Syntactic Competence and Reading Ability in Children* · Haskins Laborarories Status Report on Speech Research 1989, SR-99/l00,180-194 Syntactic Competence and Reading Ability in

Haskins Laborarories Status Report on Speech Research1989, SR-99/l00, 180-194

Syntactic Competence and Reading Ability in Children*

Shlomo Bentin,t Avital Deutsch,t and Isabelle Y. Libermantt

The effect of syntactic context on auditory word identification and on the ability to detectand correct syntactic errors in speech was examined in severely reading disabled childrenand in good and poor readers selected from the normal distribution of fourth graders. Thepoor readers were handicapped when correct reading required analysis of the sentencecontext. However, their phonological decoding ability was intact. Identification of wordswas less affected by syntactic context in the severely disabled readers than in either thegood or poor readers. Moreover, the disabled readers were inferior to good readers injudging the syntactical integrity of spoken sentences and in their ability to correct thesyntactically aberrant sentences. Poor readers were similar to good readers in theidentification and judgment tasks, but inferior in the correction task. The results suggestthat the severely disabled readers were inferior to both good and poor readers in syntacticawareness, and in ability to use syntactic rules, while poor readers were equal to goodreaders in syntactic awareness but were relatively impaired in using syntactic knowledgeproductively.

Fluent reading involves a complex interaction ofseveral parallel processes that relate visualgraphemic stimuli to specific entries in the lexiconand combine the semantic and syntacticinformation contained in those entries toapprehend the meaning of sentences. Some ofthese processes relate to the decoding of thephonological code from print while others relate tothe assignment of meaning to the phonologicalunits. Although the decoding of the phonologicalcode can, in principle, be based solely on "bottom­up" application of grapheme-to-phonemetransformation rules, it is well documented thatthis process is supported by "top-down" streamingof lexical knowledge and contextual information.The common denominator of the "bottom-up" and"top-down" processes in reading is that both arecomponents of the human linguistic endowment(see Perfetti, 1985; Rozin & Gleitman, 1977).

This study was supported by the Israel FoundationsTrustees to Shlomo Bentin. Isabelle Liberman was supportedin part by National Institute of Child Health, and HumanDevelopment Grant HD-Ol994 to Haskins Laboratories. Theuseful comments of Anne Fowler are much appreciated. Wegratefully acknowledge the cooperation of the principals andteachers of the integrative schools "Luria" and "Stoane" inJerusalem, and of the School for Remediation Teaching inHertzelia, Israel.

180

The relative contribution of context dependent(top-down) processes to visual word recognition isdetermined by many factors among which readercompetence is particularly important. Althoughsome authors have argued that as fluencydevelops, the reader increasingly relies oncontextual information during word recognition(Smith, 1971), more recent studies havedisconfirmed this hypothesis. For example,semantic priming facilitates lexical decisions morein children than in adults and more in youngerthan in older children (Schvaneveldt, Ackerman,& Semlear, 1977; West & Stanovitch, 1978).Similarly, context effects are greater in poorreaders than in good readers both in lexicaldecision (Schwantes, Boesl, & Ritz, 1980) andnaming tasks (Perfetti, Goldman, & Hogaboam,1979; Stanovitch, West, & Feeman, 1981). Withinthe same subject, larger context effects occurwhen bottom-up processes are inhibited bydegrading or masking the target stimuli (e.g.,Becker & Killion, 1977; Massaro, Jones, Lipscomb,& Scholz, 1978; Meyer, Schvaneveldt, & Ruddy,1975). These results imply that the increased roleof higher-level contextual processes in visual wordrecognition is caused by a need to compensate fordeficiencies in lower level, decoding processes,

Syntactic Competence and Reading Ability in Children 181

such as those that occur with poor readers ordegraded stimuli (Perfetti & Roth, 1981;Stanovitch et al., 1981).

The observation that context effects are largerin poor than in good readers does not imply,however, that poor readers are better or moreefficient users of contextual information than goodreaders. In fact, the opposite may be true. BothPerfetti et al. (1979) and Schwantes et al. (1980)reported that when subjects are required to usethe context for predicting the target word beforeseeing it, skilled readers do better than poorreaders. The magnitude of semantic primingeffects in visual word recognition may therefore bea rather poor indicator of the real ability to usecontextual information. A relatively unbiasedmethod for comparing good and poor readers'awareness of contextual information would be toevaluate context effects on word recognition insituations that either eliminate the need fordecoding of print, as in auditory presentation, orthat force all readers, regardless of skill, to usecontextual processes to the same extent.

