All words are
listed in Table 1, which shows the list of the
correct words and the two comparison words
Figure 1. Computer screen layouts. The top por- that appeared in the selection pool. A total
tion depicts the screen layout for preliminary training. of 24 words were used; each word contained
The critical-difference and multiple-difference training four letters, with two initial consonants, a
conditions were identical except that four-letter words
were shown instead of three-letter words. The bottom
vowel, and another consonant (CCVC). The
portion shows the screen for the constructed-response same words were used in both critical- and
tests. multiple-difference training conditions, but
the two comparison words in the selection
pool differed between conditions.
As far as possible, the 24 words with ini-
 STIMULUS CONTROL OF SPELLING ERRORS 73

tial CCs were common English words that Preliminary Screen Training
have been found to result in spelling errors The object of the preliminary training was
(Treiman, 1993), but this was heavily con- to teach the children to respond appropri-
strained by the availability of three critical- ately to the auditory stimuli by clicking the
difference comparison words. The compari- mouse on screen words or letters. This train-
son words in the selection pool for the two ing used five three-letter words with no CCs,
training conditions were based on different but was otherwise identical to the matching-
criteria. In the critical-difference training to-sample discrimination training and con-
sessions, the comparison words were mini- structed-response test conditions described
mally different from the correct word that below. The children were taught in three
was being trained. Only the second conso- phases: first, to select the spoken letter from
nant in the first comparison word differed a selection of ten letters; second, to select a
from the correct word, and the vowel and spoken three-letter word from a pool of
final consonant were different in the second three words; and last, to select the correct
comparison word. For example, when the letters to spell those same three-letter words.
correct word was SNOW, the first compar- On the first trial of each phase, only the
ison word was SLOW and the second com- correct words or letters were available for se-
parison word was SNAP. In order to adhere lection by making the other two gray and
to these selection criteria, four of the selec- inoperative. Therefore, the children got all
tion pool words were nonsense words, al- five correct. On the second trial, all words
though they were phonetically correct in En- or letters were black and could be selected,
glish: SKED, SMEW, GWAT, and TWAY. but the correct one shook in an obvious
In the multiple-difference training sessions, manner as a prompt. After five correct words
the first and second comparison words in the or letters in a row, the third trial began with
selection pool differed completely from the all words or letters available and with no
correct word. The comparison words all prompts. When a child made five consecu-
contained four letters and could include ini- tive incorrect responses, the second trial pro-
tial, final, or no CCs. For example, when the cedure reintroduced with its obvious
correct word was SNOW, the two compar- prompts. All children completed this entire
ison words were NICE and REST. preliminary screen training in less than 50
min, at which point they had been trained
PROCEDURE to select the question mark for the presen-
Each child was given an initial oral spell- tation of the spoken stimulus words and
ing test consisting of the 24 training words. then to select words or letters by using the
No reinforcement was given for correct or mouse.
incorrect responses. Words that were spelled
correctly in this initial oral test were not in- Experimental Conditions
cluded in the experimental conditions for At the beginning of the first session, con-
those children. Out of the 24 words, the structed-response tests were given (0 in Ap-
children answered only 5, 6, 3, 2, 9, and 6 pendix) of the three critical-difference con-
correctly (Anna to Fiona, respectively). They dition words and the three multiple-differ-
made 10, 13, 19, 26, 9, and 12 CC errors ence condition words about to be trained—
(Anna to Fiona, respectively). The Appendix words that the children had spelled incorrectly
shows the incorrect spellings of those words on the initial oral test (P in Appendix).
chosen for training (P). Each session consisted of 30 training pre-
74 BEVERLY BIRNIE-SELWYN and BERNARD GUERIN

