files.eric.ed.gov · Another behavioral process that was clearly explicit in Verbal Behavior was emergent re-combination of behavioral units to form new units. Unit recombination

RECOMBINATIVE READING DERIVED FROM PSEUDOWORD INSTRUCTION IN A MINIATURELINGUISTIC SYSTEM

ELENICE S. HANNA1,2, MARINA KOHLSDORF

1,2, REGIANE S. QUINTEIRO1,2, RAQUEL MARIA DE MELO

1,2,DEISY DAS GRACAS DE SOUZA

2,3, JULIO C. DE ROSE2,3, AND WILLIAM J. MCILVANE

2,4

1UNIVERSIDADE DE BRASILIA2INSTITUTO NACIONAL DE CIENCIA E TECNOLOGIA SOBRE COMPORTAMENTO, COGNICAO E ENSINO

3UNIVERSIDADE FEDERAL DE SAO CARLOS4E. K. SHRIVER CENTER AND UNIVERSITY OF MASSACHUSETTS MEDICAL SCHOOL

A miniature linguistic system was used to study acquisition of recombinative symbolic behavior. Threestudies evaluated the teaching conditions of conditional discriminations with printed and spokenpseudowords that could potentially generate recombinative reading. Fifty-four college students acrossall studies learned to match 12 printed pseudowords to 12 spoken pseudowords. Some also matchedpictures to the same spoken words. Each two-syllable pseudoword was formed by symbols from anarbitrarily created alphabet composed of four vowels and four consonants. Letters had univocalcorrespondence with phonemes. Recombinative receptive reading, comprehensive reading, and textualresponding to pseudowords were periodically assessed. Experiment 1 (n 5 20) showed thatrecombinative reading increased as the number of trained words composed of the same symbolsincreased. Experiment 2 (n 5 14) showed that overtraining the same two words did not producerecombinative reading for most participants. Experiment 3 (n 5 20), in which training with pictures wasomitted, showed that elemental control by within-syllable units can develop even when the trainedpseudowords are meaningless (not related to pictures). The present results support the utility of theminiature linguistic system methodology for identifying and controlling environmental determinants ofrudimentary reading skills.

Key words: reading, recombination of behavioral units, stimulus equivalence, textual behavior,miniature linguistic system, college students, button presses

_______________________________________________________________________________

In Verbal Behavior (1957), Skinner providedwhat he termed an ‘‘exercise in interpreta-tion,’’ endeavoring to relate various commu-nication repertoires (listening, speaking, read-ing, writing, etc.) to fundamental processesthat were originally defined in basic researchwith laboratory animals. For some years, thisexercise did not seem to inspire muchtheoretical extension or empirical analysis,but that is no longer true. For example,stimulus equivalence (Sidman, 1994) is onebehavioral process that was perhaps implicit inVerbal Behavior and is clearly necessary for theanalysis of verbal relations. That topic hasinspired voluminous research that (1) has ledto fundamental insights about complex hu-man behavior (e.g., Hayes, Barnes-Holmes, &Roche, 2001) and (2) has come to berecognized through basic studies as a funda-mental behavioral process (Catania, 2001;Sidman, 2000). Equivalence class formationhas thus become a standard process-levelanswer to the historical criticism that behavioranalysis could not explain the emergence ofnovel behavior.

The research was supported by The National Council forScientific and Technological Development (CNPq, Grants310230/2007-1 and 400359/2008-1), Brasilia, Brazil, by theFoundation for Research Support in the State of Sao Paulo(FAPESP, Grants 03/09928-4 and 05/56502-8) and byFINATEC. E.S.Hanna, D.G. de Souza, and J. de Rose weresupported by a research productivity fellowship from CNPq.William McIlvane was supported by grants from the USNational Institute of Child Health and Human Develop-ment (HD04147, HD 25995). All authors are currentlyaffiliated to Instituto Nacional de Ciencia e Tecnologiasobre Comportamento, Cognicao e Ensino, supported byFAPESP (Grant # 08/57705-8) and CNPq (Grant #573972/2008-7) which provided support for preparationof this manuscript. Some data of Experiments 1 and 3 comefrom Regiane S. Quinteiro’s master dissertation supervisedby Elenice Hanna in the Post-Graduation Program inPsychology, University of Brasılia, Brazil.

We thank the following students for their help in datacollection: Adriana Pinheiro Carvalho, Camila Akemi,Cinthia Camimura, Fernanda Trancoso de Morais, LicioAndrade, Luciano Oliveira, Patrıcia Serejo de Jesus, RobertaLadislau, Talita Leao, Thiago C. Costa and Virgınia D. Fava.We thank two anonymous reviewers and Douglas Greer fortheir criticisms that improved the manuscript.

For additional information about the study or reprintrequests, contact Elenice Hanna, Instituto de Psicologia,Universidade de Brasılia, Campus Darcy Ribeiro, 70.910-900, Brasılia – DF, Brazil (e-mail: [email protected]).

doi: 10.1901/jeab.2011.95-21

JOURNAL OF THE EXPERIMENTAL ANALYSIS OF BEHAVIOR 2011, 95, 21–40 NUMBER 1 ( JANUARY)

21

Another behavioral process that was clearlyexplicit in Verbal Behavior was emergent re-combination of behavioral units to form newunits. Unit recombination also results in theemergence of novel performances, thus add-ing another important process-level answer tothe historical criticism. In this light, it issurprising that there have been virtually nobasic research studies exploring recombina-tive processes in behavioral journals. We knowof only one such study (Goldstein, 1983), andnone have appeared in JEAB (stimulus substi-tutability in sequence classes, e.g., Wulfert &Hayes, 1988, seems to be the closest relatedtopic).

Some unit recombination research hasbeen done by behavior analysts interested inacquisition of rudimentary reading reper-toires and other relations established con-ventionally by the verbal community (e.g.,de Rose, de Souza, & Hanna, 1996; deSouza, de Rose, Faleiros, Bortoloti, Hanna,& McIlvane, 2009a; Hubner, Gomes, &McIlvane, 2009; Matos, Avanzi, & McIlvane,2006; Mueller, Olmi, & Saunders, 2000).Valuable as these studies are, they do notconstitute basic research in the usualsense—exactly because they are concernedwith conventional relations that participantsmay have encountered in some form before.This same problem, of course, had to bedealt with in stimulus equivalence research.After a few initial studies with conventionalrelations (e.g., Sidman, 1971), basic stimu-lus equivalence research studies have cometo use arbitrary stimuli that are unrecogniz-able as verbal stimuli and which are virtuallyimpossible to name or sometimes even todescribe. Goldstein (1983) seems to bealone in using that approach to study unitrecombination, usually called recombinativegeneralization1 (Goldstein, 1993; Sucho-wierska, 2006).

Notably, other branches of behavioral sci-ence (e.g., psycholinguistics) have been inter-ested in the same issues of experimentalcontrol that have inspired the current stimulusselection practices in stimulus equivalenceresearch. Moreover, they have applied them

to study recombinative processes. One com-mon, longstanding solution has been toemploy what have been termed miniaturelinguistic systems or miniature artificial languages(Berko, 1958; Braine, Brody, Brooks, Sudhal-ter, Ross, Catalano & Fisch, 1990; Esper, 1925;Foss, 1968; MacWhinney, 1983).

A miniature linguistic system consists of a setof stimuli especially created for researchpurposes that may vary along one or moredimensions (e.g., shape, position) and a set ofspecific responses to be made in the presenceof those stimuli (Foss, 1968). Use of stimuliwith no extraexperimental history allows oneto do well-controlled basic studies of recombi-native processes with language-experiencedparticipant populations such as college stu-dents and verbal children.

This article describes the use of a miniaturelinguistic system to study acquisition of behav-ioral unit recombination in college students.Our object was to model experimentallynaturally occurring unit recombination inconventional reading, writing and other sym-bolic communication tasks. A series of system-atic replications with minor variations was setto allow potentially informative post hoccomparisons of the results of these studies.

A pseudoalphabet of eight ‘‘letters’’ wascreated, four corresponding to vowels and fourto consonants. Tests of recombinative ‘‘read-ing’’ of pseudowords (constructed from theseletters) were alternated with training sessionsof basic relations between spoken words andabstract pictures (AB) and between spokenwords and printed words (AC). As the numberof trained relations increased, new AC rela-tions, pseudoword comprehension (BC/CB),and oral textual responding (CD) to novelpseudowords were tested.