As a language in which efficient use ofcontextual information is essential for fluentreading, Hebrew provides an excellent channelthrough which to examine syntactic effects. InHebrew orthography, letters represent mostlyconsonants, while vowels are represented bydiacritical marks placed below, within, or abovethe letters. The vowel marks are usually omittedin writing except in poetry, holy scripture, andchildren's literature. Because different words maybe represented by the same consonants butdifferent vowels, when these vowels are absent upto seven, eight, or more different words may berepresented by the same string of letters. Inaddition to being semantically ambiguous, theseHebrew homographs are also phonemicallyequivocal because the (absent) vowels of the wordsthat are responsible for the different meaningsvary from word to word. Therefore, fluent readingof "unvoweled" Hebrew requires heavy reliance oncontextual information.

Reading instruction in Hebrew starts, as a rule,with the "voweled" orthographical system inwhich the diacritical marks are presented withthe consonant letters. The vowels are graduallyomitted from school texts starting at thebeginning of the third grade. During the thirdgrade, the children begin to learn to read withoutvowels. By grade four, they are expected to befluent readers of unvoweled texts. Informaldiscussions with teachers, however, revealed thatthe transition from reading voweled to reading

unvoweled material is not equally easy for allchildren. According to teachers, some children aregood readers as long as the diacritical marks arepresent but are slow in acquiring the skill ofreading without the vowels. Because withoutvowels the context of the sentence is a primarysource of phonological constraints on reading, wesuspected that the children in this group,although knowing the grapheme-to-phonemetransformation rules, do not (or can not) usecontextual information efficiently. Thus, in spiteofbeing phonologically skilled, those children maybe poor readers. Thus, Hebrew may be aconvenient medium through which to test thehypothesis that at least some deficient readers areless able than good readers to use context.

We decided to manipulate syntactic rather thansemantic context in the present research. Themain reason for this decision is that syntax isprobably a more basic linguistic ability thansemantics (see Chomsky, 1969) and less affectedby reading experience (Lasnik & Crain, 1985). Inaddition, syntactic violations are more clearlydefined than are manipulations of semanticassociation strength.

The effect of prior syntactic structure on theprocessing of a visually presented target word hasbeen investigated in a number of studies. Lexicaldecisions regarding target words are faster whenthey are preceded by syntactically appropriateprimes than when preceded by syntacticallyinappropriate primes (Goodman, McClelland, &Gibbs, 1981; Lukatela, Kostic, Feldman, &Turvey, 1983). Lexical decision (West &Stanovitch, 1986; Wright & Garret, 1984) andnaming (West & Stanovitch, 1986) are facilitatedwhen targets are syntactically congruent withpreviously presented sentence fragments relativeto when targets follow a syntactically neutralcontext.

Although the syntactic context effect on wordrecognition is reliable, information on the relationbetween this effect and reading ability iscomparatively scarce and controversial (for areview see Vellutino, 1979). Several studies reportthat poor readers are inferior to normal readers indealing with complex syntactical structures inspeech (Brittain, 1970; Bryne, 1981; Cromer &Wiener, 1966; Goldman, 1976; Guthrie, 1973;Newcomer & Magee, 1977; Vogel, 1974). Otherauthors, however, have challenged a simpleinterpretation of these results (Glass & Perna,1986). For example, Shankweiler and Crain (1986)have suggested that the poor readers' apparentdeficiency in processing complex syntactic

182 Bentin et aI.

information may be an epiphenomenon oflimitations of the working memory processor,rooted in a difficulty in generating phonologicalcodes.

The present study sought to examine therelation between word recognition and syntacticawareness; i.e., sensitivity to syntactic.structureand the ability to use syntactic knowledgeexplicitly, in children who vary in readingcompetence. Experiment I, using a group of adultfluent readers of Hebrew, was designed to

. establish the validity of the procedure of testingthe effect of syntactic context on the identificationof auditorily presented words masked by whitenoise. Experiment 2 examined syntactic contexteffects in two groups of children. One group wascomposed of chil~ren with learning disordersdrawn from a population of students who wereselected by the school system for specialsupplementary training in reading; a comparisongroup was formed of good readers drawn from thepopulation of fourth-graders of two elementaryschools. In Experiment 3, the same group of goodreaders was compared with a group of poorreaders from the same elementary schools. Thepoor readers were matched with the good readersfor their ability to apply grapheme-to-phonemetransformation rules in reading voweledpseudowords, but they were significantly inferiorin reading sentences when the words were printedwithout the diacritical vowel marks.