sentations, five for each of the six words be- response. The word that was selected ap-
ing trained at any one time. At any time, peared in the box just above the selection
there were usually three words being trained pool. The child then clicked on the END
using the multiple-difference condition and key to indicate that he or she had finished
three words using the critical-difference con- that presentation. Three seconds of blackout
dition. Each session of 30 training presen- were presented after an incorrect response; a
tations was immediately followed by six correct response received one of the five dif-
nonreinforced constructed-response tests, ferent feedback screens. The next trial fol-
one for each of the six words being trained lowed immediately.
in that session. Occasional probes on words Six words were presented to each child
to be trained in future sessions were also giv- five times each session. Three words were
en after the constructed-response tests (see trained with each of the two different train-
Appendix, lower case words). Usually, two ing conditions, and no word was trained in
sessions were completed each day, with a both conditions for any 1 child. The train-
break of an hour or more between sessions. ing condition used for a word was varied
The selection pool for training consisted of across the 6 children. Three children (Bella,
whole words, whereas in the constructed-re- David, and Fiona) received training on
sponse test condition, the selection pool words selected from the first 12 words in
consisted of 10 letters from which to con- Table 1 by the multiple-difference method
struct (spell) the sample word. and on words from the second 12 words by
Multiple-difference and critical-difference the critical-difference method. The words
training presentations. The operating condi- for the other 3 children (Anna, Colin, and
tions, consequences for feedback, blackout, Elena) were the reverse: critical-difference
and the layout of the computer screens were training on the first 12 words and multiple-
the same for both the multiple-difference difference training on the second 12 words.
and the critical-difference conditions. The The actual words used depended upon each
only differences between the two training child’s initial oral tests. The training condi-
methods were the incorrect comparison tions were also alternated within each ses-
words that appeared for selection during sion. For example, the first word might re-
training. ceive multiple-difference training, the sec-
The children were presented with the ond critical-difference training, the third
screen shown at the top of Figure 1. They multiple-difference training, and so on.
first clicked on the question mark (an ob- Constructed-response transfer test. After
serving response) to produce a ‘‘speech bub- completing the 30 training presentations of
ble’’ from the face on the screen and the each session, the six words being trained at
auditory stimulus. The auditory stimulus that time were each tested once. No feed-
consisted of an oral presentation of the word back or blackout conditions were presented
being trained and a sentence containing that during this test. The computer screen layout
word (e.g., ‘‘Snow. [pause] Snow is white.’’). for the constructed-response test condition
The child could repeat the observing re- is shown at the bottom of Figure 1. The
sponses before responding. The child then auditory stimuli were the same as in the crit-
selected the word from the selection pool ical- and multiple-difference training condi-
that matched the auditory stimulus. Only tions. The child was required to click on the
one word from the selection pool could be question mark to produce the auditory stim-
chosen, because all the words in the selection ulus and then to construct a word by select-
pool became inoperable (gray) after the first ing up to six letters from the selection pool
 STIMULUS CONTROL OF SPELLING ERRORS 75

area. The letters appeared in the construct- the cluster (excluding reverse CC errors)
ed-response box just above the selection pool (e.g., SENOW instead of SNOW).
after they were selected. The END key was
again in operation, and no more than six
letters could appear in the constructed-re- RESULTS
sponse box. The letters in the selection pool MATCHING-TO-SAMPLE
became inoperable (gray) when they had DISCRIMINATION TRAINING
been selected; in this way a child could not Overall Percentage Correct in Training
select the same letter twice.
The top panel of Figure 2 shows the per-
Data from the constructed-response test
centage of correct responses made by each
condition were analyzed at the end of each
child during the training presentations, with
session. When a child had met criterion by
the results separated for the critical-differ-
constructing a word corrrectly for three con-
ence and multiple-difference conditions.
secutive sessions, that word was dropped and
Children selected whole words during this
another word to be trained using the same training, and did not construct words from
type of condition was introduced. All new individual letters. The data illustrate that a
words were pretested again before training. greater percentage of correct responses were
If a word was spelled correctly, it did not recorded in the multiple-difference condi-
undergo training, and pretesting continued tion than in the critical-difference condition
until a new word was not correctly spelled. for all 6 children.
That word was then introduced in training.
CONSTRUCTED-RESPONSE
Dependent Measures TRANSFER TESTS
The results that were of most interest were Overall Percentage Correct
(a) the percentage of correctly selected words The middle panel of Figure 2 shows the
during training, (b) the number of sessions percentage of correct responses made in con-
required to learn to construct the words cor- structed-response tests, separated by condi-
rectly, (c) the percentage of words spelled tion. The data illustrate that the percentage
correctly during the constructed-response of correct responses was higher for all chil-
tests, and (d) the percentage of CC errors dren in the critical-difference condition than
during the constructed-response tests. CC in the multiple-difference condition.
errors were categorized as follows. A CC er-
ror was the incorrect spelling or omission of Number of Training Sessions to Meet
the first or second consonant in each word. Constructed Word Criterion
A first omission CC error was the complete The Appendix depicts the complete in-
omission of the first consonant in any given dividual data for the constructed-response
word (e.g., NAP instead of SNAP). A sec- tests over all sessions separately for the two
ond omission CC error was the complete conditions. Words shown in capital letters
omission of the second consonant in any were being trained when a constructed-re-
given word (e.g., WEN instead of WHEN). sponse test was given. From this, the number
A reverse CC error was the reversal of the of training sessions required for each word
second consonant and third vowel in any can be found. For example, with the critical-
given word (e.g., SONW instead of SNOW). difference training, Anna required three
Finally, an insert error occurred when a letter training sessions for the word SNOW before
was inserted between the two consonants of meeting criterion and 11 training sessions
76 BEVERLY BIRNIE-SELWYN and BERNARD GUERIN