The first experiment examined the acquisi-tion process and the emergence of recombi-native performances and pseudoword namingas the amount of training with new pseudo-words increased. The second experimentevaluated whether effects of the teachingprocedure would be maintained when thesame two pseudowords were overtrained in-stead of adding new combinations. The thirdexperiment omitted the history of matchingpictures to spoken words, and evaluated theamount of behavioral unit recombination thatemerged when only spoken word–printedword relations were taught.

1 Since the use of the word generalization is controversialin the context of discriminative control by elements (Alessi,1987), verbal or nonverbal, the terms recombinative readingand emergent textual behavior will be preferred in this paper.

22 ELENICE S. HANNA et al.

EXPERIMENT 1

Translational and applied studies of recom-binative generalization showed that recombi-nation increases with amount of training andwith the number of relations mastered by thestudent (e.g., de Rose et al., 1996; Goldstein,1983; Hanna, de Souza, de Rose & Fonseca,2004; Matos et al., 2006; Mueller et al., 2000).The goal of this experiment was to takeadvantage of the experimental control allowedby the miniature linguistic system to study thiseffect. The number of trained relations in-creased along six cycles of training and testingconditions. Each cycle was designed to yieldequivalence classes consisting of two spokenwords and the corresponding pictures andprinted words, as well as textual responses tothe printed words (i.e., oral reading of thesewords). Recombination was probed after eachcycle. The experiment also aimed to investi-gate whether training with syllabic recombina-tion would produce reading of novel pseudo-words with within-syllable recombination.

METHOD

Participants

Twenty undergraduate students (11 males),enrolled in Introductory Psychology classes,volunteered to participate. Their ages rangedbetween 18 and 24 years (mean 20), and theirnative language was Portuguese. Half of themwere students of Natural Sciences and Engi-neering and the other half were students ofHealth or Social Sciences. Participants had twoor three periods of 30 min per week availablefor the sessions. They read and signed aninformed consent form, which stated that theaim of the research was to investigate symbolicbehavior, that participation was voluntary, andthat they could withdraw at any time. Studentsearned course credits for research participa-tion that could increase their IntroductoryPsychology grades up to 5%.

Setting and Apparatus

A sound-attenuated room of 9 m2 housed aMacintosh Performa 6230 computer thatprogrammed reinforcement contingenciesand recorded data through the MTS 11.6.7software (Dube, 1991). Participants selectedstimuli using the computer’s mouse. Theexperimenter sat to the right of the partici-pant, with the keyboard in front. An indepen-

dent observer sat at the right diagonal cornerof the room. Experimenter and observer werepresent during all sessions and recorded oralresponses on response forms. They could notsee each other’s recorded observations. Anaudio recorder and tapes also recorded oralresponses.

Stimuli

Stimuli were spoken pseudowords (A),pictures (B), and printed pseudowords (C).All spoken pseudowords were formed by twoconsonant–vowel (CV) syllables (e.g., FALE),therefore consisting of four phonemes. Spo-ken words were pronounced with stress on thelast syllable (e.g., |fale9|), different from thePortuguese language in which most wordshave stress on the penultimate syllable. Re-corded male and female voices were producedby two external computer loudspeakers.Twelve pseudowords were used in training,each formed by two of the four possiblesyllables NI, BO, FA and LE (e.g., NIBO andFALE). Fourteen two-syllable words, formed bynew combinations of the letters, were used fortesting (e.g., BENA and LOFI). The pictureswere ambiguous forms that resembled animalsand toys, all selected from a stimulus library(Dube, 1991). Printed words were written withletters from an invented alphabet whosecharacters resembled Greek letters. Lettershad a univocal correspondence with B, F, L,N, A, E, I and O and their respectivephonemes in the Portuguese language. Noneof the words had meaning in Portuguese.Figure 1 shows all visual stimuli and phonetictranscriptions of auditory stimuli.

Stimuli used for training sessions combinedsyllables NI, BO, FA and LE, so that all syllableswere used once in each cycle (see Figure 1). InAC relations (match printed word to spokenword sample), the correct comparison was theprinted word that corresponded to the spokenword according the rules of the miniaturelinguistic system. Incorrect comparisons werealso two-syllable words which had only onecommon letter and in the same position as thecorrect word (S+). The other three letters ofthe S2 did not occur in the S+. In Cycle 1, forexample, when the sample was |fale9|, theprinted word FALE was the S+ and F BO,BANI, NOLI or BINE could be S2 (all stimuliwere written with the pseudoword alphabet).The choice of S2 pseudowords was designed

RECOMBINATIVE READING IN MLS 23

to prevent restricted control by one of theletters in one location.

Interobserver Agreement Measures

Interobserver agreement checks were con-ducted for oral textual responses duringtraining and tests. Two independent observers(including the experimenter) recorded oralresponses. Each of these responses was con-sidered a unit and scored as an agreement ordisagreement. Interobserver agreement wasobtained using a point-by-point formula (Kaz-din, 1982): the total number of agreementswas divided by agreements plus disagreementsand multiplied by 100. Agreement scores wereobtained for 3578 responses (98% of thetotal), and agreement was 96.7%. For most

disagreements, listening to the audiotape ofthe session made clear which responses shouldbe included in the data analysis.

Procedure

All participants initially learned to matchtwo patterns to two related geometric forms, aspretraining for the arbitrary matching proce-dure.

Table 1 shows the procedures and order ofconditions of Study 1. For each set of words,training and testing followed a sequence oftasks consisting of a cycle, as exemplified inTable 1. Six cycles were conducted. In eachcycle, participants learned stimulus relationsAB (matching picture to spoken-word sample)and AC (matching printed-word to spoken-

Fig. 1. Spoken pseudowords (A), pictures (B), and written pseudowords with the invented alphabet (C) used astraining and testing stimuli in each cycle of Experiment 1.


word sample), as well as textual responding(CD - oral reading) of two printed words untilcriterion was reached. Cycles ended with testsfor (1) emergence of equivalence classes (BCand CB matching with trained words) or wordcomprehension, (2) recombinative readingcomprehension (BC and CB matching withnovel words), (3) recombinative oral reading(textual responses for novel words), and (4)recombinative receptive reading (AC match-ing with novel printed pseudowords).

AB and AC relations with two different two-syllable words were taught in each cycle. Testsused the words FALO and BENA in all cycles(‘‘constant’’ testing words) that were com-posed of elements of the training stimuli.Participants were never explicitly taught torelate these printed words to the respectivespoken words. Students learned, however, torelate these words, in spoken form, to pictures,to allow for testing of the relation of thecorresponding printed words to the respectivepictures (recombinative reading comprehen-sion). Each cycle tested recombinative perfor-mances of two additional words which weredifferent in each cycle (e.g., NIBE and LOFI inCycle 1; see Figure 1). This permitted a test ineach cycle of recombinative reading of words

that were completely new. Note that these newwords had not been related to pictures, so theyhad no ‘‘meaning’’ in this invented systemand, therefore, it was not possible to testreading comprehension of these words.

AB training – matching pictures to spokenpseudowords. AB training consisted of 60matching trials to establish relations betweenfour pictures and four corresponding spokenwords (see Table 1). Two of these words(called training words) were later used in ACtraining, and the other two (called testingwords) were used later to test recombinativereading comprehension. Cycles that followedthe first one did not include review trials withtraining words from previous cycles. Testingwords were repeated every cycle.

Instructions, presented on the screen at thebeginning of each session, consisted of aPortuguese version of: ‘‘Listen and click onthe black square presented in the center of thescreen and then choose one of the picturespresented in the bottom. If you get it right, thecomputer will show stars or the word ‘Correct’on the screen’’.

Matching trials began with the presentationof the sample stimulus: A black square wasdisplayed on the center key of the screen

Table 1

Sequence of training and test conditions of Experiment 1.