EXPERIMENT 1

The purpose of the present experiment was toestablish the effect of syntactic context on theidentification of auditorily presented words thatwere masked by white noise. The auditorymodality was used to attenuate the deficientreaders' excessive reliance on contextualinformation in reading (which is presumablycaused by the need to compensate for theirdifficulty in decoding the print). Stimulus maskingwas incorporated in the procedure becauseprevious studies suggest that degradationincreases the tendency of all subjects to usecontextual information for word recognition(Becker & Killion, 1977; Stanovitch & West,1981). We chose to use identification rather thanreaction time measures in order to keep themeasurement as simple and direct as possible.

Subjects were presented with a list of three- orfour-word sentences that were pre-recorded ontape. In each sentence, white noise wassuperimposed on one or several (target) words.The subjects were instructed to identify the

masked words. Half the targets in the list werecongruent with the syntactic structure of thesentence in which they appeared whereas theother targets were incongruent, that is, caused asyntactic violation. We predicted that thepercentage of correctly identified targets would behigher for syntactically congruent words than forsyntactically incongruent words.

Method

SubjectsThe subjects were 28 undergraduate students

(14 males) who participated in the experiment forcourse credit or for payment. They were all nativespeakers of Hebrew with normal hearing.

Test MaterialsThe auditory test included 104 three- or four­

word sentences. Each sentence was used in twoforms: a) syntactically correct and b) syntacticallyincorrect. The incorrect sentences wereconstructed by changing the correct sentences inone ofthe following 10 ways.

Type 1 - In this category there were 12sentences in which the gender compatibilitybetween the subject and the predicate was altered.In six of these sentences a masculine subject waspresented with a feminine predicate and in theother six a feminine subject was presented with amasculine predicate. The masked target was thepredicate, which was always the last word in thesentence.

Type 2 - In this category there were 12sentences in which the compatibility of numberbetween subject and predicate was altered. In sixof these sentences a singular predicate followed asubject in plural form, and vice-versa in the othersix. The masked target was the predicate whichwas the last word in each sentence.

Type 3 - There were eight sentences in thiscategory. The compatibility of gender and numberbetween the subject and the predicate was alteredin each sentence. The masked target was thepredicate which was the last word in eachsentence.

Type 4 - In Hebrew, prepositions related topersonal pronouns (e.g., "on me") become oneword. The violation in this category consisted ofthe decomposition of the composed pronoun intotwo separate words (the pronoun and thepreposition) which kept the meaning but weresyntactically incorrect. Eight differentprepositions were used, four times each, totaling32 sentences. The masked target words were thecomposed or decomposed pronouns. Because the

Syntactic Competence and Reading Ability in Children 183

decomposition of the pronoun and prepositionadded one word to the sentence, another word wasexcluded from each of the incorrect sentences.

The syntactical violations of types 5 to 10 werebased on changing compulsory order of parts ofthe sentence. Therefore, the sentences of thesetypes were masked from the first to the last word.

Type 5 - In the ten sentences of this type, theorder of the attribute and its nucleus wasreversed.

Type 6 - In each of the six three-word sentencesof this type, the predicate was incorrectlyintroduced between the subject and its attribute.

Type 7 - In Hebrew, the negation always comesbefore the negated predicate. We altered this fixedorder in six sentences, using three differentnegation words.

Type 8 - In six sentences, the interrogativeword was moved from its fixed place at thebeginning of the sentence to the second place.

Type 9 - In six sentences, the fixed order ofpreposition and noun was reversed so that thenoun appeared before the preposition.

Type 10 - In six sentences, the copula thatshould occur between the subject and thepredicate was moved to the beginning or the endof the sentence.

All 104 correct and 104 incorrect sentences wererecorded on tape by a female native speaker ofHebrew. The tapes were sampled at 20KHz. Themasked intervals were marked and white noisewas digitally added to the marked epochs with asignal-to-noise ratio of 1:2.75. This ratio wasdetermined on the basis of pilot tests so thatcorrect target identification level was about 50%.