Figure 2. The top two panels show the percentage of words that were correctly selected during training
and constructed-response tests for both the critical-difference (open bars) and multiple-difference (black bars)
training conditions. The bottom panel shows the percentage of words spelled during constructed-response
testing that contained CC errors for both the critical-difference and multiple-difference training conditions.
 STIMULUS CONTROL OF SPELLING ERRORS 77

for the word WHEN. Over the six words probes were correctly spelled without train-
trained with critical-difference training, she ing (SCAB, SHOW, PLAY, SKIM).
averaged 5.3 training sessions to meet crite-
rion. Consonant Cluster Errors
These results show that all children took Of the 180 errors made in this experi-
fewer sessions to learn critical-difference ment, 61% were CC errors. Of these 109
words than multiple-difference words. The CC errors, 30 (27%) were made in the crit-
6 children (Anna to Fiona, respectively) took ical-difference condition and 79 (73%) were
an average of 5.3, 5.4, 4.8, 4.8, 3.7, and 4.7 made in the multiple-difference condition.
sessions per word for the critical-difference Consonant cluster errors for each word. The
training words and 6.0, 7.8, 11.7, 5.7, 5.3, bottom panel of Figure 2 shows the per-
and 6.4 sessions per word for the multiple- centage of words that contained CC errors
difference training words. for both training conditions. The results
Fourteen of the 24 words were trained by show that all 6 children made more CC er-
both methods with at least 1 child per meth- rors in the multiple-difference condition.
od, thus allowing a between-child compari- As can be seen in the Appendix, the chil-
son. For example, 5 children were trained on dren frequently made no CC errors for a
the word SKIM, 2 in the critical-difference word in the critical-difference condition,
condition and 3 in the multiple-difference whereas they made numerous CC errors in
condition. The Appendix indicates that for the multiple-difference condition. Anna’s
9 of these 14 words, children took fewer ses- data show that four of the six words trained
sions to complete training in the critical-dif- in the critical-difference condition recorded
ference condition than in the multiple-dif- no CC errors compared to only one of the
ference condition. Those words were SKIM five words in the multiple-difference condi-
(5, 4 vs. 5, 9, 9), STAY (3, 3, 7 vs. 11, 4, tion. Bella’s data show that in the critical-
8), SCAB (3, 4 vs. 6, 7), WRIT (5, 3, 3 vs. difference condition, four of the seven words
9, 8, 5), BLEW (4, 10, 3 vs. 10, 4, 9), recorded no CC errors, but all six words in
CHAP (4, 3 vs. 7, 6), FROG (6 vs. 10, 7), the multiple-difference condition recorded
GNAT (3, 3, 3 vs. 5, 3, 5), and CROP (3, CC errors. The results for Colin show that
6 vs. 12). Children took fewer sessions to five of the nine words trained in the critical-
complete training in the multiple-difference difference condition recorded no CC errors
condition for only 2 of the 14 words: compared to one out of nine in the multiple-
WHEN (11 vs. 6, 6) and TWIN (9, 9, 7 difference condition. David’s results show
vs. 5, 5, 5). The remaining three words that seven of the nine words in the critical-
(SNOW, FLAY, and TRIG) did not show a difference condition recorded no CC errors
clear difference in the number of sessions re- compared to only three of the nine words in
quired to complete training. the multiple-difference condition. For Elena,
Another result included in the Appendix in the critical-difference condition, five of
is that for both training conditions, children the seven words recorded no CC errors, but
improved their spelling of novel words by in the multiple-difference condition two of
the end of the experiment. For all children, the six words recorded no errors. Fiona’s re-
after six or seven words had been trained, sults show that six of the eight words in the
the novel words presented as untrained critical-difference condition recorded no CC
probes were constructed correctly. For ex- errors, whereas two of the seven words in the
ample, Anna was trained on six words with multiple-difference condition recorded no
critical-difference training, and the next four errors.
78 BEVERLY BIRNIE-SELWYN and BERNARD GUERIN