Conditiona Task Stimuli Exampleb

Training conditions of each cycle

1 Matching Pictures to Spoken Words (AB) 4 words/cycle NIBO, FALE, FALO, BENAc

2 Matching Printed to Spoken Words andTextual Responding (AC/CD)

2 words/cycle NIBO, FALE

Tests of each cycle

3 Assessment of Equivalence Classes: Picture-Printed Word and Printed Word-PictureMTS (BC/CB)

2 training words/cycle NIBO, FALE

4 Recombinative Reading Comprehension: Picture-Printed Word and Printed Word-PictureMTS (BC/CB)

2 test words/cycle FALO, BENA

5 Recombinative Oral Reading: Textual responding(CD)

4 test words/cycle NIBE, LOFI, FALO, BENA

6 Recombinative Receptive Reading: Printed toSpoken Word MTS (AC)

4 test words/cycle NIBE, LOFI, FALO, BENA

Final Test

7 Textual Responding (CD) 12 training words see Figure 114 test words

a Conditions 1 to 6 comprise an Experimental Cycle. Six Experimental Cycles were performed with training of twodifferent pseudowords in each cycle.

b Example based on Cycle 1; see Figure 1 for other cycles.c FALO and BENA were test stimuli for recombinative reading. The teaching of relations between these spoken words

and pictures allowed for testing recombinative reading comprehension in Condition 4.


monitor and a word was spoken and repeatedevery 15 s. A mouse click on the black squaredisplayed pictures on the bottom keys. Thespoken word continued to repeat until theparticipant selected one of the comparisonstimuli. A 1-s display of flashing stars accom-panied by a series of computer-generatedchimes or the presentation of the printedword ‘‘Correct’’ and a beep followed correctkey selection. Incorrect responses were fol-lowed by 3 s of black screen and the repetitionof the trial. The correct choice stimulus didnot appear in the same position on more thantwo consecutive trials. The intertrial interval(ITI) was 1.5 s.

Training trials were distributed across 15successive blocks that gradually increased thenumber of samples and comparisons. Table 2shows the sequence of training blocks. A singleAB relation was trained initially, with the samesample displayed on successive trials. Thenumber of comparisons increased from oneto three over successive trials (blocks 1 to 3,Table 2). Two errors within blocks of two orthree comparison stimuli led to a decrease inthe number of choices (backup procedure).After training two relations, samples wererandomly alternated with two- and three-comparison stimuli. Once all relations hadbeen presented with three comparison stimuli,the cumulative final block of trials (block 15)presented three trials of each sample with thecorrect comparison stimulus once in eachposition. Training in a particular cycle endedif the participant scored 100% in the finalblock, otherwise the cumulative block could berepeated up to two additional times. If thecriterion was not reached after the thirdpresentation of the cumulative block, thesession ended and the training was repeatedfrom the beginning in the next session.

AC/CD training – matching printed to spokenpseudowords and textual responding. AC Trainingwas similar to AB training, with the differencesthat only two AC relations were trained in eachcycle, rather than four relations (as in ABtraining), and that probes verifying textualbehavior to the printed pseudowords (CD)were interspersed within matching blocks.Table 2 shows the sequence of training blocks.Training in the first cycle illustrates thetraining procedures, which repeated withdifferent words in all other cycles. RelationA1C1 was trained first, beginning with only the

correct comparison stimulus and increasingthe number of comparisons to two and threein subsequent trials. Two probe trials thenverified oral reading (CD) of the printed wordC1. If reading was not correct, the experi-menter said the correct pseudoword and theA1C1 training trials with three comparisonswere repeated until the participant read C1correctly in oral reading trials. Subsequenttrials taught the A2C2 relation, starting withonly the correct comparison and increasingthe number of comparisons to two. Subse-quent trials alternated samples A1 and A2,starting with two comparisons and increasingto three. A2C2 trials with three comparisonswere presented twice as often as A1C1. Twotrials then verified reading of C2. Any readingerror repeated these AC trials, alternatingagain samples A1 and A2 with three compar-ison stimuli. AC training ended with a finalblock of six trials with A1 and A2 samples in amixed sequence (Block 10), followed by ablock of four textual responding trials (Block11), two with each word, in a mixed sequence.Errors in either of those two blocks (10 and11) repeated the matching-trial block (Block10) until 100% correct performance wasachieved before the final block of textualresponding trials was presented (Table 2).The return to the matching block could occurup to three times. If an error was made in thethird block, the session ended and training wasrepeated from the beginning in the followingsession. If all responses were correct, sessionended.

On textual responding (CD) trials, a singleprinted pseudoword was displayed in thecenter-bottom key. On the first trial, theexperimenter asked ‘‘What are those sym-bols?’’ A drawing of a male face with openmouth above the word signaled the request ofan oral response. The experimenter (as well asthe observer) recorded the participant’s utter-ance and pressed previously defined keys onthe keyboard for a correct or an incorrectresponse for providing feedback to the partic-ipant. In a few cases participants changed aresponse immediately after it was made, beforethe experimenter operated the keyboard toprovide response consequences. In thesecases, both responses were recorded but thelast was considered for feedback and analysis.The correct model spoken by the experiment-er followed incorrect responses. The partici-


pant was not asked to make a correctedresponse at that point. Correct textual re-sponses were followed by the same conse-quences described for matching trials. Theminimum number of AC/CD training trials ineach cycle was 40 matching trials and 8 textualresponse trials.

The next session started with a cumulativeblock of matching trials with differentialreinforcement programmed in all trials (Block10, Table 2). A second block with the sametrials in a different order programmed inter-mittent feedback for correct or incorrectresponses (every third trial, on average). This

Table 2

Structure of AB Training and AC/CD Training: trial type, number of trials with each sample,positive comparison (S+), number of negative comparisons (number of S2), learning criterionto move to the next block (criterion max error) and number of the block to back up when thecriterion was not met (back up block) for each block.

Block Trial typeNumber of

trials Sample S+ Number of S2Criterion max

error Back up block

AB Training1 AB 1 A1 B1 0 - -2 AB 2 A1 B1 1 1 13 AB 3 A1 B1 2 1 24 AB 1 A2 B2 0 - -5 AB 2 A2 B2 1 1 46 AB 2 A1 B1 1 1 5

2 A2 B2 17 AB 3 A1 B1 2 1 6

3 A2 B2 28 AB 1 A3 B3 0 - -9 AB 2 A3 B3 1 1 8

10 AB 2 A3 B3 1 1 92 A2 B2 12 A1 B1 1

11 AB 3 A3 B3 2 1 103 A2 B2 23 A1 B1 2

12 AB 1 A4 B4 0 - -13 AB 2 A4 B4 1 1 1214 AB 2 A4 B4 1 1 13

2 A3 B3 12 A2 B2 12 A1 B1 1

15 AB 3 A4 B4 2 1 15*3 A3 B3 23 A2 B2 23 A1 B1 2

1 AC 1 A1 C1 0 - -2 AC 4 A1 C1 1 1 13 AC 6 A1 C1 2 1 24 CD 2 C1 - - 0 35 AC 1 A2 C2 0 - -6 AC 4 A2 C2 1 1 57 AC 4 A2 C2 1 1 6

4 A1 C1 18 AC 6 A2 C2 2 1 7

3 A1 C1 29 CD 2 C2 - - 0 8

10 AC 3 A2 C2 2 1 10a

3 A1 C1 211 CD 2 C1 - - 0 10

2 C2 - -

Note. Within a block, trials with different samples were mixed.a If errors persisted after two repetitions, training repeated from Block 1 in a new session.


block could be repeated three times if an erroroccurred. After the third repetition, thesession was interrupted and training wasresumed in the next session, beginning withBlock 1. Tests began after 100% correctresponding on the training block with inter-mittent reinforcement.

Testing equivalence classes and recombinativereading. In each cycle, after participants mas-tered the AB, AC and CD relations, the cycleended with tests of stimulus equivalence andthree tests with novel recombined words.

BC and CB relations with training stimuliwere tested in blocks of 12 trials (3 with eachsample). The two types of trials were mixed ina quasirandom order, with no more than twoconsecutive trials of the same type. Threeblocks of trials were programmed but testingcould end earlier if the participant scored100% correct on the first or second block(responses consistent with equivalence wereconsidered correct).

