The generative language matrix: A comprehensive clinical analysis of generative language classes,
conditional discriminations, ecobehavioral functions, abstract comprehension, and natural language
development
Eric V. Larsson, Kara Riedesel, Angela Keene, Leslie Davis
LIFE, University of Minnesota, University of Kansas
April 30, 2003
Permission to reproduce must be requested from the authors at:
Eric V. Larsson, Ph.D.
LIFE Midwest
2925 Dean Parkway, #300
Minneapolis, MN 55416
elarsson@lovaas.com
Generative Language Matrix
Page 2
The treatment of autism
Natural language and social behavior may be analyzed in many ways. When the purpose of the
analysis is to develop natural social and language skills, then it is helpful to integrate the analysis into
a comprehensive organizational framework. In intensive early intervention with young children with
autism, the great variety of natural language skills can be developed within such a coherent
conceptual framework. As a result, the framework will enable productive treatment planning and
program evaluation, as well as efficient language programming. Intensive therapy will be most costeffective when both language and social skills are developed through an organized, as opposed to
disordered, curriculum. The mark of an integrated curriculum is that it has content and coherent
validity. Given that natural social behavior is highly inter-dependent with language skill, the valid
curriculum should integrate both areas of skill development.
In the assessment of the needs of a child who has been diagnosed with autism, three areas of
development typically appear to deviate from the normal range. The child will have a distinctly
atypical developmental repertoire of both language and social behavior. The child will also show
some form of stereotyped behavior. Each child will show a highly individualized pattern of these
developmental deviations (Committee on Children With Disabilities, 2001). Therefore the
curriculum should integrate a therapeutic approach to addressing all three areas, but in a framework
that allows for significant individualization, as opposed to a simple progressive cookbook.
In order to attain the best possible outcomes for the child, behavior therapy should result in natural
patterns of behavior in all of the typical environments of childhood. If these outcomes are thought
of in terms of the natural repertoire of a typical six-year-old child, the goals of such treatment can be
described generally as follows. The child will empathize and share affection with his family. The
child will independently make and keep mutual friendships. The child will succeed independently in
school. The child’s social interaction will both be responsive and dynamic. The child will be
effective in social situations. Finally, the child will meet the natural expectations for social behavior
and self-control of their behavior in the environment. The child will possess all of these typical
behavior patterns in the appropriate environments without requiring specialized supports. Instead,
these typical behavior patterns will independently arise in the context of the natural ecology of the
environment.
To more operationally define these general aspects of typical development, the treatment goals are
for the child to show developmentally typical patterns of the following behaviors.
o Generalized imitation
o Generalized compliance
o Distal compliance
o Response to novel adults
o Compliance with group instructions
o Attending in small and large groups
o Intelligible speech
o Generalized speech
o Colloquial speech
o Creative story telling
o Social comprehension
o Cooperative play with adults
o Cooperative play with peers
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
o
o
o
o
o
o
Page 3
Congruent social play
Maintenance of mutual friendships
Participation in group conversations
Thorough social responsiveness
Self control of stereotypy
Self control of excessive activity levels
When a child has mastered these goals, by spontaneously using these skills as appropriate in the
natural environment, he is likely to have reached a point at which he is independent in the typical
environments of childhood. He can enter and graduate from first grade without requiring special
supports. He attains all of the typical developmental milestones. He can attain normal scores of
cognitive functioning in standardized testing environments. He does not qualify for any diagnosable
mental disorder. He responds equally appropriately in all environments, and with his parents, peers,
siblings, and all natural care-providers, as well as with trained therapists. He does not require
ongoing specialized therapy to develop further. These results have been found when behavior
therapy is delivered as intensive early intervention. These outcomes might be the best possible, and
in published research, children who attain these goals have a high likelihood of maintaining
independence throughout childhood (McEachin, Smith, & Lovaas, 1993).
In order to accomplish these ambitious aims, a large body of research in Applied Behavior Analysis
has been pursued over the past 40 years. The implication of this research is that behavior constantly
responds to its environment. This interaction of behavior and environment results in all of the
learned behavior patterns exhibited in a child’s life. Whether interactions with parents and staff are
planned or unplanned in their intent, these responses still interact with the child’s behavior patterns
to result in development. The developing behavior can follow a path that results in autistic
behavior. In the development of autistic behavior, stereotyped tendencies generally become
stronger as a result of the interaction of autistic behavior patterns with the environment throughout
the 24-hour day. Inconsistent treatment approaches do not readily alter this dysfunctional pattern
nor do they result in progress toward normalcy. However, consistent treatment is regularly shown
to result in appropriate development. Therefore, to alter the developmental course of autism,
treatment must transform the child’s home and community into a 24-hour therapeutic ecology. The
most well-proven therapeutic ecology is based upon the results of Applied Behavior Analysis.
More specifically, to transform the autistic behavior patterns into typical social behavior patterns,
effective treatments must establish natural behavior interactions in their typical environments.
Behavior therapy is most effective when the clinical focus is upon treating dysfunctional behavior
problems, such as stereotyped tendencies, which are functioning to prevent typical development.
To eliminate these behavior problems is then to result in accelerated progress toward typical
development. In addition to the remediation of these dysfunctional patterns, behavior therapy will
also entail a direct teaching approach which establishes the prerequisite skills necessary for natural
behavior patterns to emerge. Therefore, effective behavior therapy is a two-fold process. Not only
must typical skills be established through direct teaching, but also, for these skills to develop
naturally and be used effectively in the natural environment, the child’s clinical behavior problems
must be remediated. As such, there are both structural and functional goals in behavior therapy.
Regarding the clinical focus, a large number of the dysfunctional behavior problems have been
identified in clinical practice. These behavior problems include:
o Rigidity
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
o
o
o
o
o
o
o
o
o
o
o
Page 4
Preference for routine and sameness
Perseveration
Frustration avoidance
Retreat from overwhelming complex environments
Social disengagement
Failure to learn from the environment
Ecological effectiveness of autistic behaviors
Self-motivation as opposed to social orientation
Stereotyped reinforcement
Oppositional behavior
Generalization decrements
Each of these clinical behavior problems may be addressed through a task analysis of the individual
child’s specific needs in the area. Then the task analysis is followed as a series of individualized
interventions. The interventions are modified in response to a prescriptive analysis of their
effectiveness. The clinical interventions are interdependent upon the direct teaching of skills that
are also needed to help the child function in the natural environment.
An example of a sequence of intervention that is designed to address one clinical behavior problem,
rigidity, involves the following general steps. Before treatment, the child’s rigidity is interfering with
the normal development of social skills, as the child actively refuses participation in unfamiliar
activities. Then, when treatment is begun, a waiting skill is developed through proactive teaching.
Then tolerance of events, which currently provoke autistic behaviors, is developed by exposing the
child to the provocative stimulus, and then reinforcing calm waiting behavior, through providing the
opportunity to escape the stimulus after waiting. A variety of generalized forms of patience, such as
waiting for a requested reinforcer, waiting in community activities, and independent work, are then
developed through a gradual shaping process. Then flexibility in daily activities is developed
through differential reinforcement and observational learning. Finally, the skills involved in empathy
with others are developed. The empathy skills are elaborated to the point where the child will act to
meet the needs of others as well as themselves. Of course, while the component language and social
skills might be taught directly, the clinical need is not met unless the skills are established as typical
behaviors in the natural environment. The child is able to use these skills when the rigidity is not
preventing their function in the environment. In order for a skill to be considered to be mastered,
the skill must occur independently of specialized therapy, and must effectively supplant the clinical
behavior problem that had previously prevented natural development.
The development of natural language
The child with autism will typically also have substantial needs in the area of language development.
In many children, the tendency toward stereotyped forms of language will be so great as to interfere
with normal social communication and development.
The structure of behavior. In daily interaction, the child will show a highly individualized
tendency to make certain stereotyped language responses, and these stereotyped language responses
will occur at different rates. Generally each of these stereotyped structures will be rooted in rote
imitations of language that the child has heard. And, whether produced contextually or not, they
will not show evidence of true understanding of the meaning of the words used, because they will
often recombine into nonsensical or simple sentence errors. Because of the stereotyped nature of
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 5
the child’s language, the errors will then often become perseverative error patterns. Examples of
these stereotyped language structures include the following.
o Word omissions
 “I a car.”
o Filling nonsense sounds or words into a phrase
 “Look at the uba uba uba truck.”
o Incorrect or over-generalized articles
 “Go to a kids.”
o Pronoun errors
 “You give it to you.”
o Plural agreement errors
 “The colors is red.”
o Word salads
 “Balls red cups heavy.”
o Substitutions
 “The train is train.”
o Tense errors
 “The boy running.”
o Conjunction errors
 “It’s a big and red and tent.”
o Noncontextual statements
 “What’s your name?” “My address is 2492 Lake St.”
o Word order errors
 “Dogs bone eating.”
o Word association errors due to being in the wrong context
 “Terry is brushing his ‘hair’ (instead of teeth).”
The function of behavior. But beyond the immediate description of the structure of the behavior,
the function of the behavior may also be evaluated. Not only will language be defined by its
structure, but it will also be defined by its function. More specifically, different forms of language
will be found to be response classes that are controlled by a functional relationship with
environmental stimuli. A variety of functions of stereotyped language may be identified. These
include:
o Errors are shown to occur in a noncompliant pattern.
o Emotional responses occur when the stereotyped language behavior is interrupted.
o A response does not generalize from one stimulus to another, as if the response had been
learned by rote practice.
o Isolated errors occur in a perseverative pattern in an otherwise mastered response class.
o When an unmastered, or difficult task is presented, predictable errors are substituted for
correct responses in a perseverative pattern.
o Simple perseverative overuse of a word or phrase.
 A word has local momentum and continues to be used at a high rate.
o Over-generalized sentence structure in multiple contexts.
 “’it’s a’ we go outside now.”
o Overuse of a specific word in specifiable sentences or contexts.
o Using an incorrect (but often learned) sentence form in a predictable context.
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 6

o
o
o
o
o
o
o
Making a request by asking a question, “do you want to go outside?” (in this example,
the sentence form is learned by direct imitation)
A word association controls subsequent local language behavior.
 Pivoting on a word: “Let’s go outside is Dan’s car.”
Not responding to the entire context or message.
 A response is controlled by a single word in a sentence rather than the entire statement:
Instead of “I like your shirt,” “Thanks;” the following occurs: “It is a blue shirt,”
“Thanks.”
Overgeneralization of a response when learning a skill.
 The last exemplar mastered is over-used as an error during learning of the subsequent
exemplar: “What school do you go to?” “I am six years old.”
Generalized noncompliant statements.
Attention-getting response classes.