Table 2 DISCUSSION
The Number of Each Type of CC Errors for Both
Training Conditions This study was based on the suggestion
that CC errors occur because of selective
Percent- stimulus control by one of the consonants.
Critical Multiple age of
Type of differ- differ- Total CC CC The present research compared the results of
errors ence ence errors errors two matching-to-sample discrimination
First omission training conditions—critical-difference and
CC errors 1 7 8 7.3 multiple-difference training—in reducing
Second omission
CC errors 10 39 49 44.0
these spelling errors made by children on a
Reverse CC errors 11 19 30 27.5 constructed-response transfer test. Overall,
Insert CC errors 8 7 15 13.8 the constructed-response test results indicat-
Other CC errors 0 7 7 6.4 ed that the critical-difference training con-
Total 30 79 109 100
dition was more effective than the multiple-
difference training condition in reducing
CC errors and overall errors. Thus finer dis-
Types of consonant cluster errors. Table 2 crimination training at the level of single let-
lists the total number of each type of CC ters led to improved spelling performance,
error made by all the children in the con- as had been suggested by Dube et al. (1991),
structed-response tests. The results show that Lee and Sanderson (1987), and Stromer and
for all CC error types (with the exception of Mackay (1992b).
insert CC errors), more were made for words The number of training sessions required
that had been trained in the multiple-differ- before the constructed-response performance
ence condition than in the critical-difference reached criterion decreased as training pro-
condition. A total of 109 errors were made, gressed under both conditions. This suggests
30 in the critical-difference condition and that the critical-difference training condition
79 in the multiple-difference condition. focused the children’s attention on critical
Only one first omission error was made aspects of the CC combination and that this
in the critical-difference condition, but seven generalized to performance under the mul-
were made in the multiple-difference con- tiple-difference condition. It is very likely
dition. As expected from past research on that performance during the multiple-differ-
initial CC words, the most frequent CC er- ence condition would have been worse in
ror was the second omission error (a total of the absence of any critical-difference train-
49), which accounted for nearly half of all ing. However, because training and time
CC errors recorded. Most of these second were confounded, it is possible that the over-
omission CC errors (80%) occurred in the all improvement in performance was due to
multiple-difference condition. Thirty reverse extraexperimental experiences. The fact that
CC errors were recorded in total, the second some words were constructed correctly in
most frequent type of error. Of these, 63% the absence of specific discrimination train-
were made in the multiple-difference con- ing strengthens the argument for the effect
dition. The combined total of reverse and of such extraexperimental factors.
insert CC errors, both letters of the CC be- The results from the training conditions
ing present but not in the correct sequence, in which whole words were matched to sam-
accounted for nearly half of the CC errors ple showed that a greater number of errors
(45). The two training conditions did not were made in the critical-difference condi-
differ in insert errors. tion than in the multiple-difference condi-
 STIMULUS CONTROL OF SPELLING ERRORS 79

tion. These results therefore show that train- the components of the cluster—the initial
ing with the multiple-difference condition consonant.
was easier than with the critical-difference Another interesting finding was that re-
condition, as would be expected. The data verse and insert CC errors combined ac-
from the training condition alone do not es- counted for nearly half of the CC errors.
tablish the controlling factors that governed Both of these errors involve the sequencing
the children’s behavior during the matching- rather than the occurrence of the two con-
to-sample training because selection was sonants; the consonants were there but were
made from among whole words. However, not in the correct sequence. This means
the fact that more overall errors and CC er- that, as suggested by Stromer and Mackay
rors occurred in the constructed-response (1992b), further discrimination training for
tests with words trained in the multiple-dif- sequencing may be required to overcome
ference condition indicates that the children such CC errors. The training methods used
were not identifying all four letters of the here appeared to reduce only the reverse CC
compound stimulus during training, which errors. The techniques used by Stromer and
resulted in more mistakes being made in the Mackay (1992b) and Stromer, Mackay, Co-
constructed-response tests. These data are hen, and Stoddard (1993) seem to be prom-
consistent with the results of Allen and Fu- ising for reducing sequencing errors.
qua (1985), who used other types of stimuli. The computer tasks were extremely useful
With regard to the types and frequency of in training the spelling of nondisabled chil-
initial CC errors, Treiman (1991) found that dren. However, these children exhibited
second omission errors accounted for 42% some skills that might not have been present
of the CC errors. Other researchers have in the persons with intellectual disabilities
found that second omission errors accounted who participated in previous research (Dube
for 33% of the CC errors (Miller & Limber, et al., 1991; Stromer & Mackay, 1992b),
1985). Although the number of second and supplementary training might be need-
omission errors that occurred in the critical- ed to replicate the present results with such
difference condition of the present study was populations. Further applications of basic
very low compared to the number that oc- stimulus control procedures to spelling
curred in the multiple-difference condition, should result in improved spelling. One ex-
second omission errors accounted for 44% ample is the finer discriminative control of
of the total number of errors, close to that sequencing mentioned above (Green, Strom-
found by Treiman (1991). The second omis- er, & Mackay, 1993).
sion errors in both conditions therefore sub- Of most practical significance was that the
stantiate past findings that the most com- computer-trained spelling, as in the work of
mon error made by children trying to spell Dube et al. (1991) and Stromer and Mackay
words that contain initial CCs is the omis- (1992a, 1992b), did not require direct in-
sion of the second consonant in the cluster tervention by a teacher. If normal reading
(Bruck & Treiman, 1991; Marcel, 1980; was to be supplemented by computer tasks
Miller & Limber, 1985; Treiman, 1991). of the kind used here and elsewhere (Strom-
The higher rate of second omission errors in er & Mackay, 1992b), some spelling prob-
the multiple-difference condition compared lems, which might occur because natural
to the critical-difference condition provides contingencies are not sufficient to shape the
further support that the spelling of CCs in finer discrimination of individual letters,
words from the multiple-difference condi- could be circumvented without costly inter-
tion came under the control of only one of vention by a teacher or a special trainer. The
80 BEVERLY BIRNIE-SELWYN and BERNARD GUERIN