The first recombination test with novelstimuli evaluated the reading comprehension(BC/CB) of the ‘‘constant’’ recombinationwords, FALO and BENA. Twelve trials verifiedmatching of printed words to pictures andmatching of pictures to printed words withthese two pseudowords. The two types of trialswere mixed in a quasirandom order, with nomore than two consecutive trials of the sametype. During this test new ambiguous formswere used as S2, since the use of trainingpictures could artificially inflate scores by‘‘exclusion’’ of the trained pictures (e.g.,Dixon, 1977; McIlvane & Stoddard, 1981).The second eight-trial recombinative testassessed textual responding (oral reading –CD) of the two constant and the two variablenovel pseudowords. The third recombinationtest programmed eight matching trials ofprinted words to spoken words (AC relations)with the two constant and the two variablerecombination words used in each cycle. Inthese matching trials, three comparison stim-uli were displayed on all trials.

Instructions were presented on the screen atthe beginning of each test and consisted of thePortuguese version of: ‘‘Look/listen and clickon the black square, picture or sequence ofsymbols presented in the center of the screen.Then choose one of the stimuli presented inthe bottom. The computer will not tell you whenyour response is right’’.

After the sixth cycle, a general textualresponding test was conducted with all 12training words and 14 recombination words(one trial for each printed word). Tests had noprogrammed consequences for correct orincorrect responses.

RESULTS AND DISCUSSION

Few or no errors occurred during training ofmatching abstract pictures to spoken pseudo-words. With few exceptions, the number oftrials to criterion in this condition was close tothe minimum of 60 programmed trials (medi-an 62 and range 60–105).

In general, few or no errors (median 0 andrange 0–6) occurred during training of match-ing printed to spoken pseudowords and intextual responding to printed pseudowords.Trials to criterion and interparticipant vari-ability decreased across cycles. In the sixthcycle all participants ended the training with47–49 trials, with only 4 showing errors. Forsome participants, however, the number oferrors was higher than in AB training. Aprocedure of fading-in the incorrect compar-ison stimuli in blocks of trials with the samesample proved effective such that few errorsoccurred in training.

Matching printed words to pictures and vice-versa (equivalence or reading comprehension)emerged for all participants in all cycles. Tenstudents scored 100% correct on both trialtypes of all cycles. The remaining 10 scoredless than 100% correct in only one cycle. Thus,the training conditions were sufficient toestablish the equivalence between 12 pictures,12 spoken and 12 printed words. These resultsreplicate previous studies with pseudoalpha-bets used with literate and preschool children(Albuquerque, Hanna & Ribeiro, 1998) andthose employing meaningful Portuguesewords (e.g., de Rose et al., 1996; Matos et al.,2006; Matos, Hubner, & Peres, 1997; Matos,Hubner, Serra, Basaglia, & Avanzi, 2003) andEnglish words (e.g., Sidman, 1971).

Figure 2 shows accuracy on tests that evalu-ated three types of performances with newwords, formed by recombining elements oftraining words, conducted after training oneach cycle: equivalence or reading compre-hension, listening comprehension, and pseu-doword textual responding. Mean percentageof correct responses for the group of partici-pants was calculated for tests of matching


pictures to printed words and vice-versa (BC/CB), for tests of matching printed to spokenwords, and for emergent pseudoword namingor textual responding (CD) tests. Accuracy inall tests increased across successive cycles.Mean scores for matching performances werehigher than those for textual responding;large standard deviations (vertical line) indi-cate interparticipant variability, however.These results indicated that the number ofinstructional trials was an important factor inthe development of recombinative reading.The words used in each cycle contained thesame four syllables (NI, BO, LE, FA). Thesmall number of letters (8) and the repeatedexposure to the same syllables in differentwords and in different positions within a word(the first or the last) allowed for developmentof control by textual units smaller than thewhole words used in training. Thus, theseresults support Skinner’s assertion that controlby minimal units may develop from exposureto larger units (Skinner, 1957); once thestimulus control is established, the stimulusunit continues to control the correspondingresponse unit even when recombined withother stimulus units (Alessi, 1987).

Most participants showed some degree ofrecombinative reading after training with sixwords (third cycle) when each of the foursyllables had been presented in three differentwords and in the two possible positions. Groupanalysis (Figure 2) shows that for textual

behavior, larger changes on recombinationscores occurred in the first three cycles than inthe last three, although there was an increas-ing trend in accuracy until the last cycle. After12 words had been trained, 18 out of 20participants read at least one new word in therecombination test of the sixth cycle.

An analysis of recombinative textual re-sponding (not shown) revealed higher scoresfor the constant words than for the differentwords in all cycles. These differences could bedue to the repeated exposure but also to theparticipants’ history in the AB training, inwhich the constant words were related topictures. Subsequent study sought to clarifythis difference.

Figure 3 shows individual scores on the finaltest of textual responding to printed pseudo-words (oral reading), conducted at the end ofthe study with all words. Scores for training(upper panel) and recombined (lower panel)words were similar for all participants. Thir-teen students correctly named more than 50%of the 12 training words and more than 75% ofthe 14 recombined words. For the 7 remainingparticipants, scores were less than 20% correctfor training and recombined words. Thus,scores for training and novel words were highlycorrelated and high accuracy was found for65% of the participants in tests of recombina-tive reading.

Control by within-syllable units developedgradually after participants learned condition-al relations between pictures and spokenwords, and between printed and spokenwords. Characteristics of the miniature lan-guage system may have facilitated the devel-opment of control by within-syllable textualunits and these units continued to exertaccurate control when recombined in novelstimuli. The few elements to discriminate,systematic manipulation of the syllables’ posi-tion within training words (so that eachsyllable was trained the same number of trialsas the first and as the last component of awhole word), and the univocal correspon-dence between letters and phonemes mayhave been important aspects of the discrimi-native training which helped to establishcorrespondence between syllables and theirrelated sounds as well as the emergence ofrelations between printed letters and pho-nemes. The demonstration of recombinativereading on the level of printed letters and

Fig. 2. Percentage of correct selection responses(group mean) in tests of recombinative reading in eachexperimental cycle of Experiment 1: matching printedwords to spoken word (SW–PW MTS, triangles), matchingpictures to printed word and printed words to picture(Pic–PW/PW–Pic MTS, squares) and textual respondingto printed word (Textual Beh, circles). Vertical lines showone standard deviation from the mean.


phonemes is a robust and additional demon-stration (e.g., de Rose et al., 1996; Mueller etal., 2000) that minimal units not presentedindependently in a contingency of reinforce-ment can develop from larger units (Skinner,1957).

However, 7 students did not show recombi-nation of repertoires even after the trainingwith 12 words. It is possible that theseparticipants would show recombination if theyhad been exposed to additional training, sinceparticipants showed individual differences in

the amount of training needed for recombi-nation. The 12 training words, however,exhausted all the possible combinations ofthe four syllables into different two-syllablewords, so that increasing the amount oftraining would require the use of three-syllablewords or the use of a larger number ofsyllables.

Each cycle in this study was independentand taught specific words. This facilitatedlearning of the conditional relations butretention in the final test of naming printed

Fig. 3. Individual scores (percent correct) in the final test of textual responding to training (upper panel) andrecombined pseudowords (lower panel) of Experiment 1.


words (textual behavior) was low for theparticipants who showed inaccurate recombi-nation. These results contrast with previousstudies with Portuguese words (e.g., de Rose etal., 1996; de Souza et al., 2009a) that used acumulative training in which words learned inprevious cycles continued to be trained asbaseline words. Although some participants inthe de Rose et al. study showed poor recom-bination scores, they all showed high scores fortraining words, presumably due to the cumu-lative training. Because these studies usedstimuli with which the participants had extra-experimental histories, it was not possible toascertain whether the high scores on trainingwords were a joint product of the instructionand past experience. In the present study, bycontrast, the miniature linguistic system meth-ods eliminated concerns about previous histo-ry with all stimuli in determining the results(although it is conceivable that a history ofabstraction in one language may influenceacquisition in another or in a contrived one).

EXPERIMENT 2

The results of Experiment 1 showed thatrecombinative reading increased across cyclesas the number of trained words increased. Ineach cycle, participants learned to match andread two different words formed by differentcombinations of the same four syllables. Theposition of the syllables was varied betweencycles. This procedure included two aspectsthat should be evaluated for their importancein explaining recombinative reading: (1)behavior in the presence of the same foursyllables was reinforced during training of allsix cycles; and (2) stimulus elements (syllables)were recombined to form new words and newresponses controlled by these training wordswere established in each cycle. It is possible,however, that mere exposure to training withthese syllables in the same position couldproduce recombinative reading. Thus, Exper-iment 2 also used six cycles, but the proce-dures trained the same two words in everycycle. In this study the syllabic units weretrained the same number of times as inExperiment 1 and behavior was reinforced inthe presence of the four syllables duringtraining of six cycles. However, participantswere not exposed to stimuli that placed thesyllables in different positions.