Escape behavior.
Behavior that is reinforced by coercing tangible reinforcement.
Some functions may be very difficult to discern without repeated assessment. This is especially the
case with socially nonresponsive language behavior – behavior that does not directly respond to
specifiable social stimuli, while still showing a functional relationship to the stimuli. Observable
functions across time include:
o The correct form is not imitated after a model, even though imitation has already formed as
a response class.
o The same error is repeated after the child had complied with a correction procedure for the
error.
o While engaged in the stereotyped language behavior, there is no response to another social
stimulus, such as an appropriate interruption.
o A stereotyped error in response to a specific stimulus is manipulable by differential
reinforcement.
Behavior therapy for natural language development. The functions of autistic language, that are
described above, may cause clinical programming to be ineffective. As a result, the child may make
slow progress with a high error rate; have significant levels of stereotyped language patterns; fail in
effective communication in natural interactions; and lack maintenance of skills. Therefore, to avoid
these problems, the objectives of behavior therapy should be natural language responses which both
have typical functions in the environment and typical structures. To attain these objectives,
behavior therapy may follow this sequence. Behavior therapy is initiated by building on strengths –
generalizing simple forms of language and expanding the vocabulary within those forms as early in
the child’s development as possible. (In this initial phase, both expressive and receptive single-term
labels and requests are developed into generative response classes). Then the individual language
terms are combined into generative conditional discriminations, or sentences. Once conditional
discriminations begin to reach the generative stage, more advanced language forms are developed as
single terms (possessives, plurals, etc.), and other language forms are developed (reciprocals, social
initiations, etc.). Then concrete visual and auditory comprehension skills are developed. After that,
abstract auditory comprehension skills are developed, along with abstract production skills.
Throughout this process, the language forms that are acquired are generalized to function in natural
activities, while simultaneously, the 24-hour ecology of the stereotyped language is altered to weaken
the stereotyped language and replace it with typical language forms. Finally these natural language
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 7
skills are generalized to more challenging activities and settings (distracting settings, complex
environments, etc.)
Therefore, as has been discussed, the simple design of the curriculum is not enough to ensure
success. In addition, an intensive, systematic clinical program must be provided to address all of the
needs for support of language development. In a systematic intervention program, the child is
assured of sufficient hours of one-to-one instruction, effective staff training and management,
complete (consistent 24-hour follow-through) involvement of the parents and typical care-takers,
and regular planning and case-management to ensure an optimum rate of development and
individualization to meet the child’s special challenges.
Generative response classes. The foundation of language programming is the development of
generalized imitation, which has been shown to be a class of behavior (Baer & Sherman, 1964; Baer,
Peterson, & Sherman 1967; Garcia, Baer, & Firestone, 1971; Schroeder & Baer, 1972). The concept
of generative response classes is integral to the design of the current system. A response class has
been defined as a collection of behaviors which, when measured as a group and subjected to
environmental controls, vary together and produce “smooth curves” in graphic representation of
their variability (Skinner, 1936). When a class is formed, it is said to be generic. In other words,
when an environmental manipulation, such as reinforcement, is applied consistently to the members
of the class, their rate co-varies with the rate of reinforcement so as to show a consistent effect of
the reinforcement across the members of the class (Baer, Peterson, & Sherman, 1967). This effect
can also be described as: the behaviors are functionally related as a class.
In the development of language, a subsequent concept, generative responding, may also be used to
define membership in a class. Here, various exemplars of a class may be taught singly in the context
of discrete trials. When later-introduced members of the class acquire discriminative control more
rapidly than did earlier members, intra-class generalization occurs; suggesting a functional
relationship between the members and, therefore, membership in a common class. When a laterintroduced member of the class is shown to be acquired virtually immediately, the class may be said
to be “generative” (Schumaker & Sherman, 1970). Training of some members of the class generates
the acquisition of subsequent members (Baer & Guess, 1971; Clark & Sherman, 1975; Goldstein &
Mousetis, 1989; Guess & Baer, 1973; Halle, Baer, & Spradlin, 1981; Schumaker & Sherman, 1970;
Stokes, Baer, & Jackson, 1974). In many cases, no apparent training whatsoever is required for
subsequent members to occur.
The following figure (1) illustrates generative training of a simple label (horse). Here, each column
represents the number of trials required to develop discriminative control over a specific exemplar (a
given toy horse, picture of a horse, or live horse). A trial is counted each time the child is presented
with an SD of the given horse exemplar, or a distractor exemplar (a toy cow, for example), whether
or not the child correctly responds, “horse.” Mastery is defined, simply, as when, on the first trial of
a new day in which the child is presented with yesterday’s exemplar, the child independently makes
the correct response, “horse.” In this example, the first column shows that the child required 100
such trials before independent responding on the first trial of a new day occurred when the target
was a specific three-inch-long, black, plastic, toy horse. The second column shows that the child
required 98 trials before the independent response to a ten-inch-long, brown, painted, toy horse was
acquired. And so on with different toys or pictures being the focus of each subsequent column.
The last column shows that the child required only one trial acquire the correct response, without a
prompt, to a live horse which was standing in a field alongside the road as the family drove by. The
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 8
child’s expressive label, “horse,” was now generative, as the child no longer needed a prompt in
order to spontaneously and correctly label a novel horse.
Figure 1: Generative development of the class: “Horse.”
100
90
Number of Trials to Mastery
80
70
60
50
40
30
20
10
0
1
2
3
4
5
6
Successive Exemplars of a Horse
7
8
9
The process of developing generative language classes is a foundation of the process of establishing
functional language. Until training in some exemplars of the class can generalize to all related
members, training is incomplete. Once training has generalized to all related members, the child
shows evidence that “the child has acquired a true understanding of the term or concept.” The child
is no longer showing a response that was memorized by rote, but instead shows the same language
comprehension that any typical child shows. Therefore, “generative responding” describes the
child's responses that have not been demonstrated earlier and have not been directly taught (Baer,
Peterson, & Sherman, 1968). By training generative responding, the child can exhibit novel
responses to novel stimuli. Further, individual language terms (i.e., subject, action, preposition,
adjective, object, possessive pronoun, pronoun, singular form, plural form, past tense, present tense,
future tense) can be conceptualized as generative response classes. By training varied and numerous
exemplars of each specific language term, one can observe the emergence of new and untrained
exemplars of that language term in response to novel stimuli. Thus, behavior therapy is developing
generative response classes rather than a large set of rote responses.
A variety of language terms have been demonstrated to be generative response classes in the
experimental analysis of language. These have included the past tense, the present progressive tense
(Schumaker & Sherman, 1970), and the /-s/ and /-z/ allormorphs utilized for object pluralization
(Sailor, 1969). In a specific example, Guess, Sailor, Rutherford, and Baer (1968), chose the
productive use of the plural morpheme as the language term for analysis. Through imitation and
differential reinforcement, a child was taught to correctly label singular and plural object labels in
response to single objects and pairs of objects. After training across several exemplars of the two
forms, the child produced plural object labels that had been taught only in the singular form. This
generalization to novel plural forms, showed that the plural morpheme was a generative response
class.
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 9
In intensive early intervention, a basic programming question often is: “how many exemplars of a
label should be taught?” The answer is: the number of exemplars necessary to reach a mastery
criterion in which the child can immediately generalize to all members of the class. The
development of generative forms of language is incompatible with the development of stereotyped
forms of the same language term. Generative training has also been shown to produce better
response maintenance (Whitehurst, 1971; Baer & Guess, 1973). This is very likely the case because,
once the class has generalized to novel members which don’t require training, the class has become
independent of direct training – it should maintain on its own.
Similar to the development of a class of language responses, a concept may be acquired through
successive discrimination training across multiple exemplars (Stokes & Baer, 1977). In this case,
successive training of different exemplars of the concept results in generalization to novel members.
Again, once this occurs, the concept is said to be generative. The following figure (2) illustrates
generative training of a concept, Animals, in much the same way as the training of the concept,
Horse, proceeded.
Figure 2: Generative development of the concept: “Animals.”
100
90
Number of Trials to Mastery
80
70
60
50
40
30
20
10
0
1
2
3
4
5
6
Successive Exemplars of Animals
7
8
9
In the curriculum for language development, then, the foundation for moving on from the
development of single words to that of multiple-term sentences is the generative label. A child
should first fully acquire the generative labels and concepts that will then be combined into
sentences. Not only may single terms (horse) be found to be generative when generalization to
novel exemplars of the term occurs; but also concepts and even forms of speech may be found to be
generative. A concept (animal) is generative when generalization to novel exemplars of animal
occur. Even a form of speech (adjective) can be shown to be generative, when subsequent
exemplars of the form of speech (big, long, red, hairy) are acquired with minimal or no training
(Schumaker & Sherman, 1970).
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 10
The matrix of natural language responses. In clinical practice, various stimulus modes are often
found to be more rapidly acquired by individual children than are others. Several modes are
commonly employed – three dimensional materials (the child labels toys), two dimensional materials
(the child labels pictures), the first person (the child labels their own behavior), or the second person
(the child labels the behavior of the therapist). The third person is also used, in which the child
labels the behavior of someone other than the therapist, as are written communication modes. In
addition, there are often strong individual differences in the rate of acquisition of receptive
(responding to the language behavior of others) versus expressive (using language as a stimulus for
the behavior of others) modes. Beyond these common modes, written stimuli and responses are
frequently found to be an essential mode of language instruction in autism. In clinical practice, with
the matrix curriculum, the child will acquire generative language most efficiently if the therapy
begins with the most effective mode. Further, the process of reducing errors also often addresses
the clinical needs of the child (preventing the development of perseverative errors, or minimizing
frustration). Once generative language is acquired in one mode, the successful mode may be used to
more rapidly teach responding in the less effective modes, through using the successful mode either
as a prompt or for behavioral momentum. These multiple modes can be arrayed as a twodimensional matrix of stimulus and response modes (See Table 1).
Table 1: Matrix of stimulus and response modes.
Response Mode
Two
Dimensional
Stimulus Mode
Three
First
Dimensional Person
Second
Person
Written
Matching
Imitation
Requesting
Receptive
Labeling
Expressive
Labeling
Therapy typically begins with teaching generative expressive and receptive repertoires at the 1-term
discrimination level for each of the individual exemplars (i.e., specific subjects, actions, prepositions,
etc.). Of course, as a prerequisite, the children must be able to accurately imitate the phonemes and
number of syllables found for each particular response. In addition, the class may be developed
through matching, until generative exemplars are acquired, prior to introducing the receptive or
expressive SDs. Individual children present different strengths in either receptive or expressive
labeling skill, and this may be related to the types of differences between these modes themselves.