present research suggests that the application ing: What to teach and how to teach it. Journal
of Educational Psychology, 72, 686–700.
of selective stimulus control research to Lovaas, O. I., Koegel, R. L., & Schreibman, L.
spelling merits further attention. (1979). Stimulus overselectivity in autism: A re-
view of research. Psychological Bulletin, 86, 1236–
1254.
REFERENCES Lovaas, O. I., Schreibman, L., Koegel, R. L., & Rehm,
R. (1971). Selective responding by autistic chil-
Allen, K. D., & Fuqua, R. W. (1985). Eliminating dren to multiple sensory input. Journal of Abnor-
selective stimulus control: A comparison of two mal Psychology, 77, 211–222.
procedures for teaching mentally retarded children Marcel, T. (1980). Phonological awareness and pho-
to respond to compound stimuli. Journal of Ex- nological representation: Investigation of a specific
perimental Child Psychology, 39, 55–71. spelling problem. In U. Frith (Ed.), Cognitive pro-
Bruck, M., & Treiman, R. (1990). Phonological cesses in spelling (pp. 373–405). London: Academ-
awareness and spelling in normal children and ic Press.
dyslexics: The case of initial consonant clusters. Miller, P., & Limber, J. (1985, October). The acqui-
Journal of Experimental Child Psychology, 50, 150– sition of consonant clusters: A paradigm problem. Pa-
178. per presented at the annual Boston University
Cuvo, A. J., Ashley, K. M., Marso, K. J., Zhang, B. Conference on Language Development, Boston.
L., & Fry, T. A. (1995). Effect of response prac- Moxley, R. A., & Warash, B. (1990–1991). Spelling
tice variables on learning spelling and sight vocab- strategies of three prekindergarten children on the
ulary. Journal of Applied Behavior Analysis, 28, microcomputer. Journal of Computing in Child-
155–173. hood Education, 2, 47–61.
Dube, W. V., McDonald, S. J., McIlvane, W. J., & Neef, N. A., Iwata, B. A., & Page, T. J. (1980). The
Mackay, H. A. (1991). Constructed-response effects of interspersal training versus high-density
matching to sample and spelling instruction. Jour- reinforcement on spelling acquisition and reten-
nal of Applied Behavior Analysis, 24, 305–317. tion. Journal of Applied Behavior Analysis, 13,
Foorman, B. R., Francis, D. M., Novy, D. J., & Lib- 153–156.
erman, D. (1991). How letter-sound instruction Ollendick, T. H., Matson, J. L., Esveldt-Dawson, K.,
mediates progress in first grade reading and spell- & Shapiro, E. S. (1980). Increasing spelling
ing. Journal of Educational Psychology, 83, 456– achievement: An analysis of treatment procedures
469. utilizing an alternating treatments design. Journal
Gettinger, M. (1985). Effects of teacher-directed ver- of Applied Behavior Analysis, 13, 645–654.
sus student-directed instruction and cues versus Read, C. (1975). Children’s categorization of speech
no cues for improving spelling performance. Jour- sounds in English (NCTE Research Rep. No. 17).
nal of Applied Behavior Analysis, 18, 167–171. Urbana, IL: National Council of Teachers of En-
Gettinger, M. (1993). Effects of invented spelling and glish.
direct instruction on spelling performance of sec- Reynolds, B. S., Newsom, C. D., & Lovaas, O. I.
ond-grade boys. Journal of Applied Behavior Anal- (1974). Auditory overselectivity in autistic chil-
ysis, 26, 281–291. dren. Journal of Abnormal Child Psychology, 2,
Green, G., Stromer, R., & Mackay, H. A. (1993). 253–263.
Relational learning in stimulus equivalence. The Schover, L. R., & Newsom, C. D. (1976). Overse-
Psychological Record, 43, 599–616. lectivity, developmental level, and overtraining in
Greenwood, C. R., Dinwiddie, G., Bailey, V., Carta, autistic children. Journal of Experimental Child
J. J., Dorsey, D., Kohler, F. W., Nelson, C., Roth- Psychology, 4, 289–298.
olz, D., & Schulte, D. (1987). Field replication Schreibman, L., Koegel, R. L., & Craig, M. S. (1977).
of classwide peer tutoring. Journal of Applied Be- Reducing stimulus overselectivity in autistic chil-
havior Analysis, 20, 151–160. dren. Journal of Abnormal Child Psychology, 5,
Griffith, P. L. (1991). Phonemic awareness helps first 285–289.
graders invent spellings and third graders remem- Stevens, K. B., Blackhurst, E., & Slaton, D. B.
ber correct spellings. Journal of Reading Behavior, (1991). Teaching memorized spelling with a mi-
23, 215–233. crocomputer: Time delay and computer-assisted
Lee, V. L., & Pegler, A. M. (1982). Effects on spelling instruction. Journal of Applied Behavior Analysis,
of training children to read. Journal the Experi- 24, 153–160.
mental Analyses of Behavior, 37, 311–322. Stromer, R., & Mackay, H. A. (1992a). Delayed con-
Lee, V. L., & Sanderson, G. M. (1987). Some con- structed-response identity matching improves the
tingencies of spelling. The Analysis of Verbal Be- spelling performance of students with mental re-
havior, 5, 1–3. tardation. Journal of Behavioral Education, 2, 139–
Lewkowicz, N. K. (1980). Phonemic awareness train- 156.
 STIMULUS CONTROL OF SPELLING ERRORS 81