METHOD

Participants

Participants were 14 undergraduate studentsaged 17 to 24 years (M 5 20.5), 7 male and 7female. Half of them were from NaturalSciences and Engineering and the other halfwere enrolled in Health and Social Sciencescourses.


Setting, apparatus, stimuli, and interobserv-er agreement measures were the same as inExperiment 1. Agreement scores were ob-tained for 1572 responses (99% of the total),and agreement was 98.4%.

Procedure

The procedure was the same described forExperiment 1, except that the words (spokenand printed) used in training conditions of allcycles were always NIBO and FALE. Thetraining conditions of Cycles 2 to 6 were thesame as Cycle 1.

Six participants were exposed to all trainingtrials in every cycle (Participants 24, 28, 29, 30,31 and 34). For the remaining 8 participantstraining trials were reduced for Cycles 2 to 6:AB Training from 60 to 12 trials and AC/CDTraining from 47 to 20 trials. The reductionwas in response to participant complaintsabout the repetitive, boring nature of the task.For these 8 students, the final cumulativeblocks were conducted twice (block 15 of ABTraining and blocks 10 and 11 of AC/CDTraining). For all participants a score of 100%correct in the cumulative block was required toend the session. Correction procedures werethe same as those described in Experiment 1.


This study evaluated whether effects of theteaching procedure would be maintainedwhen the same two pseudowords were over-trained instead of adding words with recombi-nation of the syllables. Each syllable waspresent once in every cycle, as in Experiment1, but a syllable never changed position(beginning or ending of the word) and alwaysbelonged to a single word. Performance in thefirst cycle was similar to Experiment 1. Averagenumber of trials to criterion in matchingpictures to spoken words was 76.3, rangingfrom 62 to 162. In matching printed to spoken


words and textual responding to printedwords, average number of trials to criterionin Cycle 1 was 50.5, ranging from 47 to 55.

For 6 participants the following cycles (2through 6) overtrained these tasks with thesame number of trials and the others weretrained in the cumulative block until criterionwas reached. None of the participants mademore than one error in one cycle and manymade no errors at all. All 14 participantsscored 100% correct across the six cycles thattested matching printed words and picturesand vice-versa with training stimuli, showingthat, for training words, equivalence betweenprinted words, spoken words, and picturesalready emerged in the first cycle.

Results of the recombinative tests are shownin Figure 4. The percentage of correct selec-tions averaged for all participants during thethree tests of recombinative reading conduct-ed in each cycle (reading comprehension,listening comprehension and emergent oralreading or textual responding) was similar tothat of Experiment 1 in the first cycle. In thefollowing cycles, contrary to what was found inExperiment 1, there was no increase inmatching accuracy and there was only amoderate increase in textual responding accu-racy along the first three cycles. As inExperiment 1, the analysis of recombinativetextual responding (not shown) revealedhigher scores for the constant words than forthe different words in all cycles. Thesedifferences could be due to the repeatedexposure, but also to the participants’ historyin the AB training, in which these words wererelated to pictures.

Only 4 of the 14 participants scored higherthan 50% correct in the final test of textualresponding to novel pseudowords (Figure 5).Participants 31 and 34 had the same numberof training trials as in Study 1, and the other 2(Participants 32 and 33) were exposed to thereduced training. Although students readcorrectly the two training words in this finaltest (except Participants 21 and 22), thetraining with only two words did not producerecombinative reading for the majority of theparticipants. Thus, these lower scores in thisstudy cannot be attributed merely to thereduced training, since some participantsexposed to the regular and some exposed tothe reduced training produced similarly lowscores (respectively, 4 and 6 participants),

while others presented similarly high scores(2 participants under the reduced trainingand 2 participants under the regular training).

The results of Experiment 1 (but not ofExperiment 2) systematically replicated, withtwo-syllable pseudowords, the results of Muel-ler et al. (2000) with English one-syllablewords. The stimuli in both studies were wholewords (dictated and printed words), but theprocedures manipulated the within-word com-ponents (syllables or onset and rhymes) bysystematically varying sets of words with over-lapping syllables (Experiment 1) or letters(Mueller et al.). This overlapping of compo-nents was missing in Experiment 2. Since theovertraining with the same components didnot prove sufficient for generating recombi-native reading, the results suggest that therecombination of elements of the trainingstimuli is a critical factor in recombinativereading. This finding is consistent with studiesshowing that multiple exemplars varying ele-mental position appear critical for abstractionof novel words (e.g., de Rose et al., 1996;Mueller et al.). In some previous studies, thisvariable was confounded with number ofwords, which also employed components ofdifferent words included in training, althoughthe recombinations were not as systematic asthose planned in Experiment 1 or in the studyof Mueller et al. For example, de Rose et al.taught the words SELO and BOLO (present-



ing the syllable LO in two different words);VACA and CAVALO (thus presenting thesyllables VA and CA in different words and indifferent positions within the words; and

training the syllable LO in one more word);such examples increased as the study pro-gressed. In light of the results of Experiment 2(i.e., the lack of recombinative performances),

Fig. 5. Individual scores (percent correct) in the final test of textual responding to training (upper panel) andrecombined pseudowords (lower panel) of Experiment 2. Participants with reduced training are marked with asterisks.


one could infer that the use of recombinedstimuli during training (and not merely theincrease in the number of training words) wasa critical aspect in those previous studies.Future investigation could evaluate whetherthe mere increase in the number of trainedwords without recombining any word elementswould result in recombinative reading oftrained elements presented only in tests.However, the robust effects in Experiment 1,combined with those of Mueller et al., stronglysuggest that the experience with repeatableunits in different words and word positions is anecessary condition for (1) the establishmentof abstraction of those units, and for (2) theintegrity of the accurate stimulus control thatthey exert when inserted within or recom-bined with other stimuli (see Goldstein, 1983;Striefel, Wetherby, & Karlan, 1976; Wetherby& Striefel, 1978, for similar processes withother verbal behaviors).

EXPERIMENT 3

Experiments 1 and 2 were based on thestimulus equivalence paradigm: Baseline rela-tions (AB and AC) were taught and emergenceof other relations was tested. Previous reportsshowed that once an individual learns to selectboth a picture and a printed word uponhearing the corresponding spoken word,s/he will also select the picture upon seeingthe printed word and vice-versa, performancesthat have been considered to be a rudimentaryform of reading comprehension (Sidman,1971). Textual responses to novel pseudo-words, which require the recombination oflinguistic units, are not necessarily meaningful(e.g., many Portuguese-speaking persons canname words in Latin even when they do notknow the meaning of these words). In thisstudy, we evaluated the importance of learningto match pictures to spoken words for theacquisition of recombinative textual responses,given that the crucial prerequisite seems to bestimulus control by textual units over theproduction of speech sounds. Do equivalenceclasses formed by combinations of pictures,spoken words and printed words promote thedevelopment of the control by smaller linguis-tic units?

To address this question, Experiment 3 useda procedure similar to Experiment 1, but weomitted the training of conditional relations

between pictures and spoken words. Thetextual-response test with recombined wordswas programmed after training printed-pseu-doword-to-spoken-word matching in each cycleof the study. The number of participants whodeveloped recombinative reading with thisprocedure was thus available for comparisonwith Experiment 1. If substantially fewerparticipants exhibited recombinative reading,then that outcome would support the conclu-sion that teaching spoken word–picturematching could facilitate recombinative read-ing. If not, then it would appear that recombi-native reading could emerge merely from theteaching relations involving spoken and print-ed pseudowords.

METHOD

Participants

Twenty undergraduate students (10 maleand 10 female) ranging in age from 17 to30 years (mean 20.8) participated. Half werefrom Natural Sciences and Engineering andthe other half were from Health and SocialSciences courses.