For example, the receptive task requires a response to an auditory SD, while the expressive task
requires a response to a visual SD. Further, the visual stimulus component of the receptive
compound discriminations are typically simultaneous discriminations, whereas the visual component
of the expressive compound discriminations are typically successive. However, the auditory
component of the receptive discrimination is necessarily successive. These differences may play to
specific functional strengths of the individual child, and the analysis of these differences offers
strategies for training one mode by generalizing from a previously mastered mode through graded
steps of successively more similar approximations of the target mode.
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 11
Typically, the most readily acquired language classes are concrete terms, as opposed to relative
terms. The ability to label or request objects, actions, and subjects is usually acquired first.
Although here, with individual children, actions may prove to be sufficiently relative (the same boy
may be running or sitting, whereas he is always the same boy) to require additional effort to train.
Also, subjects (specifically, the proper names of people) may be (at first glance) surprisingly difficult
to teach. Although within the context of autism (a child is more focused on the child’s behavior
than that of others), it may not be surprising that the child does not attend to the identity of others
in their environment. Adjectives and prepositions are commonly more difficult to teach, as each are
clearly relative to the concrete objects in the child’s environment. These terms, when proven
generative, fall along a third dimension of the language matrix (See Table 2).
Individual
Response
Classes
Table 2: Terms.
Objects
Horse
Car
Table
Rocket
Doll
Generative Terms
Subjects
Actions
Adjectives
Mommy
Run
Long
Frank
Slide
Round
Barbie
Grab
Scary
Auntie Jane
Crush
Enormous
Donald Duck Slurp
Smooth
Prepositions
On
Under
Next to
Between
By
Conditional discriminations. Once two separately developed terms become generative, the
curriculum may then progress to sentences, or the combination of multiple terms (Karlan, BrennWhite, Lentz, Hodur, Egger, & Frankoff, 1982; Lutzker & Sherman, 1974; Mineo & Goldstein,
1990; Striefel, Wetherby, & Karlan, 1976). The use of a sentence may be best considered as a
conditional discrimination. As a simple illustration of a conditional discrimination, in Figure 3, a
successive conditional discrimination is diagrammed where a response to SD1 is reinforced only in
the presence of SDA and a response to SD2 is reinforced only in the presence of SA. Therefore,
reinforcement of the response to SD1 is conditional upon the presence of SDA. In the case of a
sentence, when there are at least two terms in the sentence, each of which require a discrimination,
then the correct response to the entire sentence would similarly require a conditional response – a
correct response to one of the terms is not reinforced unless it is in the presence of a correct
response being made to the second term also.
Figure 3: Successive conditional discrimination.
SDA

SD1
SD2


R1
R1


Sr+
Sr+
SD1

R1

Sr+
SA

SD2

R1

Sr+
As a specific example of the process of moving from the single-term discrimination to the
conditional discrimination, single-term labels (receptive adjectives) are first taught as simple
discriminations: fat versus skinny; round versus square, cold versus warm, etc. until the receptive
adjective is generative (hard versus soft is mastered in one prompt or less). Now, new members of
the class are acquired almost automatically. Similarly, receptive objects are also taught to a
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 12
generative level: car, truck, ball, jet, pirate, knight, etc., until the receptive object is generative (a
subsequent object, spaceman, is acquired with one prompt or less). Then, a two-term phrase is
taught as a conditional discrimination: fat pirate vs. skinny pirate versus fat knight vs. skinny knight.
Here, the discrimination task includes each possible combination of the two terms. When the
discrimination task therefore involves distracters for both terms (adjective and object), a conditional
discrimination is present. A response to “knight” is only reinforced in the presence of “fat” if “fat
knight” was the SD. Put more technically, a correct response to “knight” is only reinforced in the
presence of a correct response to “fat.” Therefore, if the child is to use a sentence that contains
more than one term, the child is making a conditional discrimination. A correct conditional
discrimination is only made when both terms are simultaneously discriminated to produce the one
correct response (out of four possible in this case).
Figure 4: Two-term conditional discrimination.
Figure 4 shows a second example of a two-term conditional discrimination. This is another
adjective-object discrimination, where the SD is, "white ball." On the floor is a black ball, a white
ball, a black car and a white car. A simple discrimination would be “ball” and a response to either
ball would be correct. The response becomes conditional when the adjective determines which ball
is correct (i.e., white ball). In effect, the adjective “white” is the conditional stimulus (similar to
touching the center key in an experimental conditional discrimination procedure). Again, in this
case, a conditional discrimination is not present unless there is an alternative SΔ, or distracter, for
each term in the conditional discrimination (black vs. white and car vs. ball). If there were only a
black and white car on the floor, then the discrimination of car vs. ball would not be necessary, and
the discrimination of white vs. black would be a single-term discrimination.
If both terms did not have a distracter present, the child may not fully attend to both terms and
possibly either learn to use the nondiscriminated term as a stereotyped phrase within sentences; or
the child may acquire latent inhibition over the nondiscriminated term (Lubow & Moore, 1959;
Lubow, 1989). In latent inhibition, the child is in effect receiving extinction training for the
nondiscriminated term (its correct use is not functional in determining reinforcement). As a result
the child may not respond to the extinguished term in future training when it is used functionally to
control access to reinforcement. Because of the potential for latent inhibition, single-term
discriminations are often developed without the use of extraneous words that are not functional
terms. For example, the use of the SD “show me running,” would be shortened to “run,” to avoid
extinguishing discriminative responses to “show me.” Therefore the first condition to be fulfilled in
acquiring sentence skill is to acquire a valid conditional discrimination of at least two individual
terms.
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 13
Conditional discrimination training has been shown to result in generalized responding (Saunders
and Spradlin, 1990). The incorporation of conditional discrimination training into language
programming has also been shown to be effective in remediating overselective responding to
multiple cues (Riedesel & Larsson, 2002). Overselective responding, in which separate cues are not
equally functional in controlling responding, is often cited as a response characteristic of children
with autism (Lovaas, Koegel, & Schreibman, 1979). However, several studies have demonstrated
that children with autism could learn to respond to multiple cues if the environment is arranged
properly through the use of conditional discrimination training (Schreibman, Koegel, & Craig, 1977;
Koegel & Schreibman, 1977). Furthermore, after presenting a series of conditional discriminations,
children with autism have responded to novel conditional discriminations without demonstrating
overselective responding (Riedesel & Larsson, 2002).
Recombinative generalization. By teaching a conditional discrimination, a new process for the
formation of a generative response class, as a conditional discrimination, becomes possible:
recombinative generalization (Goldstein, 1983). Here a generative two-term response is acquired
when the individual terms spontaneously recombine into previously untrained combinations. For
example, responses to “pushing barrel,” “filling barrel,” and “filling cup,” are prompted, but
“pushing cup” is acquired as a novel recombination – one which had never before been prompted
or reinforced. True mastery of a two-term conditional discrimination would be when a novel
combination of two terms, which had never before been prompted in a conditional discrimination,
“throwing car,” occurs. As a result, the child is again “showing true comprehension” of the
sentence by being able to respond correctly to untrained sentences. Therefore, the second condition
for acquisition of a sentence is a conditional discrimination that has reached the level of
recombinative generalization.
Therefore, there are two uses of the term “matrix” in this curriculum. The first, as has been used in
this paper to this point, is the use of the term to describe the overall interaction of language
programs throughout the language curriculum. The second (as shown in Figure 5) is to refer to the
smaller scale programming of a matrix of tasks to most effectively result in a recombinative
multiple-term conditional discrimination within a single skill development program (Wetherby &
Striefel, 1978). For example, a two-dimensional matrix, might, along one axis, include adjective
labels (i.e., blue, green and red) and along the second axis include object labels (i.e., cup, bowl and
plate). By identifying the cross sections between axes, one could develop and train a variety of
language term combinations (i.e., blue cup, blue bowl, red bowl, etc.). With a goal of recombinative
generalization, the process of training specific combinations of language terms, (i.e., blue cup and
red cup) derived from the matrix, continues until it results in the comprehension and production of
previously untrained combinations (i.e., blue bowl and green cup).
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 14
Figure 5: Two-dimensional programming matrix within a single skill program.
Blue
Green
Red
Cup
Blue Cup
Green Cup
Red Cup
Bowl
Blue Bowl
Green Bowl
Red Bowl
Plate
Objects
Adjectives (Color)
Blue Plate
Green Plate
Red Plate
Mastery of every multiple-term conditional discrimination is only achieved when novel combinations
are reliably produced (recombined) with no specific training.
This process of developing recombinative conditional discriminations is a markedly different
language acquisition process from that of acquiring scripted sentences through repetitive practice.
In the review of common stereotyped language structures presented above, many of the errors
described are errors of recombination. In practice, effective conditional discriminations may
commonly be trained to the eight-term level. These multiple-term conditional discriminations add a
fourth dimension to the matrix (See Table 3).
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 15
Table 3: Common Multiple-Term Conditional Discriminations
2-Term Conditional Discrimination
Subject-Action
Action-Object
Adjective-Object
Subject-Adjective
Preposition-Object
Subject-Preposition
Action-Preposition
Preposition-Adjective
Adjective-Adjective
3-Term Conditional Discrimination
Subject-Action-Object
Action-Adjective-Object
Subject-Action-Adjective
Action-Preposition-Object
Subject-Action-Preposition
Preposition-Adjective-Object
Action-Action-Object
4-Term Conditional Discrimination
Subject-Action-Adjective-Object
Subject-Action-Preposition-Object
Action-Preposition-Adjective-Object
Action-Adjective-Adjective-Object
Subject-Action-Preposition-Adjective
Multiple-Term Conditional Discrimination
Alternate Term Order/Conjunction
Responses are controlled by the complete compound discriminative stimulus (i.e., all of the terms
and their combination within the sentence) rather than a component of the discriminative stimuli
(i.e., one term of the sentence). As the language term combinations increase in number and
complexity, so too does the stimulus array.
In conditional discrimination training, the number of terms which are considered to be part of the
conditional discrimination are only counted by the number of terms for which there are functional
distracters present in the task, rather than by the number of structural terms present in the sentence.
For example, if a sentence such as “Don pulled the red car” is given as part of a conditional
discrimination trial, there are four structural terms presented – a subject: Don; a verb: pulled; an
adjective: red; and an object: car. However, there must be a functional distracter present for each
term – another subject: Joe; another verb: pushed; another adjective: blue; and another object: ball –
to count it as a four-term conditional discrimination. Examples of these conditional discriminations
in different modes are given in Table 4.
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 16
Table 4: Four-term conditional discrimination: subject/preposition/adjective/object
(“The horse is on the tall fence.”)