Stromer, R., & Mackay, H. A. (1992b). Spelling and Treiman, R. (1993). Beginning to spell: A study of first-
emergent picture-printed word relations estab- grade children. New York: Oxford University Press.
lished with delayed identity matching to complex Treiman, R., & Zukowski, A. (1988). Units in read-
samples. Journal of Applied Behavior Analysis, 25, ing and spelling. Journal of Memory and Language,
893–904. 27, 466–477.
Stromer, R., Mackay, H. A., Cohen, M., & Stoddard, Treiman, R., & Zukowski, A. (1990). Toward an un-
L. T. (1993). Sequence learning in individuals derstanding of English syllabification. Journal of
with behavioural limitations. Journal of Intellectual Memory and Language, 29, 66–85.
Disability Research, 37, 243–261.
Stromer, R., Mackay, H. A., & Stoddard, L. T.
Received October 26, 1995
(1992). Classroom applications of stimulus equiv-
Initial editorial decision December 20, 1995
alence technology. Journal of Behavioral Education,
Final acceptance October 25, 1996
2, 225–256.
Action Editor, Joseph E. Spradlin
Treiman, R. (1991). Children’s spelling errors on syl-
lable-initial consonant clusters. Journal of Educa-
tional Psychology, 83, 346–360.

STUDY QUESTIONS

1. What are words containing consonant clusters (CCs), and how are these words typically
misspelled?
2. Describe the rules used for generating the experimental words and both the critical-difference
and multiple-difference comparison words. Given shot as an experimental word, provide
examples of its multiple-difference and critical-difference comparisons.
3. Describe the basic procedural difference between the training trials and the constructed-
response trials.
4. How did the authors control for the different training approaches used?
5. What types of CC errors did the authors record?
6. Summarize the results in terms of (a) percentage correct during training and constructed
response trials, (b) number of trials to criterion, and (c) CC errors.
7. What explanation was provided to account for the performance differences that were ob-
served during training versus testing?
8. The authors suggested that performance during the multiple-difference condition might have
been worse than that obtained had the children not been exposed to critical-difference
training. How might the authors have controlled for the influence of critical-difference
training?
Questions prepared by SungWoo Kahng and Michele Wallace, University of Florida
82 BEVERLY BIRNIE-SELWYN and BERNARD GUERIN

APPENDIX
Individual data for the initial oral test (P) and constructed-response test sessions, shown separately for the
critical-difference and multiple-difference training conditions. At the beginning of the first session, the three words
for training were retested (0). Words that were being trained at the time of any test are shown in capitals. Occasional
probes for the next words to be trained are shown in lower case letters. Criterion was three correct consecutive
spellings, at which point a new word was introduced to training.