Setting, apparatus, stimuli and interobserveragreement measures were the same as inExperiment 1. Agreement scores were ob-tained for 3642 responses (99% of the total),and agreement was 97.5%.

Procedure

The procedures were similar to those ofExperiment 1 with the omission of training(Condition 1 in Table 1) and testing withpictures (Conditions 3 and 4 of Table 1).

Six independent cycles of training andtesting conditions were programmed. A cyclestarted with training of two conditional rela-tions between printed and spoken words (AC)and textual responding of the two printedpseudowords (CD). Two tests of recombinativereading ended the cycles: (1) Oral textualresponding of four test printed pseudowords;and (2) matching printed to spoken pseudo-words with test words. The AC/CD Trainingwas conducted with different words in eachcycle (Figure 1) as in Experiment 1. A generaltextual responding test with all training andnovel pseudowords was conducted at the endof the study.



This experiment omitted the history ofmatching pictures to spoken pseudowords,and evaluated the amount of behavioral unitrecombination that emerged when only spo-ken–printed pseudoword relations weretaught.

The number of trials to learn conditionaldiscriminations between printed and spokenpseudowords in each cycle was higher than theminimum, due to errors that required repeti-tion of blocks; however, number of trialsdecreased across cycles. In general, trainingof Cycles 1 to 3 was somewhat longer than thatfor participants of Experiment 1 (wherematching pictures to spoken words precededtraining of matching printed pseudowords tospoken words). Group mean decreased from78.8 to 51.3 trials from the first to the sixthcycle, and all participants ended training ofthe last three cycles with no more than oneerror.

In this experiment, the reading comprehen-sion tests were omitted. Although recentstudies have suggested that one need notteach the picture–print–spoken word relations(e.g., Greer & Speckman, 2009; Leite &Hubner, 2009), this would apply only to casesin which the child already learned to relatespecific pictures to the corresponding words aswas the case in the first study by Sidman(1971). When the child has a tact for a‘‘picture’’ as a result of a naming experience,training can be reduced by omitting theteaching of AB relations. In this study,although the college student participants hadan extensive repertoire of naming, it was notfunctional in the context of pseudowords, andreading comprehension required the learningof picture–spoken word matching and/orpicture tacting.

Figure 6 shows mean results from the tworecombination tests conducted in each cycleafter mastering the baseline relations: (1)matching novel printed to spoken pseudo-words and (2) emergent oral reading of novelpseudowords (textual responding). Increasingtrends were found for both measures ofrecombinative performances. The score fortextual responding in the first cycle was verylow (3.8%) and gradually increased throughCycle 4 (50%). There was no further increasein the group mean in the last two cycles, andintersubject variability was high (see vertical

bars in Figure 6 for standard deviations).Percentages were higher for matching (i.e.,listener responding) than for textual respond-ing (i.e., speaker responding).

The increases from Cycle 1 to Cycle 6 weresmaller in the present study than those foundin Experiment 1. In both studies, scores on thetwo measures of recombinative repertoiresincreased as the number of trained wordsincreased, but the gains were higher forparticipants of Experiment 1, who had ahistory of spoken word–picture matchingtraining. These results can be interpreted asevidence of a facilitative effect of a history ofmatching pictures to spoken pseudowords onrecombinative repertoires such as receptivereading (AC) and textual behavior (CD). Butwhy should establishing picture–word relationsfacilitate recombinative repertoires that theyare not dependent upon?

This difference may be related to an aspectof the procedure. In Experiment 1, ABtraining included two dictated word–picturerelations of testing stimuli in recombinationassessments (FALO and BENA, Figure 1).Results of recombinative tests showed higherscores for the constant words than for wordsthat differed in every cycle. This difference wasnot found in Experiment 3. An analysis ofrecombinative textual responding across cycles(not shown) revealed similar scores for theconstant words and for the different words inall cycles. Experiment 1 has, however, two



confounding variables: (1) exposure to thesame textual and auditory stimuli duringtesting conditions of every cycle; and (2)additional teaching of AB relations (pictureswere related to spoken words in every cycle).In Experiment 3, by contrast, there was no ABtraining but the constant words were present-ed in the recombination tests of all cycles.Repeated exposure to the words in several testswas not sufficient, therefore, to producehigher scores of oral reading of these words.Higher scores of recombinative textual behav-ior and receptive reading appear to be relatedto the additional exposure to some of thetesting stimuli during AB training.

Figure 7 shows individual scores in the finaloral reading test (textual responding) for

training (upper panel) and recombined pseu-dowords (lower panel). Twelve of the 20participants scored 70% correct or higher forboth types of words. The remaining 8 studentsshowed 25% or lower scores to training wordsand no correct textual responses to novelprinted words. Scores for training pseudo-words and for novel pseudowords were almostperfectly correlated: If participants read train-ing words accurately, then they also readrecombined words accurately; the others didnot read any words accurately. These resultssuggest a shared underlying process in theemergence of reading training and novelpseudowords. Participants who developedpseudoword reading abstracted sound–printrelations that were smaller than the whole

Fig. 7. Individual scores (percent correct) in the final test of textual responding to training (upper panel) andrecombined pseudowords (lower panel) of Experiment 3.


words during the matching trials. When thoseunits were represented in other words, theycontinued to exert accurate stimulus controlon vocal responses to text (Alessi, 1987;Skinner, 1957).

The proportion of participants who showedrecombinative reading in Experiment 3 (12 of20) was comparable to that of Experiment 1(13 of 20) and much higher than that inExperiment 2 (4 of 14). Thus, recombinativereading in a miniature language system doesnot appear to depend upon (although it maybenefit from) relating words to pictures.Rather, Experiment 3 developed true or‘‘pure’’ textual behavior, isolating it from thecomprehension component of reading. Al-though students did not relate the pseudo-words to pictures, they apparently could soundout the words and recognize their similarity towords dictated to them.

GENERAL DISCUSSION

Three studies showed recombinative read-ing of novel pseudowords with a miniaturelanguage system using three different behaviormeasures that assessed listener and speakerresponding: matching printed to spoken pseu-dowords, textual responding, and picture toprinted pseudowords (reading comprehen-sion).

Emergent Textual Responding andRecombinative Reading

In Experiment 1, 65% of the participantsresponded textually to recombined lettersembedded in new printed pseudowords afterforming 12 classes of equivalent stimuli com-posed of pictures, spoken and printed pseudo-words.

In Experiment 2, only two classes of stimuliwere established with no variation of thesyllables’ position in different training pseudo-words and no overlapping syllables. Thisprocedure was significantly less effective; re-combinative scores of 50% or greater wereachieved by only 28% of the participants. InExperiment 3, no equivalence classes wereestablished between printed and spoken pseu-dowords and pictures. Instead, students ac-quired 12 conditional discriminations betweenspoken and printed pseudowords in which thepositions of syllables varied in different train-ing pseudowords, as in Experiment 1, ensuring

training with overlapping syllables and experi-ence with recombined stimuli. Sixty percent ofparticipants achieved textual respondingscores above 50% correct, comparable to theproportion of participants who did so inExperiment 1.

Overall, this sequence of three demonstra-tion studies showed that learning the phone-mic sounds in a miniature language system ledto abstracted or derived textual responding ofnovel arrangements of the originally taughtprint–sound relations. Moreover, teachingmore print–sound relations was correlatedwith more emergent responses.

The accurate control by single sound–printrelations over elements of responses to novelstimuli, derived from the whole word trainingconditions, systematically replicates and ex-tends results reported by Mueller et al. (2000).In Experiments 1 and 3 participants wereexposed to all possible pairs of combinationswith the syllables BO, FA, LE, and NI (e.g.,NIBO, BOFA, FALE; LENI, etc); this ensuredthat (1) each syllable was presented in the firstand in the last position of a word; (2) eachsyllable participated in a compound with allthe other syllables: and (3) each syllableparticipated as a compound in the samenumber of pseudowords (six). When testsevaluated textual behavior to pseudowordswith within-syllable recombination, most par-ticipants showed recombinative reading. Theseresults were not replicated in Experiment 2, inwhich participants learned only two pseudo-words, NIBO and FALE; although thesepseudowords contained all the elementaryunits (the same four consonants and fourvowels) and were presented in all cycles, thetraining lacked the overlapping feature ofExperiments 1 and 3. Thus, it seems thatexperience with the recombination of ele-ments during training is a necessary conditionfor the development of abstraction (cf. Gold-stein, 1983; Mueller et al., 2000; Striefel et al.,1976; Wetherby & Striefel, 1976).