Distracters in the Field
Sample SD
Sample Response
D
Receptive three- S : “Put the chicken on the hard
R: The child chooses the correct
dimensional box.”
one of two subjects that is
Six toys
in a location relative to one of
four objects possessing one of
two adjective properties.
Receptive First Person SD: “Sam is on the little chair.”
R: The child places himself in the
Four toys and at least
correct location relative to
two persons in area
one of four objects possessing
one of two adjective properties.
D
Receptive Third Person S : “Is Bill under the long table?”
R: The child answers yes or no
Four toys and at least
according to the accuracy of the
two persons in area
question posed by the therapist.
Receptive two- SD: “Baby Bop is on the pink
R: The child points to the correct
dimensional bike.”
picture containing the subject/
Books/Pictures
preposition/adjective/object.
Distracters in the Field
Sample SD
Expressive three- SD: Demonstrate placing one of
dimensional two subjects in a location relative
Six toys to one of four objects possessing
one of two adjective properties.
Expressive First Person SD: Model placing yourself in a
Four toys and at least location relative to one of four
two persons in area objects possessing one of two
adjective properties and then the
child labels after they imitate.
Expressive Third Person SD: Demonstrate placing yourself
Four toys and at least in a location relative to one of four
two persons in area objects possessing one of two
adjective properties and then the
child labels what you did.
Expressive two- SD: Present a picture containing the
dimensional subject/preposition/adjective/
Books/Pictures object.
Sample Response
R: “The chicken is on the hard
box.”
R: “Bobby is on the little chair.”
R: “Kara is under the long table.”
R: “Baby Bop is on the pink
bike.”
Table 4 illustrates the concept that the structural words in a sentence are only counted as terms
when there is a distracter for them.
In Table 4, receptive labeling instructions (“Put,” “Point to,” “Go,”), and the expressive labeling
questions (“Where,” “What,” “Is,”) could have been used, but would not be counted as a distinct
term, because there are no distracters for these words. These natural sounding instructions and
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 17
questions may be used in the interest of natural generalization, but they risk becoming irrelevant
through the process of latent inhibition (the child ignores them in the SD and later does not respond
differentially when they are true discriminators of correct performance), as described above. Indeed,
in clinical practice, the “Wh” question words are often found to be poor discriminations, and it is
likely that latent inhibition is the culprit. Therefore, with a child who requires careful introduction
of terms, these words are best not even used in early programming. The receptive labeling
instructions are taught correctly as a conditional discrimination later as a separate program, and then
they are used correctly in these programs with distracters. The question words are also taught later
as a separate program with correct distracters. Once these skills have been mastered, then the
examples shown in Table 4 are valid natural goals of careful programming. In general, the decision
to use extraneous words in SDs is a significant one, and commonly, the best practice is to limit SDs to
only the least necessary words. A child should clearly demonstrate an ability to respond correctly to
the additional nondiscriminated words before they are used in regular practice.
This brings up the issue of the use of colloquial sentence structures versus the use of telegraphic
speech. The early goal in therapy is to maximally expand the functional vocabulary (as defined by
the total number of generative discriminations) rather than to produce typical sounding sentences
with relatively few discriminations, when there is a conflict between these two practical goals. This
is because the breadth of vocabulary (or number of functional language discriminations) developed
early in age is the best predictor of later intellectual capacity. Therefore with some children, the use
of articles (the, is) may be avoided in order to increase the speed of language acquisition. As a result,
the children sound as if they are speaking telegraphically (on a temporary basis). Then the use of
articles themselves is introduced as a term in subsequent responding.
Similarly, when keeping the number of terms at the most effective level, but still providing typical
sounding sentences in a matrix of distracters, nonfunctional filler words may be used. For example,
in a two-term conditional discrimination (verb-adjective), the discriminative stimulus need not be of
an abstract form (push red). Instead, a perfectly acceptable (and typical sounding) two-term
sentence might be “push the red one,” with no distracter present for “one.”
Of course the preference would be to use natural sounding language at the point at which the rate of
acquisition is similar, with or without the use of nonfunctional terms. This point leads to the next
topic of this discussion, individualization.
Individualization and the sequence of language therapy. To this point, the discussion has
focused on the development of concrete requests and labels of the visual environment, whether
receptive or expressive. In the present framework, a given program to develop a specific multipleterm skill would contain the following significant procedures. First, each component term of the
conditional discrimination has been taught until generative. Then the multiple-term skill is first
taught as a matching or imitation task. This ensures the successful discrimination of the complex
stimulus array, before adding on the additional task of attaching a verbal label to the array. The
match or imitation itself would be developed until recombinative generalization occurs. In order to
ensure that the conditional discrimination is valid, a sufficient matrix of distracters for each term in
the conditional discrimination is included. This will ensure that the conditional discrimination is
truly a conditional discrimination, rather than allowing a single term to inadvertently be a relevant SD
for the conditional discrimination. Table 4 provided examples of the range of modes and necessary
distracters required to master each multiple-term skill.
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 18
The next mastery step is to develop the conditional discrimination until generalization occurs to the
first presentation of a novel combination of exemplars, embedded in a field of novel distracters (to
avoid the process of elimination serving as an inadvertent prompt). This feature ensures that the
presence of either known or unknown items will inadvertently serve as a relevant SD or S for
responding. The next feature is to program discriminations between stimulus continuums that are
relevant and those that should be irrelevant to the conditional discrimination. A large variety of
potential discriminations may be addressed, given the target skill and the tendencies of the specific
child. For example, alternate multiple-term sentences may be discriminated from the target
sentence; alternate instructional arrangements may be employed; or alternate subclasses of the
general case of a given term may be included. These discriminations, which are not of the type of
direct discriminations that have been identified in the matrix, are important to ensure that the child
is responding to the relevant language, rather than to the inadvertent prompt which is offered by the
format of the task being presented. Then the skill, which heretofore had often been taught through
repetition, is generalized to the natural environment by gradually expanding the time and variety of
intervening activities between target trials, as well as varying the SD and the location of trials, until
spontaneous generalization and maintenance in natural activities can be anticipated. Another feature
is to begin programming (earlier or later, depending upon the child’s current optimal learning style)
multi-modal activities in which unrelated stimulus and response modes are mixed into the same
activity. This is followed by generalization of the therapy activities into forms of naturally occurring
activities. In fact, one of the best conventions used for the promotion of thorough generalization
and maintenance, is to purposely design skill-training programs within the context of natural play
language activities (playing with emergency vehicles, doll houses, etc).
For maximum efficiency in therapy, the existence of the matrix is not used to dictate a standard,
lock-step progression through programs. Instead a premium is placed upon the clinical judgment of
the practitioner to advance the progress of programs in the optimum manner. Moving too slowly
through the curriculum has been found to result in an increase in the rate of stereotyped language.
Instead, progress through the curriculum is best based upon dynamic behavior principles. For
example, one skill is moved to the next as soon as generative responding is acquired. When the rate
of learning is not accelerating, the practitioner begins a prescriptive functional assessment to
determine more effective procedures (direct prompting versus behavioral momentum, for example),
or a change in therapeutic focus across modes in order to maintain optimum progress toward
generative responding. In clinical practice, a child who shows a strong individualized tendency to
learn more readily in one mode as opposed to another, will typically move through several levels of
multiple-term conditional discriminations in a single mode, before the other modes are even
addressed. This allows the child to learn the most individual terms and combinations as early in
their development as possible, rather than slowly acquiring a broad vocabulary. Normative cognitive
research suggests that the earlier the child acquires a broad vocabulary, the higher he will score in
later measures of cognitive functioning (Ramey, Campbell, & Finkelstein, 1984; Hart & Risley,
1995). Further, by acquiring fluent skill in language through a single mode, the child may more
readily acquire basic functioning in the more difficult modes.
In the language matrix, different modes are typically associated with different discrimination training
procedures. A receptive two-dimensional task is typically a simultaneous discrimination, but a
receptive first-person task is typically a successive discrimination. An expressive task is typically a
successive discrimination, while a receptive task may be either. This important difference in
procedure may be another source of individual children’s differential difficulty with one mode as
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 19
opposed to another. In addition other differences in procedure may result in significant effects
upon skill acquisition. These procedural differences include auditory versus visual stimulus modes,
speed of presentation, types of distracters available, or vocal versus physical response mode – each
of which may interact with individual error history, motor ability, tendency to impulsively respond,
or attractiveness of stimuli. Regarding the matrix of multiple-term sentences, the conditional nature
of an expressive task is very different from the same number of terms in a receptive task. The
effective use of matching or imitation as a prompt also varies depending upon the mode used.
The choice of vocabulary should be designed to promote natural language development. As follows
from the use of novel language learning as the criterion for mastery, the continuing elaboration of
this criterion suggests that the best ongoing procedure is to automatically include novel vocabulary
in all multiple-term programs so that the child is maintaining the skill of learning new words in a
single trial, as a typical child does. Some children require an ongoing vocabulary program in which
each trial is composed of a different mode (of which currently there are 23 identified modes), to
ensure that the vocabulary learning is sufficiently generative (in the first trial, the term is introduced
as a definition; in the second as a yes-no answer; in the third as a two-dimensional expressive, etc.).
A related concept here is to integrate natural social and play vocabulary in all programs, as opposed
to over-reliance upon academic concepts. The essential aim of behavior therapy for autism is to
teach social language, so two goals can be accomplished simultaneously if the exemplars have to do
with common social and play concepts (Disney villains, Hot Wheels cars, action heroes, etc. rather
than community helpers and days of the week). We are teaching children to have attractive play
skills rather than be academics (the latter often being an easy goal). The focus on social and play
language also determines the best modes of instruction. Two-dimensional tasks may be readily
available, but are the least similar to actual play activities. Therefore labeling of three-dimensional
toys and the actions of others are much more readily generalized skills.
As noted above, most children acquire language skills more readily in one mode than in others, and
this fact offers a significant strategic means of individualizing therapy. The language skills may be
best introduced in the primary, most successful mode, and then generalized to the more difficult
modes. Once sufficient skill is developed with the primary mode of instruction, related modes,
which may be more challenging to teach, can be either: generalized to; programmed through
behavioral momentum from the primary mode; or taught by using the primary mode as a prompt.