Participant Session

Anna
Critical-difference training
P sow win bid sda rat
0 sow wen bed
1 SNOW WN BED
2 SNOW WEN BED sea
3 SNOW WHEN BED
4 WEN BRED STAY ret
5 WHEN BRED STAY
6 WHEN BRED STAY
7 WHIN WRIT
8 WIEN WRET
9 WHEN WRIT
10 WHEN WRIT
11 WHEN WRIT
Multiple-difference training
P sap fog nat twn gwin
0 cip fog nat
1 CAP FOG NAT
2 CHAP FOG GAT
3 CAP FOG GNAT
4 CAP FROG GNAT twn
5 CHAP FROG GNAT gwen
6 CHAP RFOG TIWN
7 CHAP FRO TWN gwin
8 FROG TWIN GWEN
9 FROG TWIN GWEN
10 FROG TWIN GWEN
11
Bella
Critical-difference training
P cep nat towen fog trwr
0 cip nat towin
1 CLIP GNAT TEWIN
2 CLIP GNAT TIWEN fog
3 CLAP GNAT TIWN thrig
4 CLIP TWIN FROG
5 CLIP TEWIN FROG
6 CLIP TIHWN FOG
7 TWIN FROG TIRG
8 TWIN FROG TRGE
9 TWIN FROG TRIG
10 THRG
11 TIRG
12 TRIG
13 TRIG
14 TRIG
 STIMULUS CONTROL OF SPELLING ERRORS 83

APPENDIX
(Extended)

bli sca s pla scm

show

blewu play skim

BLW scab
BLEW show play skim
BLEW siow play
BLEW scab show
scab show skim

clp fla tha gro

they

clip flay grow

clip flay thay grw

clip grow
flay they
they grow

flak rrop glrad gren thay

they
glad gwen

flye crihp

glad gwen
they
FLAY CROP
FLAY CROP
FLAY CROP glad gwen
they
84 BEVERLY BIRNIE-SELWYN and BERNARD GUERIN

APPENDIX
(Continued)

Participant Session

Multiple-difference training
P sgem wene bole rit seteay
0 sim wen blu
1 SEIM WEN BLOW
2 SIGM WHEN BEW wrt wsta
3 SKIM WEN BLWE
4 SKIM WHEN BIEW
5 SKIM WHEN BELW
6 WHEN BEW WIT
7 BLWE WEIT SITY
8 BLEW WIRT STYE
9 BLEW WRIT STAY
10 BLEW WRIT SITY
11 WIRT STIY
12 WRIT STAY
13 WRIT SATY
14 WRIT SATY
15 STAY
16 STAY
17 STAY
Colin
Critical-difference training
P s low soc sme reat
0 so bw sok
1 SNOW BOLW SMOG
2 SNOW BLOW SMRG
3 SONW BLEW SMOG skm
4 SNOW BLAW SMOG bied
5 SNOW BLEW SMOG
6 SNOW BLIW SIKM
7 BLOW SIKM BRED
8 BLEW SKIM BDEM
9 BLEW SKIM BRED
10 BLEW SKIM BRED
11 BRED
12
13
14
15
16
17
Multiple-difference training
P wooi decf copi foc mit
0 roiw rik cop
1 GOW TREG CROP
2 GROW TERG GOP
3 GORW GTR CORP forg
4 GROW TREG CROP
5 GROW TROG CROP nat
6 GROW TREG CORP
7 TROG CORP FORG
8 TRIG CORP FORG
 STIMULUS CONTROL OF SPELLING ERRORS 85

APPENDIX
(Extended - Continued)

sgarb selet brerde

sacb slat bred

SCBE clat bred

SECB
CGRB bred
SCAB slat
SCAB slat
SCAB
slat

sti whit sab sow sat

sta

rit slat
STAY WRIT caeb seow
STAY WRIT SCAB
STAY WRIT SCAB sait
SCAB SHOW slat
SHOW
SHOW slat
slat

tin ilep ceni dladi fae thai

tein
86 BEVERLY BIRNIE-SELWYN and BERNARD GUERIN

APPENDIX
(Continued)