Results of the present studies taken togetheralso suggest that, although the history ofmatching picture to spoken pseudoword wasnot a necessary condition (Experiment 3), itdid influence the recombinative performance(Experiment 1), increasing scores of readingwords that were previously related to pictures(meaningful words). Experimental historywith picture-matching might select a set of


responses that are emitted more often in arelated context (Hanna, Kohlsdorf, Quinteiro,Fava, de Souza, & de Rose, 2008). When partialstimulus control is also established in thatcontext, efficient performance can resultunder the control of new stimuli (correctrecombinative reading).

In summary, the difference in the numberof participants who recombined textual ele-ments across the three studies suggests that:(1) The critical variable in fostering elementrecombination is not the number of trainingtrials overall, but rather the number ofexemplars of recombinations encounteredduring training, (2) establishing equivalenceclasses involving pictures and pseudowords(spoken and printed) is not necessary forgenerating recombinative performances, and(3) development of recombinative unit-basedinstructional technology (de Souza et al.,2009a; Mueller et al., 2000) is effective inbypassing the challenge of teaching wordattack skills or textual responses to individualswho formerly lacked them.

The present studies add evidence thatmultiple exemplar instruction promotes thedevelopment of abstracted stimulus control ofverbal operants (cf., Becker, 1992; Greer,Stolfi, Chavez-Brown, & Rivera-Valdez, 2005a;Greer, Yuan, & Gautreaux, 2005b; Striefel etal., 1976), extending previous findings toword-element abstraction.

Reading Comprehension

Identifying the critical variables that allowthe recombination of units—especially emer-gent textual behavior—is important for theunderstanding of this component of a readingrepertoire also. Clearly, a key component ofskilled reading is the speaker response as itrelates to print stimuli (i.e., textual respond-ing). Other processes are likely to be impor-tant also. There is increasing evidence, forexample, that the relation between the textualresponse and the speaker acting as listener ofhis/her own productions (Skinner, 1957) islikely to be a key component in readingcomprehension (de Rose, 2005; de Souza, deRose & Domeniconi, 2009; Greer & Ross, 2008;Greer & Speckman, 2009). The special observ-ing responses (listener responses) and pro-duction responses (speaker responses) foundin verbal behavior may be established inde-pendently (e.g., Greer, Stolfi, et al., 2005; Lee

& Pegler, 1982; Skinner, 1957). In a skilledreader repertoire, however, these otherwiseindependent verbal operants become interre-lated in a network of relations, and formationof equivalence relations may be the mecha-nism underlying the interdependence devel-opment (e.g., de Rose et al., 1996; Sidman,1994; Stromer, Mackay & Stoddard, 1992).Reading comprehension was demonstrated inthe first two studies by the emergence ofmatching pictures to printed pseudowords andvice-versa (not directly trained), thus replicat-ing the effects first described by Sidman(1971) with pseudowords and abstract pic-tures. Although an initial observational expe-rience can be the source of comprehension,where a child with a repertoire of other verbalrelations need not have a picture MTSexperience (e.g., Greer & Speckman, 2009),when using a MLS system, picture MTS couldbe useful in establishing the comprehensioncomponent in reading. In Experiment 3, thiscomponent was absent, supporting the asser-tion that responding to print was purelytextual behavior (under the control of print–sounds relations) and not reading with com-prehension.

Methodological Aspects

The present series of studies could have beenreported as a single experiment that variedbetween groups (1) the amount of trainedrelations, (2) the occurrence of overlappingelements in the training stimuli and (3) theoccurrence of AB training. Although theparticipants were all college students and theirbehavioral histories were apparently similar, wedid not (1) equate participant numbers acrossthe studies, (2) match precisely for participantvariables (chronological age, intellectual func-tioning, etc.), (3) equate every detail oftraining, and (4) analyze the data statistically,as is customary in formal matched-groupstudies. Rather, we conducted the studies as aseries of systematic replications (Sidman,1960). This tactic did allow us to demonstrate,however, replications and generalization of animportant behavioral phenomenon (the emer-gence of recombinative repertoires) amongconditions that suggested a role for apparentlycritical variables or conditions (e.g., amount ofexemplars and overlapping stimuli on recom-binative reading, the role of class formation inreading comprehension, etc.).


The robust findings of the present studiessuggest that a miniature linguistic system mayprove useful in analyzing the role of ‘‘hearing’’phonemic sounds in deriving textual respond-ing. They permit precisely controlled, fine-grained analyses that trace teaching conditionsstep by step. One can thus examine thevariables that determine the first instances ofemergence of novel behavior with respect tospecific printed stimuli. The miniature systemcertainly adds to the evidence of researchdone in conventional languages such asEnglish and Portuguese.

By limiting the number of letters, pho-nemes, and trained syllables, by inventingpseudowords with convenient configurations(e.g., the regularity in the sound–print corre-spondences), by specifying some pseudowordsto be nonsense, and by providing a similarhistory with other ‘‘meaningful’’ pseudowordsprior to testing, we identified a favorablecontext to advance the comprehension oftrained and emergent verbal skills. Thismethodology permits simulation of readingacquisition with participants who have alreadylearned to read in their native language. Thus,it may be possible via this procedure toconduct systematic investigation of variableswith readers that may be important also inreading acquisition by nonreaders.

Considering the network of verbal relationsinvolved in reading and writing, one wouldpredict that an individual who learns phonemicsounds for letters may have the behavioralprerequisites for spelling given that s/he hasderived relations between speaking and writing(e.g., de Rose et al., 1996; de Souza, de Rose,Faleiros et al., 2009). Future research with aminiature linguistic system could evaluatewhether the accuracy of emergent writtenspelling would be a function of the accuracyof control by the speech sounds, in line withresults reported by Greer, Yuan et al. (2005) ina study using English words. The presentstudies also have implications for the teachingmethodologies of future research. MTS proce-dures that gradually introduced choice stimuliacross successive trials were used to establishnumerous conditional discriminations with fewor no errors. Minimizing errors will likely helpto minimize multiple teaching failures instimulus equivalence studies that teach manybaseline relations (e.g., Fields, Matneja, Var-elas, Belanich, Fitzer, & Shamoun, 2002) and

perhaps to develop baselines for stimuluscontrol studies that are less contaminated byundesirable competing stimulus control topog-raphies (McIlvane & Dube, 2003).

The methodology used in the present studymay be effective also in guiding translations andapplications of the research findings, as it hasbeen in other fields (e.g., second languageinstruction, MacWhinney, 1983). Moreover, asin the theoretical analysis of stimulus equiva-lence (Hayes et al., 2001; McIlvane & Dube,2003), basic studies of the variables that deter-mine the circumstances under which recombi-nation of behavioral units occurs (or not) haveimportance for theoretical understanding and/or interpretation of complex human behavior.

REFERENCES

Albuquerque, A. R., Hanna, E. S., & Ribeiro, M. R. (1998).Changes in training stimulus composition and control byelements of compound stimuli. Poster presented at the 24thAnnual Convention of the Association for BehaviorAnalysis, Orlando, USA.

Alessi, G. (1987). Generative strategies and teachinggeneralization. The Analysis of Verbal Behavior, 5, 15–27.

Becker, W. (1992). Direct instruction: A twenty-year review.In R. West & L. Hamerlynck (Eds.), Design foreducational excellence: The legacy of B. F. Skinner (pp.71–112). Longmont, CO: Sopris West.

Berko, J. (1958). The child’s learning of English morphol-ogy. Word, 14, 150–177.

Braine, M. D. S., Brody, R. E., Brooks, P. J., Sudhalter, V.,Ross, J. A., Catalano, L., & Fisch, S. M. (1990).Exploring language acquisition in children with aminiature artificial language: Effects of item andpattern frequency, arbitrary subclasses, and correc-tion. Journal of Memory and Language, 29, 591–610.