For the purpose of generalization, materials that readily generalize from one mode to another may
be very strategic. Then the transfer from one mode to the other can be pursued in a series of graded
steps. For example, when a child’s strongest mode is two-dimensional, Colorforms (twodimensional vinyl pictures) or realistic stickers may be used to create the exact two-dimensional
stimulus that will also be set up as a three-dimensional stimulus with the three-dimensional toys that
are pictured in the Colorforms or stickers. The Colorforms provide for response and stimulus
generalization, in that the child can be manipulating the placement of the two-dimensional picture of
the toy in a similar manner to their manipulation of the toy. This ready generalization can then be
faded into less similar stimuli. In another example, for a child who attends well to television,
labeling a video on a screen, or a frame from a CD-ROM book, can be generalized to the same
arrangement of three-dimensional figures, using the same toys. For a child for whom active gross
motor play is highly preferred, the use of requesting in the context of playground play may be the
most effective initial mode, followed by generalization to the same play actions with, for example, a
Playmobil playground set. Similarly to generalize from the three-dimensional mode to the firstperson mode, the child may initially be moving toys around a table-top arrangement of toy furniture.
Then the child moves the same toys around similar actual furniture in the room. Actions are initially
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 20
targeted with toys for which a non-object movement is also possible (banging with a hammer can
generalize to banging a fist). Then the toys are removed and the child moves their own body around
the room. To generalize from the first-person mode to the two-dimensional, the child may “match”
two-dimensional photographs of their body in certain positions to their mirror image as they act out
the same body positions. To generalize from two-dimensional to third-person responding, the child
may match photographs of other persons who are engaged in a behavior to the actual body of the
other person as they act out the same behavior.
Sample results of expressive and receptive matrix programming
Children. For illustrative purposes, clinical data on three children's performances with language
matrix programs are presented here. Each child had been diagnosed with autism or PDD/NOS and
was participating in an in-home early intervention behavior therapy program. The level of direct
therapy hours was approximately 40 hours per week of home- and school-based intervention. The
average age at intake was 3 years old (range: 2 years 11 months to 3 years 2 months). At the onset of
intervention, the children's verbal abilities varied from five-to-six one-term receptive and expressive
labels to 50 receptive and expressive object labels. Each child demonstrated limited verbal ability.
Once the children acquired generalized matching skills and an assessment of each child's receptive
language skills was conducted, the receptive language matrix programming was implemented into
each child's therapy programming. Expressive language matrix programs were implemented once
the child was able to accurately imitate phonemes and the required number of syllables for each
expressive response.
Clinical data. Preliminary clinical data were collected for each child on each dimension of the
language matrix (i.e. response mode, stimulus mode, and the components of the conditional
discrimination). In order for each child to meet the generative mastery criteria, the child needed to
generalize responding to novel or untrained conditional discriminations in each response and
stimulus mode on the first trial of a new day independently (without prompting).
Child 1. Data for Child 1 are presented as number of days to mastery of each skill, organized by
type of program, rather than consecutively (see Figure 2). Child 1 met the generative mastery criteria
for receptive 1-term discriminations after an average of 19 consecutive days of training (range 2 -78
days). Forty or more consecutive training days were required for actions, prepositions, and
adjectives before child 1 demonstrated generalized responding to novel or untrained stimuli.
Acquisition of generalized responding for the expressive 1-term response mode occurred more
rapidly than the receptive response mode. On average, 9 consecutive training days (range 3 - 21
days) were necessary before the mastery criterion was met.
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 21
For Child 1, receptive 2-term conditional discriminations were acquired more rapidly than receptive
1-term conditional discriminations. Three training days on average (range 2-7 days) were conducted
before the child demonstrated comprehension of novel 2-term conditional discriminations.
However, additional training days were needed for the expressive response mode (average 18 days;
range 2 - 56 days).
Ten consecutive training days on average were implemented for receptive 3-term conditional
discriminations before mastery criteria was met across all stimulus modes (i.e. three-dimensional,
two-dimensional, first person, and third person). The number of training days for the expressive
response mode decreased to an average of 6 consecutive days (range 2-6 days).
For the receptive and expressive 4-term response modes, an average of 4 and 2 consecutive training
days, respectively, were required before the child demonstrated comprehension and production of
generative language. Furthermore, Child 1 generalized responding to both the receptive and
expressive 5-term conditional discriminations without training. This child participated in the
language matrix program for 22 consecutive months.
Child 2. Data for child 2 are presented both by order of stimulus mode (Figure 3) and by date of
program introduction (Figure 4). Child 2 achieved mastery criteria for 1-term discriminations on an
average of 81 and 78 consecutive training days for receptive and expressive response modes,
respectively (range 6 - 239 days). Two hundred thirty-nine training days were required for
possessive pronouns in both response modes and three of the target stimulus modes (i.e. threedimensional, first person, and third person).
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 22
The average number of training days decreased to 10 (range 1-27) consecutive training days for
receptive 2-term conditional discriminations. In addition, the average number of consecutive
training days decreased to 13 (range 3-31) for the expressive response mode. The accelerated
acquisition of more complex conditional discriminations continued for 3-term conditional
discriminations. It took an average of 2 and 6 consecutive days to meet mastery criteria for the
receptive and expressive response modes, respectively (receptive range 1-4; expressive range 2-14).
For both receptive and expressive 4-term conditional discriminations, averages of 4 and 2
consecutive training days were conducted before the child exhibited generalized responding to
untrained combinations (receptive range 1 - 8; expressive range 1 - 5). Also, minimal training was
required for receptive and expressive 5-term conditional discriminations (receptive average and
range 2; expressive average 2.5 days and range 2 - 3 days). The data for Child 2 illustrated
accelerated generalization of responding to more complex conditional discriminations across various
response and stimulus modes. This child participated in the language matrix program for
approximately 12 months.
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 23
Child 3. Data for Child 3 are presented both by order of stimulus mode (Figure 5) and by date of
program introduction (Figure 6). Child 3 required an average of 42 consecutive training days for
both receptive and expressive 1-term discriminations. Receptive 1-term discriminations ranged
from 1 - 136 training days; whereas expressive 1-term discrimination ranged from 2 - 94 training
days.
The training time required to meet mastery criteria for more complex conditional discriminations
consistently decreased as the complexity increased for both response modes and all stimulus modes.
Receptive and expressive 2-term conditional discriminations required an average of 15 and 11
training days (receptive range 1 - 59 days; expressive range 2 - 29 days), respectively. Averages of 9
and 6 training days were conducted to achieve generalized responding for receptive and expressive
3-term conditional discriminations (receptive range 2 - 14; expressive range 2 - 22).
Additionally, Child 3 continued to require minimal training for both 4 and 5-term conditional
discriminations in both response modes and all stimulus modes. The average and range of
consecutive training days was 8 for the receptive mode and 16 for the expressive mode. Child 3
participated in the language matrix program for 11 months.
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 24
Summary of clinical data. Overall, the average length of time for the three children to complete
the basic language matrix program was 14 months with a range from 10.5 months to 22 months. All
three children were able to comprehend novel complex instructions and produce generative
language that contained at least 5-terms. Furthermore, the time required to meet the generative
mastery criteria in each response and stimulus mode generally decreased for each child as the
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 25
conditional discriminations became more complex. The alternation of stimulus modes in treatments
offers an alternating treatments design for analysis.
While these are typical results, it should be noted that, for whom the clinical focus is noncompliance, rather
than language skill per se, the pattern of results may be quite different. In such a case, the child might begin
engaging in stereotyped noncompliance only when certain modes at a certain level of complexity were
reached. The difficulty with those modes may, for example, begin to surface in the three-term
discriminations. The length of time to mastery in such cases may be due to the time required to problemsolve an effective clinical solution, by engaging in a sequential functional analysis of alternative interventions
(Riedesel & Larsson, 1999). Other reasons for a different progression of acquisition rates would include a
case in which there is a functional barrier to the child’s attending skills, that is only addressed at a certain level
of complexity. Other clinical focuses are also found.
Advanced programming issues. In many children, the use of reading and writing is an effective
format for developing language responding (Lovaas & Lovaas, 1999); and the sequence of written
statements can be developed according to the principles discussed here. It is often noted that a
reading response results in more rapid vocal language acquisition than does a picture-based system.
This may be because both vocal and written language responses are arbitrary representations of
visual discriminations, while a picture-based language response involves much more direct
representations of the visual discriminations. This distinction concerning arbitrary language modes
suggests an important refinement to the concept of symbolic language. Nonarbitrary picture and
gestural systems are much more primitive than are arbitrary written and vocal systems.
Much of a child’s everyday language involves surprisingly large conditional discriminations. In the
second phase of matrix training, five-term conditional discriminations are established. But many
language interactions involve up to 12-term conditional discriminations or larger. Not every
conditional discrimination is a highly structured sentence that follows grammatical rules. Much
everyday language is a string of phrases, which must nevertheless be comprehended or initiated
(“It’s a bear, it’s big and brown, and it’s chasing your Daddy away from the fish he caught!”)
Therefore the conditional discriminations will be generalized into a variety of grammatical (and
“nongrammatical”) statements comprising many terms. Another common multiple-term variation
involves conjunctions (and, or, before, instead of, rather than, then, if-then). Many of the terms are
used in combinations of two or three in the same conditional discrimination (adjectives – big and
hairy; prepositional phrases – in the bucket under the bed; subjects – Bert and Ernie; or verbs –
dancing and singing). Also, alternate term orders are also significant generalization steps (“the frog
is under the table” vs. “under the table is where you’ll find the frog”).
After establishment of basic language skill in the requesting and labeling areas, more complex
variations or forms of concrete language may be developed as expressive and receptive labels (Risley
& Baer, 1973). Such further objectives to be addressed include syntax (Goldstein, Angelo, &
Mousetis, 1987), word forms (Baer & Guess, 1973), and tenses (Kuczaj, 1977). Many skills that
often pose challenges to therapy may more readily be developed when taught through this
framework. For example, pronouns, possessives, tenses, and syntax (or grammatical rules) may be
more readily developed when introduced upon a base of generative language responding. In fact,
these advanced language skills may be taught through the same conditional discrimination process as
were the basic skills. Conversely, the simple generative programming of sufficient sentence forms
may result in natural generalization to these specialized forms, without the need for direct
programming. Alternatively, these forms may never need direct programming, when, for example,
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 26
the original function of the delayed language development was a clinical behavior challenge such as
noncompliance or social avoidance. In these cases, thorough matrix programming is not necessary
because the clinical focus processes of generalized compliance training and generalized imitation
training meet the functional needs for normal development.
The focus on controlling language development, according to the planned development of terms
and modes, may substantially shape the curriculum. For example, in an alternative early intervention
system, many early labeling and requesting skills may be taught as full sentences (“I want _;” “I have
a _;” “Show me _.”) However, each of these language responses incorporate three terms. Further,
they incorporate pronouns. Finally, each of these sentences use an abstract term (want, have, show).