Participant Session

9 TRIG CORP FORG

10 TREG CORP FORG
11 TRIG CROP FROG
12 TRIG CROP FROG
13 TRIG CROP FROG
14 GNAT
15 GNAT
16 GNAT
17
18
19
David
Critical-difference training
P croiw corp tewn cilp tha
0 grw cop tiwen
1 GROW CROUP TEWIN
2 GROW CROP TIWN
3 GROW CRUP TAWIN cihp toey
4 CROP TWIN CLIB
5 CROP TWIN CLP
6 CROP TEWIN CLIP
7 TWIN CLIP THAY
8 TWIN CLIP THEY
9 TWIN THEY
10
11 THEY
12
13
14
15
Multiple-difference training
P sgi soiw wen pla beed
0 skm soiw wen
1 SKM SOW WEN
2 SIM SHOW WEN
3 SKM SHOW WIEN pla bied
4 SKM SOW WHEN
5 SKIM SHOW WHEN
6 SCIM SHOW WHEN
7 SKIM SHOW PLAY
8 SKIM PLAY BED
9 SKIM PLAY BED
10 BED
11 BRED
12 BRED
13 BRED
14
15
16
17
 STIMULUS CONTROL OF SPELLING ERRORS 87

APPENDIX
(Extended - Continued)

cilp
gaed flay
thay
TIN CIP gwen giad flay
TIN CLIP GEWN GLAD they
TWIN CILP GEN GAD flay
TWIN CLIP GWEN GLAD they
TWIN CLIP GWEN GLAD
CLIP GWEN GLAD flay they

cap gen nat gilad flaa trik

cehap win
nat glad
CAP glad
CHAP GWN flay tig
CHAP GWN GNAT glad
CHAP GWEN GNAT flay TRG
GWEN GNAT TRIG
GWEN flay TRIG
TRIG

rit bliw sta so siat sab

rit bw
stae slat
sw scab
RIT BLUEW slat
RIT BLEW scab
WRAIT BLEW slat
RIT BLEW
RIT STA SOW scab
WRIT STAY SNOW
WRIT STAY SNOW
WRIT STAY SNOW
88 BEVERLY BIRNIE-SELWYN and BERNARD GUERIN

APPENDIX
(Continued)

Participant Session

Elena
Critical-difference training
P sa blue rite scm bed
0 sae ble rit
1 STAE BLEW WRIT
2 SATY BLEW WRIT
3 STAY BLEW WRIT sm bed
4 STA SKM BRED
5 STAY SKIM BRED
6 STAY SKIM BRED
7 STAY SKIM
8
9
10
11
Multiple-difference training
P sap nat tri flai tin
0 cab nat tik
1 CAP GNAT TIG
2 CHAP GHAT TRIC
3 CAP GNAT TRIG flai twn
4 CHAP GNAT TRIG
5 CHAP GNAT TRIG
6 CHAP FLAE TWIN
7 FLAI TIWEN
8 FLA TWIN
9 FLAY TWIN
10 FLAY TWIN
11 FLAY
Fiona
Critical-difference training
P fil tik gad lip gwn
0 fai tik gad
1 FLAI TIG GLAD
2 FLAI TRI GLAD
3 FILAY TRIG GLAD cilp gwenm
4 FLAY TIRG CLIP
5 FLAY TRIG CLP
6 TRIG CLIP
7 TRIG CLIP GWEHN
8 CLIP GWEN
9 GWEN
10 GWEN
11
12
13
14
 STIMULUS CONTROL OF SPELLING ERRORS 89

APPENDIX
(Extended - Continued)

plai sab smo

plai

scrb smog
PLAY
PLAY SAB
PLAY SCAB smog
SCAB smog
SCAB

wen cip

gwin
clip
GWEM
GWIN
GWEN
GWEN clip
GWEN clip

twn fog chab nat

twnm frog
cap
nat
TWN frog
TWUN CHAP
TWIN CHAP
TWEN CHAP GNAT
TWIN frog GNAT
TWIN GNAT
TWIN
90 BEVERLY BIRNIE-SELWYN and BERNARD GUERIN

APPENDIX
(Continued)

Participant Session

Multiple-difference training
P st blu sgb sm rit
0 sate blu cab
1 STA BLW SAB
2 STAI BILW SCAB
3 STAY BLUW SAB sm
4 STAE BLIW SGAB rit
5 STAE BLEW SCAB
6 STAY BLIW SCAB
7 STAY BLEW SCAB
8 STAY BLEW SGIM
9 BLEW SKIM RIT
10 SKM RIT
11 SKM WRIT
12 SKIM WRIT
13 SIM WRIT
14 SKIM
15 SKIM
16 SKIM
 STIMULUS CONTROL OF SPELLING ERRORS 91

APPENDIX
(Extended - Continued)

sat smoc sow bireed

sat show
smok
bred

SLAT show bred

SLAT
SLAT
SOG
SMOG show bred
SMOG
SMOG