Catania, A. C. (2001). The roles of responses and stimuli infunctional equivalence. European Journal of BehaviorAnalysis, 2, 42–45.

de Rose, J. (2005). Behavior analysis of learning to read andspell. Brazilian Journal of Behavior Analysis, 1, 29–50.

de Rose, J. C., de Souza, D. G., & Hanna, E. S. (1996). Teachingreading and spelling: Exclusion and stimulus equivalence.Journal of Applied Behavior Analysis, 29, 451–469.

de Souza, D. G., de Rose, J. C., Faleiros, T. C., Bortoloti, R.,Hanna, E. S., & McIlvane, W. J. (2009). Teachinggenerative reading via recombination of minimaltextual units: A legacy of Verbal Behavior to childrenin Brazil. International Journal of Psychology and Psycho-logical Therapy, 9(1), 19–44.

de Souza, D. G., de Rose, J. C., & Domeniconi, C. (2009).Applying relational operants to reading and spelling.In R. A. Rehfeldt & Y. Barnes-Holmes (Eds.), Derivedrelational responding: Applications for learners with autismand other developmental disabilities (pp. 171–207). Oak-land, CA: New Harbinger Publications.

Dixon, L. S. (1977). The nature of control by spoken wordsover visual stimulus selection. Journal of the Experimen-tal Analysis of Behavior, 27, 433–442.


Dube, W. V. (1991). Computer software for stimulus controlresearch with Macintosh computers. Experimental Anal-ysis of Human Behavior Bulletin, 9(2), 28–30.

Esper, E. A. (1925). A technique for the experimentalinvestigation of associative interference in artificiallinguistic material. Language Monography, 1, 5–45.

Fields, L., Matneja, P., Varelas, A., Belanich, J., Fitzer, A., &Shamoun, K. (2002). The formation of linkedperceptual classes. Journal of the Experimental Analysisof Behavior, 78, 271–290.

Foss, D. J. (1968). An analysis of learning in a miniaturelinguistic system. Journal of Experimental Psychology,76(3), 450–459.

Goldstein, H. (1983). Training generative repertoires withinagent-action-object miniature linguistic systems with child-ren. Journal of Speech and Hearing Research, 26(1), 76–89.

Goldstein, H. (1993). Structuring environmental input tofacilitate generalized language learning by children withmental retardation. In A. P. Kaiser & D. B. Gray (Eds.),Enhancing children’s communication: Research foundations forintervention (pp. 317–334). Baltimore: Paul H. Brookes.

Greer, R. D., & Ross, D. E. (2008). Verbal behavior analysis:Developing and expanding verbal capabilities in childrenwith language delays. Boston: Allyn & Bacon/Merrill.

Greer, R. D., & Speckman, J. (2009). The integration ofspeaker and listener responses: A theory of verbaldevelopment. The Psychological Record, 59, 449–488.

Greer, R. D., Stolfi, L., Chavez-Brown, M., & Rivera-Valdez,C. (2005). The emergence of the listener to speakercomponent of naming in children as a function ofmultiple exemplar instruction. The Analysis of VerbalBehavior, 21, 123–134.

Greer, R. D., Yuan, L., & Gautreaux, G. (2005). Noveldictation and intraverbal responses as a function of amultiple exemplar history. The Analysis of VerbalBehavior, 21, 99–116.

Hanna, E. S., de Souza, D. G., de Rose, J. C., & Fonseca, M.L. (2004). Effects of delayed constructed-responseidentity matching on spelling of dictated words.Journal of Applied Behavior Analysis, 37, 223–228.

Hanna, E. S., Kohlsdorf, M., Quinteiro, R. S., Fava, V. M. D.,de Souza, D. G., & de Rose, J. C. (2008). Diferencasindividuais na aquisicao de leitura com um sistemalinguıstico em miniatura [Individual differences inreading acquisition with a miniature linguistic system].Psicologia: Teoria e Pesquisa, 24, 45–58.

Hayes, S. C., Barnes-Holmes, D. & Roche, B. (Eds.). (2001).Relational Frame Theory: A Post-Skinnerian account ofhuman language and cognition. New York: Plenum Press.

Hubner, M. M. C., Gomes, R. C., & McIlvane, W. J. (2009).Recombinative generalization in minimal verbal unit-based reading instruction for pre-reading children.Experimental Analysis of Human Behavior Bulletin, 27, 11–17.

Kazdin, A. E. (1982). Single-case research designs. New York:Oxford University Press.

Lee, V. L., & Pegler, A. M. (1982). Effects on spelling oftraining children to read. Journal of the ExperimentalAnalysis of Behavior, 37, 311–322.

Leite, M. K. S., & Hubner, M. M. C. (2009). Aquisicao deleitura recombinativa apos treinos e testes de dis-criminacoes condicionais entre palavras ditadas eimpressas [Acquisition of recombinative reading aftertraining and testing of conditional discriminationsbetween dictated and printed words]. Psicologia:Teoria e Pratica, 11(3), 63–81.

MacWhinney, B. (1983). Miniature linguistic systems astests of the use of universal operating principles insecond-language learning by children and adults.Journal of Psycholinguistic Research, 12(5), 467–478.

Matos, M. A., Avanzi, A. L., & McIlvane, W. J. (2006).Rudimentary reading repertoires via stimulus equiva-lence and recombination of minimal units. TheAnalysis of Verbal Behavior, 22, 3–19.

Matos, M. A., Hubner, M. M. C., & Peres, W. (1997).Oralizacao e copia: Efeitos sobre a aquisicao de leiturageneralizada recombinativa [Oral response and copy:effects on the acquisition of recombinative general-ization of reading]. Temas em Psicologia, 1, 47–65.

Matos, M. A., Hubner, M. M., Serra, V. R. B. P., Basaglia, A.E., & Avanzi, A. L. (2003). Redes de relacoescondicionais e leitura recombinativa: Pesquisando oensinar a ler [Conditional relation network andrecombinative reading: Searching the teaching toread]. Arquivos Brasileiros de Psicologia, 54, 284–303.

McIlvane, W. J., & Dube, W. V. (2003). Stimulus controltopography coherence theory: Foundations and ex-tensions. The Behavior Analyst, 26, 195–213.

McIlvane, W. J., & Stoddard, L. T. (1981). Acquisition ofmatching-to-sample performances in severe mentalretardation: Learning by exclusion. Journal of MentalDeficiency Research, 25, 33–48.

Mueller, M. M., Olmi, D. J., & Saunders, K. J. (2000).Recombinative generalization of within-syllable unitsin prereading children. Journal of Applied BehaviorAnalysis, 33, 515–531.

Sidman, M. (1960). Tactics of scientific research. New York:Basic Books.

Sidman, M. (1971). Reading and auditory-visual equiva-lences. Journal of Speech and Hearing Research, 14, 5–13.

Sidman, M. (1994). Equivalence relations: A research story.Boston, MA: Authors Cooperative.

Sidman, M. (2000). Equivalence relations and the rein-forcement contingency. Journal of the ExperimentalAnalysis of Behavior, 74, 127–146.

Skinner, B. F. (1957). Verbal behavior. New York: Appleton.Striefel, S., Wetherby, B., & Karlan, G. R. (1976).

Establishing generalized verb-noun instruction-follow-ing skills in retarded children. Journal of ExperimentalChild Psychology, 22, 247–260.

Stromer, R., Mackay, H. A., & Stoddard, L. T. (1992).Classroom applications of stimulus equivalence tech-nology. Journal of Behavioral Education, 2, 225–256.

Suchowierska, M. (2006). Recombinative generalization:Some theoretical and practical remarks. InternationalJournal of Psychology, 41(6), 514–522.

Wetherby, B. C., & Striefel, S. (1976). Establishinggeneralized verb–noun instruction-following skills inretarded children. Journal of Experimental Child Psychol-ogy, 22, 247–260.

Wetherby, B. C., & Striefel, S. (1978). Application ofminiature linguistic system or matrix training proce-dures. In R. Schiefellbusch (Ed.), Bases of languageintervention (pp. 317–356). Baltimore: University ParkPress.

Wulfert, E., & Hayes, S. C. (1988). Transfer of aconditional ordering response through conditionalequivalence classes. Journal of the Experimental Analysisof Behavior, 50, 125–144.

Received: May 17, 2010Final Acceptance: September 1, 2010


Documents

files.eric.ed.gov · Another behavioral process that was clearly explicit in Verbal Behavior was emergent re-combination of behavioral units to form new units. Unit recombination