In the language matrix formulation, programming would begin with single term labels and only
move to such phrases once concrete three-term sentences are recombinative conditional
discriminations. Then the pronouns would be taught as single term discriminations, being faded
into proper noun discriminations, and keeping the receptive and expressive forms distinct until
generative mastery of each mode is mastered. As the earlier examples suggested, these multiple-term
sentences may then be taught as analogue concrete conditional discriminations (using terms that are
visual). Once the concrete conditional discriminations are established, generalization to the abstract
terms are more readily acquired as recombinative terms.
The focus on generative labeling also suggests other significant deviations from traditional
curriculums. For example, in alternative systems, sequencing two-dimensional stimuli is often
introduced as an early skill. However, the commonly employed sequencing cards usually vary by
three or more terms as the sequence progresses. In order to avoid merely teaching rote associations,
which may easily be irrelevant to the sequencing concept, it is preferable to first teach the child to
label each stimulus card (after acquiring the multiple-term conditional discrimination) and then
sequence the cards. Therefore, in the matrix system, this skill will typically be introduced after a year
of therapy, rather than early in programming.
Natural language programming. The simple development of language skills in a one-to-one
therapeutic setting is not the intended outcome of this approach. As indicated above, the true goals
of therapy reflect the natural use of natural language in natural activities. Therefore the use of these
skills must be directly related to the use of these skills in the natural environment. Highly structured
programming, as described above, may be necessary to develop generative language responding.
Then natural programming techniques may be employed to generalize the language skills throughout
the day. A common procedure is to gradually fade the structure of the language programs into that
of a natural play situation as each skill is developed. A significant fading step is to increase the
number of modes used within the same activity until the natural variety of modes is arrived at.
Maintenance programming is then assessed in the natural play activities rather than in the structured
activities. Again, the strongest modes of language development may be used to develop the desired
natural uses in the environment. Incidental trials may be inserted in natural activities in order to
create behavioral momentum for natural language usage and then planned reinforcement may be
used to establish the necessary rate of the desired skill. After establishing the desired skill in the
natural environment, the contingencies may be faded to the natural rates found in the environment.
Conversely, while therapy is progressing, natural language activities that demand a higher order
(number of terms, unmastered mode) of language skill than that which has been mastered in direct
therapy, should be avoided. Premature exposure to unsuccessful tasks may inhibit natural language
development, by counter-establishing stereotyped language behavior in those activities.
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 27
It is usually essential to target natural statements in the home, school, and community; as well as to
generalize statements to other persons in the environment, such as relatives and peers. Again, this
programming usually proceeds with a minimum of stereotyped errors if the language skills used in
these natural environments follow the sequence of generative programming, rather than not being
coordinated with that sequence.
Of course, the child’s therapy will progress most effectively if initial learning is programmed through
the most highly successful structure for that individual child. After establishing the skill, the skill is
generalized to all naturally occurring activities. One of the most significant areas for generalizing the
skill is to operant responding. Free operant language responding takes many functions, but
regardless of the function, the complexity of the language should be programmed in concert with
matrix progress. The spontaneous operant may be directly programmed by establishing labels that
are not dependent upon a specific vocal SD. Rather, the conditional discriminations are presented as
the materials themselves, and the prompts are faded until the child naturally labels a visual array.
Then similar arrays can be arranged in natural settings, such as the living room, bathroom, or
playground. The child is then given repetitive errand trials to go to the setting and label a series of
arrays. After momentum is established, the artificial array is removed, and the child is given a new
errand trial to go to the same setting. The likely result is that the child will now label a natural array
in the setting. This “spontaneous” label can then be operantly reinforced.
Auditory comprehension programming. A focus on concrete requesting and labeling skills
enables the use of straightforward direct teaching procedures to establish a relatively high degree of
language skill, before moving into less concrete language skills, which are typified by auditory
comprehension skills. Initially, the comprehension skills can be established as direct concrete
analogues of the labeling skills. These comprehension skills are first developed as comprehension
questions about visual stimuli, and then comprehension questions about analogous auditory stimuli.
Then these comprehension skills can be developed into abstract and inferential comprehension
skills. A significant aspect of the comprehension skill for the treatment of autism is social
comprehension. Upon this base, the development of attention span, communication of complex
information, and conversational skill will also be developed. Simultaneously all of these skills are
generalized to all of the natural activities in the child’s day, ensuring that their skill is fully established
as a bona fide resident of the ecology of the child’s behavior.
The progression of advanced communication skills follows this general course:
o Requesting
o Expressive labeling
o Receptive labeling
o Concrete reciprocals
o Generative concept formation
o Concrete visual comprehension
o Concrete auditory comprehension
o Abstract auditory comprehension
o Inferential auditory comprehension
o Social comprehension
o Naturally occurring language interactions
o Complex language production
o Abstract conversations
o Comprehension of conversations
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 28
In the auditory comprehension phase, the prerequisites (all generative) are seven-term conditional
discriminations and relevant concept categorization. Concept development is a particularly
important realm of language responding, as it forms a straightforward base for abstract
comprehension skill. Indeed most, if not all, knowledge can be conceptualized as a hierarchy of
categories.
After development of the multiple-term labeling and requesting skills, comes concrete development
of language concepts through matching, sorting, categorizing, and labeling related to these skills.
For example, a relatively high level of discriminative ability can be developed through nested and
related concept matching (i.e., horse to horse; farm animal to farm animal; mammal to mammal;
transportation to transportation; pet to pet; “cowboy stuff” to “cowboy stuff;” big animal to big
animal (two-term matching)). Not only are these concepts generatively matched, but relevant labels
are generative (given a class, name a member; given a member, name a class; given an attribute,
name a class; given a class, name an attribute). When all of these skills become generative, the stage
is set for the auditory skills. Both generalized vocabulary skills and higher-order concepts (typically
taught through categorization) can be developed according to these principles. In particular, with
concept development, the language tasks (labeling members of a category, characteristics of a
category, etc) can conform to the child’s current skill level in the matrix of conditional
discriminations.
The preliminary phase of comprehension programming is visual, in which the child is presented with
a visual display, and then given an SD in the form of a question. A particular component of the
visual comprehension task that should always be checked, but may not need programming, is the
“Wh” term discrimination. An individual child may need single term programming to discriminate
Who-What-Where as a term, especially if these words have been used “indiscriminately” in earlier
programming – resulting in potential latent inhibition. Normally, the “Wh” discrimination is begun
as a two-term discrimination in the appropriate matrix of distracters (“who is jumping”). Before
each of the subsequent auditory comprehension tasks are delivered, the analogue visual task should
be mastered.
Then the first phase of auditory comprehension is detail comprehension for each of the single
terms. Programming in this phase typically involves a discriminative stimulus which is composed of
two two-term statements and one two-term question (Ed is running and Fred is sitting. Who is
sitting?). This is developed in a series of single trials, with each trial commencing with new
statements, because the content of the terms themselves is no longer being programmed. Each trial
is a partial conditional discrimination for the sake of attention span, but through the course of
repeated trials, every possible combination is pursued. Typical questions involve who, what, where,
which, and yes-no. The skill can be directly taught, based upon the foundation of labeling, by using
the direct visual labeling analogue of the comprehension skill. The visual analogue can be used as
either a prompt or for behavioral momentum. After generative responding at this level, the program
progresses to three-term statements and four-term questions (counting the “wh” word as a term).
Finally, attention span is built by gradually increasing the number of sentences in a story and then
asking a single question.
A sample two-term “wh” discrimination is as follows:
Arrange a layout with two sets of three unambiguous items and do actions with the two subjects
(Ernie is in a racecar and Bert is next to the trash):
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
SD: “Who is in?”
SD: “Where is Bert?”
SD: “What did Ernie do?”
SD: “Who sneezed?”
Page 29
R: “Ernie”
R: “Next to”
R: “Drive”
R: “Bert”
A sample detail comprehension task embedded in a story would be as follows:
“Piglet and Pooh went to Eeyore’s birthday party. Pooh brought honey and Piglet brought a
balloon. The balloon broke and Eeyore said it was alright. Who brought the balloon?”
The second phase is concrete reasoning comprehension. The common question words are when,
how, why, if-what, and yes-no. This reasoning comprehension skill is termed “concrete” because of
the existence of natural cue words (because, since, so, before, after, etc.) in the discriminative
stimulus (in this phase, these cue words are not prompts). Here the direct response is to give the
answer based upon giving the content that is related to the cue word. This skill progresses through
the same term structure as detail comprehension, until a full story is given and the correct response
is to give the answer based upon the cue word. The use of two cue words forces an active
discrimination that usually ensures accurate attending and responding.
A sample concrete reasoning task would be as follows:
“Mary and Margaret went to the store for dessert. When they got there, the man told them that
there were no more cookies. So they went home. When did the man tell them there were no more
cookies?”
Note that a “When” question is a more relative comprehension question than is “who,” “what,” or
“where.” In the concrete reasoning, there are many common cue words, such as “before,” “after,”
“while,” “first,” “second,” and “then.” The child’s discrimination of these time concepts can be
programmed generatively. These time-concept, cue words can also be used in the receptive and
expressive modes as a term: “push the ball then throw the ball.” Therefore, the child can master
these relative time concepts in the labeling mode before being challenged in the comprehension
mode, and the labeling can be used as a prompt for the comprehension. In addition, it can be noted
that the time concept is then an additional term in the sentence: “hug and tickle the doll” requires
only a three-term conditional discrimination, whereas “hug then tickle the doll” requires a four term.
Once the concrete reasoning is mastered, abstract reasoning analogues can be built using the same
forms of statements and questions, but with no cue words. The same progression is followed as
before. A sample abstract reasoning task would be as follows:
“Jenny couldn’t get into the house when she got home. She looked and looked but couldn’t find a
key. She went to Mary’s house and called her mother. Her mother came home and unlocked the
door. How did Mary get into her house?”
Then inferential comprehension is developed. The distinction between abstract and inferential
comprehension is that no correct answer is given in the SD. The correct answer is a creative, but
relevant, answer to the question. Inferential comprehension questions include: “why do you think,”
“how do you think,” “what do you think will happen next,” “if-what,” and “yes-no.” The same
progression is followed as before.
To distinguish again between a matrix-based curriculum and a traditional early intervention
curriculum, common factual comprehension skills may not be programmed until mastery of these
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 30
reasoning skills is achieved in order to avoid the development of stereotyped “memorization” of the
answers to the factual questions without true comprehension. Instead, as the examples have shown,
each of these tasks are plainly generative, in that the child is learning to answer questions about
novel content, and mastery is assessed when the generalization to novel material occurs.
Social comprehension. At the heart of the focus on language comprehension, for most children
with autism, is social comprehension. Such tasks can be the most abstract, as the concepts are
typically based upon an analysis of a series of interactions between two or more persons. Some of
the initial social comprehension tasks can be concrete (“who pushed the girl?”), but most become
very abstract (“why didn’t they want to play with him?”). As such, the social comprehension tasks
usually follow development of inferential comprehension. Then, when the social comprehension
tasks are initiated, they are based upon a story that allows for such social comprehension questions.
A sample task would be as follows:
Read a story about Arthur’s dog ruining the decorations for his surprise party for Francine. His
friends come over early and help him make new decorations just in time for the party. At
appropriate moments in the book, stop and ask the following questions.
o What does Arthur want? Why?
o How does he feel? Why?
o How do his friends feel? Why?
o What can they do for Arthur? Why?
o What do you think will happen next? Why?
o What would you do if you were Arthur’s friend? Why?
In social comprehension, to establish thorough social competence, each common social concept is
programmed across multiple exemplars. For example, various commercial stories, which contain the
same concept, can be presented in randomized fashion until correct generative responding is made
spontaneously to a novel story regarding the same concept. The modes of the stories can be
customized to allow for the child’s strengths (video, picture book, written passage, computerized
book) and ultimately generalized to answering questions regarding a role-played situation or
conversation, as well as acting out the roles in a play. This is often presented as dramatic play based
upon a familiar book or video, but not following a memorized script) and mastery is assessed both
upon making the appropriate responses in the dramatic play, and in answering comprehension
questions about the dramatic play. Sample social concepts, across which multiple exemplars are
programmed, include:
o Other children want to choose what to do
o Children don’t want to do the same thing all the time
o Children want people to like them
o Children want friends
o Boys and girls like different things
o Children don’t like to have their things ruined
o Friends help each other
o Children want to win
o Sometimes children don’t want to be together
o Children get mad for a reason
o Children can laugh when they are teased
o Children like to be good at something
o Families love each other
o Children want attention from others
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
o
o
o
o
o
Page 31
Children help someone who is hurt
Children should be good winners
Children should keep a secret
Children try to be like the others
Children don’t like bullies
Complex language production. A particularly important aspect of language development is
creative speech. While much of the programming discussed throughout this chapter is creative in its
outcome (novel recombinative generalization), a significant creative language skill is expression of
original ideas. In the matrix program, this progression may be taught through a story-telling
framework. Once the required multiple-term complexity is developed, the child’s basic story telling
skill is developed through expressive labeling of sequencing tasks. After the concrete sequencing
tasks are mastered (expressive labeling and sequencing of a series of related two-dimensional
stimuli), more ambiguous two-dimensional stimuli are given. These ambiguous stimuli still are
related and tell a sequential story, but there is sufficient complexity and variety faded into successive
sequences that the child’s expressive labeling of each picture is necessarily more creative. Once this
responding is established, the child is taught to tell a story using puppets or toys, through a simple
chaining process. However, a rote script is not employed. Instead the child is labeling a variety of
available stimuli in a related fashion. Variations in the array of materials provided can force creative
story telling. For example, uncommon combinations of materials can be presented so that the child
is highly likely to produce novel expressive labels of the materials available. Instructing creativity
and differential reinforcement of creative terms and sentences are also significant techniques. Once
the story telling is established with visual materials, the next step is to establish nonvisual story
telling. This can be done using similar materials for either momentum or prompts. A related
process is to use pictures of actual events, to develop the skill of telling about daily events.
Conversational skills. An important aspect of natural language development is conversational
skill. Conversational responding is a different function from the labeling, requesting, and questionanswering skills which have here-to-fore been discussed. In conversational responding, the child
reciprocates statements to those made by another. The skill involves creative speech, but also a
necessary degree of responsiveness to the trend of the conversation. In the present framework, the
appropriate reciprocals are developed only after mastery of the conditional discriminations is
accomplished, so that the conversational skills are generative rather than rote. To begin
programming conversations, the reciprocal interactions can be programmed as visual reciprocals
(manipulating a toy while labeling the action). To establish common conversational responding, the
reciprocal response is to respond to at least one term of the conditional stimulus, and add to that, in
the common way that a natural conversation often involves a series of tangential statements. They
are each related to the prior, but move in a connected series of tangents, as opposed to directly
mimicking the form of the stimulus. For example, the SD may be, “the car crashed the truck,” and
the reciprocal response is, “the truck’s on fire.” This tangential relationship is more conversational
than mimicking the syntax as in, “the car crashed the truck,” reciprocated by, “the man pushed the
button.” In this manner, the various modes of the labeling skills may be used as concrete prompts
for the reciprocal statements. Before transitioning to purely auditory conversations, the creative
speech programming is also very helpful as a prerequisite. Then, creative reciprocals may be
developed by repeatedly prompting “random” responses – never giving the same statement twice in
the prompting, until the child learns to spontaneously give a new response. In addition, effective
differential reinforcement is critical; and the use of sophisticated clinical judgment is called for, in
determining whether the child’s response was related enough to the original SD to qualify as a
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 32
relevant reciprocal response, versus an autistically unrelated response. Observational learning
procedures have proven extremely valuable in rapidly developing conversational skill. Additional
programming for responsiveness to the partner involves complex contingencies such as cooperation
and competition; and responsive peer tutoring. Auditory comprehension skills are critical for the
development of conversational skill, and comprehension of conversations will be specifically
targeted.
As described above, this language acquisition process is designed to culminate in the generalization
of natural language forms functionally into the child’s typical daily activities. However, the
development of complex social interactions which involve language may need to be programmed
directly. For example, differential reinforcement of target social skills in the natural environments is
commonly required (Buell, Stoddard, Harris, & Baer, 1968; Hart, Reynolds, Baer, Brawley, & Harris,
1968). Children have been effectively learned to recruit social praise for appropriate behavior in the
classroom (Connell, Carta, & Baer, 1993; Stokes, Fowler, & Baer, 1978). Target social skills may be
prompted using a two-dimensional stimulus in the natural setting (Curl, Rowbury, & Baer, 1985).
Creative play may be differentially reinforced (Goetz & Baer, 1973). And prompt fading and
differential reinforcement may be used to establish spontaneous initiations of sharing (RogersWarren, & Baer, 1976; Pinkston, Reese, LeBlanc, & Baer, 1973). In addition, peer programming and
observational learning techniques have been used to develop appropriate social behavior in the
natural settings (Odom & Strain, 1986; Tryon & Keane, 1986).
The consideration of the clinical focus will also dramatically affect the level of social communication
that generalizes from therapy to natural interactions. For example, if the child’s extreme acting-out
behavior is currently being reinforced by the reactive accommodations of caregivers, then this
pattern may prevent the generalization of more natural social language into those activities. The
remediation of the challenging behavior may be necessary before natural social language occurs in
those natural settings. For example, it has been found necessary to establish compliance as a
competing response for attention-getting behavior, and then use differential reinforcement to alter
the rates of the two forms of interaction (Baer, Rowbury, & Baer, 1973; Pinkston, Reese, LeBlanc, &
Baer, 1973). In another example, children’s lack of generalization from one-to-one to group
language activities was accomplished by programming the presence of individual peers into the oneto-one activities (Larsson & Larsson, 1983).
Nonverbal communication. An important feature of social communication is the skill of
comprehending nonverbal communication or body language. Throughout this paper, we have
referred to modes of language behavior that are produced by the mouth as vocal behavior, rather
than verbal behavior; and have referred to verbal behavior as any mode of language behavior that
involves a representational (or symbolic) form of language (such as written words, pictures, signs,
gestures, or vocal words). However, the use of the term, “nonverbal” here is to refer explicitly to
language responses that are conditional upon the body language or inflection used in language. For
example, a receptive statement might be: “pick it up.” If only one item is on the table, the statement
is unambiguous, and merely a two-term conditional discrimination. However, if two items (a book
and a candle) are on the table, the correct conditional response depends upon the glance of the
speaker. If the speaker is looking at the book, then the correct three-term response is to pick up the
book instead of the candle. In this simple case, the additional term is the body language (direction
of gaze) of the speaker. By constructing such tasks, the child can learn to attend to the body
language of others as communication.
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 33
Nonverbal language is also commonly composed of vocal inflections. For example, the statement:
“My car is under the table,” has a different meaning depending upon which of the six words is
emphasized. To comprehend the statement when “my” is emphasized would be to respond that the
speaker is discriminating his car from someone else’s. The correct receptive labeling response to
that statement would be to pick up the speaker’s car instead of another person’s. To comprehend
the same statement when “car” is emphasized would be to respond that the speaker is discriminating
his car from some other possession. The correct receptive labeling response to this statement would
be to pick up the speaker’s car instead of another of the speaker’s possessions. A more complex
receptive labeling response would be to have the speaker’s car and truck under the table and the car
and truck of another person under the table. However, as should be obvious, in a situation such as
this, the words alone will discriminate the correct response, whether or not an inflection is used. So
programming at the level of labeling can only be usefully done as a vocal comprehension task, in
which the child uses the correct inflection in answering a question regarding the possessive, object,
etc. In this case, interestingly, the two-term task described above would confound the correct
discrimination, and instead the SD would be comprised of: the child’s car is under the table, and the
parent’s car is on the table, and the question is “whose car is under the table?” This could be done
as a visual or as an auditory comprehension question. Of course the receptive tasks could be used to
model the correct expressive inflections.
Prevention of stereotyped language. If the language-based programs for typical social activities
are controlled by the progress of the language curriculum, common stereotyped language behavior
may be avoided. As introduced above, the following stereotyped language behaviors such as word
omissions, filling nonsense sounds or words into a phrase, incorrect or over-generalized articles,
word salads, substitutions, conjunction errors, noncontextual statements, and word association
errors are avoided through direct teaching of generative and recombinative language responding.
Other specific language errors such as pronoun errors, plural agreement errors, tense errors, and
word order errors are addressed through direct programming of the appropriate target skills within a
generative framework.
Many of the functions of delayed language development may also be addressed through this
generative process. For example, functions of stereotyped language are addressed through this
process as follows. Lack of generalization from one stimulus to a physically similar stimulus, as if
the response had been learned by rote, is addressed through generative response class development.
The substitution of predictable errors in a perseverative pattern in an unmastered, or difficult task, is
addressed through development of generative responding with the difficult task. Over-generalized
sentence structure is addressed through recombinative generalization of the sentence forms. Using
an incorrect (but often learned) sentence form in a predictable context, is addressed by generative
request development. A word association controlling subsequent local language behavior is
addressed through recombinative generalization of the word. Predictable “error patterns” are
prevented through generative programming in the language tasks in which the patterns would occur.
Lack of responding to the entire context or message, is taught through gradual shaping of the length
of the language task to which the child must respond. A response being controlled by a single word
in a sentence, rather than the entire statement, is prevented through recombinative generalization.
Overgeneralization of a response when learning a skill is prevented by focusing on the successful
development of prerequisite skills to the generative level.
The intent of this system for language acquisition is to develop the complete range of typical
language behavior while preventing or minimizing the impact of common stereotyped language
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 34
responses of autism. With the programming principles described above, this intent can be
accomplished through multiple phases of direct teaching and simultaneous generalization
throughout the child’s natural daily life. The organization of the language curriculum can be used to
control the pacing of related social skill development in a systematic manner. This will prevent
stereotyped responding and develop functional skills for communicating complex social
information. When the language interventions extend through all of the social language functions of
the typical child, including conversational behavior, the child’s language and social behavior may
attain the typical range of functioning of the first-grade child.
References
Baer, A.M., Rowbury, T., & Baer, D.M. (1973). The development of instructional control over
classroom activities of deviant preschool children. Journal of Applied Behavior Analysis, 6, 289-298.
Baer, D.M. (1991). Tacting "to a fault." Journal of Applied Behavior Analysis, 24, 429-431.
Baer, D.M. (1993). Quasi-random assignment can be as convincing as random assignment.
American Journal on Mental Retardation, 97, 377-379.
Baer, D.M., & Guess, D. (1971). Receptive training of adjective inflections in mental retardates.
Journal of Applied Behavior Analysis, 4, 129-139.
Baer, D.M. & Guess, D. (1973). Teaching productive noun suffixes to severely retarded children.
American Journal of Mental Deficiency, 77, 497-505.
Baer, D.M., & Sherman, J.A. (1964). Reinforcement control of generalized imitation in young
children. Journal of Experimental Child Psychology, 1, 37-49.
Baer, D.M., Peterson, R.F., & Sherman, J.A. (1967). The development of imitation by reinforcing
behavioral similarity to a model. Journal of the Experimental Analysis of Behavior, 10, 405-416.
Buell, J., Stoddard, P., Harris, F.R., & Baer, D.M. (1968). Collateral social development
accompanying reinforcement of outdoor play in a preschool child. Journal of Applied Behavior
Analysis, 2, 167-173.
Clark, H.B., & Sherman, J.A. (1975). Teaching generative use of sentence answers to three forms of
questions. Journal of Applied Behavior Analysis, 8, 321-330.
Committee on Children With Disabilities (2001). Technical Report: The Pediatrician's Role in the
Diagnosis and Management of Autistic Spectrum Disorder in Children. Pediatrics, 107, e85.
Connell, M.C., Carta, J.J., & Baer, D.M. (1993). Programming generalization of in-class transition
skills: Teaching preschoolers with developmental delays to self-assess and recruit contingent
teacher praise. Journal of Applied Behavior Analysis, 26, 345-352.
Curl, R.M., Rowbury, T.G., & Baer, D.M. (1985). The facilitation of children’s social interaction by a
picture-cue training program. Child and Family Behavior Therapy, 7, 11-39.
Foss, D. (1968). Learning and discovery in the acquisition of structured material. Effects of number
of items and their sequence. Journal of Experimental Psychology, 77, 341-344
Garcia, E.E., Baer, D.M., & Firestone, I. (1971). The development of generalized imitation within
topographically determined boundaries. Journal of Applied Behavior Analysis, 4, 101-113.
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 35
Garcia, E.E., Guess, D., & Byrnes, J. (1973). Development of syntax in a retarded girl using
procedures of imitation, reinforcement, and modeling. Journal of Applied Behavior Analysis, 6, 299311.
Goetz, E.M., & Baer, D.M. (1973). Social control of form diversity and the emergence of new forms
in children’s block building. Journal of Applied Behavior Analysis, 6, 209-217.
Goldstein, H. (1983). Recombinative generalization: Relationships between environmental
conditions and the linguistic repertoires of language learners. Analysis and Intervention in
Developmental Disabilities, 3, 279-293.
Goldstein, H., Angelo, D., & Mousetis, L. (1987). Acquisition and extension of syntactic repertoires
by severely mentally retarded youth. Research in Developmental Disabilities, 8, 549-574.
Goldstein, H., & Mousetis, L. (1989). Generalized language learning by children with severe mental
retardation: Effects of peers’ expressive modeling. Journal of Applied Behavior Analysis, 22, 245-259.
Guess, D., & Baer, D.M., (1973). An analysis of individual differences in generalization between
receptive and productive language in retarded children. Journal of Applied Behavior Analysis, 6, 311331.
Guess, D., Sailor, W., & Baer, D.M. (1976). Functional speech and language training for the severely
handicapped. Lawrence, KS: H & H Enterprises.
Guess, D., Sailor, W., Rutherford, G., & Baer, D.M. (1968). An experimental analysis of linguistic
development: The productive use of the plural morpheme. Journal of Applied Behavior Analysis. 1,
292-307.
Halle, J.W., Baer, D.M., & Spradlin, J.E. (1981). Teacher’s generalized use of delay as a stimulus
control procedure to increase language use in handicapped children. Journal of Applied Behavior
Analysis, 14, 389-409.
Hart, B.M., Reynolds, N.J., Baer, D.M., Brawley, F.R., & Harris, F.R. (1968). Effect of contingent
and non-contingent social reinforcement on the cooperative play of a preschool child. Journal of
Applied Behavior Analysis, 1, 73-76.
Hart, B., & Risley, T. (1995). Meaningful differences in the everyday experiences of young American children.
Baltimore: Paul H. Brookes.
Karlan, G. R., Brenn-White, B., Lentz, A., Hodur, P., Egger, D., & Frankoff, D. (1982). Establishing
generalized productive verb-noun phrase usage in a manual language system with moderately
handicapped children. Journal of Speech and Hearing Disorders, 47, 31-42.
Kuczaj, S. A. (1977). The acquisition of regular and irregular past tense forms. Journal of Verbal
Learning and Verbal Behavior, 16, 589-600.
Larsson, D.G., & Larsson, E.V. (1983). Manipulating peer presence to program the generalization
of verbal compliance from one-to-one to group instruction. Education and Treatment of Children, 6,
109-122.
Lovaas, N., & Lovaas, O.I. (1999). The Reading & Writing Program: An Alternative Form of
Communication. Los Angeles, CA: Lovaas Institute for Early Intervention.
Lubow, R.E. (1989). Latent Inhibition and Conditioned Attention Theory. London: Cambridge University
Press.
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 36
Lubow, R.E., & Moore, A.U. (1959). Latent inhibition: The effect of nonreinforced preexposure to
the conditioned stimulus. Journal of Comparative and Physiological Psychology. 52, 415-419.
Lutzker, J. R., & Sherman, J. A. (1974). Producing generative sentence usage by imitation and
reinforcement procedures. Journal of Applied Behavior Analysis, 7, 447-460.
Mineo, B. A., & Goldstein, H. (1990). Generalized learning of receptive and expressive action-object
responses by language-delayed preschoolers. Journal Speech and Hearing Disorders, 55, 665-678.
Odom, S.L., & Strain, P.S. (1986). A comparison of peer-initiation and teacher-antecedent
interventions for promoting reciprocal social interaction of autistic preschoolers. Journal of
Applied Behavior Analysis. 19, 59-72.
Peterson, G.B., Larsson, E.V., & Riedesel, K.L. (2003). A conceptual toolkit for intensive early
behavioral intervention teachers. Journal of Behavioral Education.
Pinkston, E.M., Reese, N.M., LeBlanc, J.M., & Baer, D.M. (1973). Independent control of a
preschool child’s aggression and peer interaction by contingent teacher attention. Journal of
Applied Behavior Analysis. 6, 115-124.
Ramey, C. T., Campbell, F. A., & Finkelstein, N. W. (1984). Course and structure of intellectual
development in children at high-risk for developmental retardation. In P. Brookes, R. Sperber, &
C. McCauley (Eds.), Learning and cognition in mental retardation. Hillsdale, NJ: Erlbaum Associates.
Riedesel, K, & Larsson, E.V. (May, 2002). Overselective responding or incomplete task analysis?
Review and recommendations for generative language training. In E.V. Larsson (Chair). The
development of generative classes of language behavior in intensive early intervention. Symposium presented at
the 28th Annual Convention of the Association for Behavior Analysis. Toronto CA.
Risley, T.R., & Baer, D.M. (1973). Operant behavior modification: The deliberate development of
behavior. In B. Caldwell, & H. Ricciuti (Eds.). Review of child development research, Vol. III: Child
development and social policy. Chicago: University of Chicago Press.
Rogers-Warren, A., & Baer, D.M. (1976). Correspondence between saying and doing: teaching
children to share and praise. Journal of Applied Behavior Analysis. 9, 335-354.
Saunders, K. J., & Spradlin, J. E. (1990). Conditional discrimination in mentally retarded adults: The
development of generalized skills. Journal of The Experimental Analysis of Behavior, 54 (3), 239-250.
Schroeder, G.L., & Baer, D.M. (1972). Effects of concurrent and serial training on generalized vocal
imitation in retarded children. Developmental Psychology, 6, 293-301.
Schumaker, J. & Sherman, J. A. (1970). Training generative verb usage by imitation and
reinforcement procedures. Journal of Applied Behavior Analysis, 3, 273-287.
Skinner, B.F. (1936). On the generic nature of the concepts of stimulus and response. Journal of
General Psychology.
Stokes, T.F., Baer, D.M., & Jackson, R.L. (1974). Programming the generalization of a greeting
response in four retarded children. Journal of Applied Behavior Analysis, 7, 599-610.
Stokes, T.F, & Baer, D.M. (1977). An implicit technology of generalization. Journal of Applied Behavior
Analysis, 10, 349-367.
Stokes, T.F., Fowler, S. A., & Baer, D. M. (1978). Training preschool children to recruit natural
communities of reinforcement. Journal of Applied Behavior Analysis, 11, 285-294.
© April 30, 2003, Eric V. Larsson, Ph.D.
Generative Language Matrix
Page 37
Striefel, S., Wetherby, B., & Karlan, G. (1976). Establishing generalized verb-noun instructionfollowing skills in retarded children. Journal of Experimental Child Psychology, 22, 247-260.
Tryon, A.S., & Keane, S.P., (1986). Promoting imitative play through generalized observational
learning in autistic-like children. Journal of Abnormal Child Psychology. 14, 537-549.
Warren, S.F., Rogers-Warren, A., & Baer, D.M. (1976). The role of offer rates in controlling sharing
by young children. Journal of Applied Behavior Analysis, 9, 491-497.
Wetherby, B. & Striefel, S. (1978). Application of miniature linguistic system of matrix training
procedures. In R.L. Schiefelbusch (Ed.), Language intervention strategies. Baltimore: University Park
Press.
© April 30, 2003, Eric V. Larsson, Ph.D.