The 208 sentences were organized into two lists.In each list, 52 sentences were syntacticallycorrect and 52 sentences were syntacticallyincorrect. Each sentence appeared in each list onlyonce, either in correct or incorrect form. Sentencesthat were correct in list A were incorrect in list Band vice versa. Fourteen subjects were tested withlist A and the other 14 with list B. Thus, eachsubject listened to an equal number of correct andincorrect sentences, and across subjects eachsentence appeared an equal number of times ineach form.

The sentences in each list were randomized andre-recorded on tape. Subjects listened to the tapesvia Semmheiser earphones (HD-420).

ProcedureThe subjects were tested individually in a quiet

room. The experimenter listened to the stimulisimultaneously with the subject and stopped thetape-recorder at the end of each sentence. Thesubjects were asked to repeat the masked part ofeach sentence, and were encouraged to guesswhenever necessary. The subjects' responses wererecorded manually by the experimenter. Subjectswere randomly assigned to List A or B.

Results and DiscussionThe average percentage of correct responses,

across subjects and sentence types was 41.3%.Overall, correct identification was 67.0% for thesyntactically correct sentences but only 15.6% forthe syntactically incorrect sentences. Thesyntactic context effect was evident for each typeof syntactic violation (Table 1).

Table 1. Percentage of correct identification of syntactically correct and syntactically incorrect sentences in eachviolation-type category (see text).

1 2 3Type of syntactic violation

4 5 6 7 8 9 10

Syntacticallycorrect

Mean

(SEm)

Syntacticallyincorrect

Mean

(SEm)

t1J.7

62

18.7

4.9

72.6

4.4

25.6

5.0

67.9

45

27

26

83.2

1.7

20.6

52

54.3

4.6

12.2

42

67.9

6.6

10.7

6.9

71.5

7.1

14.3

6.6

69.0

6.9

82

3.4

64.3 58.4

6.1 7.0

19.0 23.7

7.6 65

184 Bentin et al.

These observations were confirmed by two­factor analyses of variance with subjects andsentences as random variables. The factors weresyntactic context (correct, incorrect) and syntacticviolation type (Type 1 to 10). The main effect ofsyntactic context was significant (F(1,26)=588.44,MSe=629, p < .0001 for the subject-analysis andF(1,91)=140.93, MSe=624, p <. 0001 for thestimulus-analysis). The main effect of sentencetype was also significant (F(9,234)=4.5, MSe=404,p < .0001 for the subject-analysis andF(9,91)=2.21, MSe=437, p < .03 for the stimulus­analysis). The context effect was conspicuous forall syntactic violation types, but, as suggested byan interaction between the two factors, itsmagnitude differed. This interaction wassignificant for the subject analysis (F(9,234)=4.73,MSe=322, p < .0001), but only marginal for thestimulus analysis (F(9,91)=1.86, MSe=624, p <.07). Tukey-A post hoc analysis of the interactionrevealed that the context effect was greater fortype 3 (gender and number), type 4 (compositepronoun), type 8 (translocation of interrogativeword), type 6 (separation of subject and attribute),and type 7 (translocation of negation word) thanfor all the other sentence-types. Within these twogroups, the context effects were similar.

The results of Experiment 1 demonstrate thatidentification of words in sentences is influencedby the syntactic coherence of the sentence.Because, across subjects, exactly the same wordswere masked and had to be identified in thesyntactically correct and incorrect sentences, thedifference in the correct identification ratebetween the two modes of presentation is probablydue to the manipulation of syntactic coherence.The magnitude of this effect seemed to vary acrossdifferent types of syntactic anomalies but it wasreliable and statistically significant for each type.Because we had neither a priori predictions aboutthe effects of particular violation-types onidentification nor clear post hoc explanations forthe observed differences, and because the type ofviolation is not directly relevant to the issuesinvestigated in this study, the syntactic violation­types will be collapsed in all further analyses.

On the basis of these results, we can use thetechnique Of Experiment 1 to assess possibledifferences in the magnitude of the effect in goodand poor readers.

EXPERIMENT 2Experiment 2 compared the magnitude of the

syntactic context effect on the identification ofauditorily presented words in children who are

good readers and children with a severe readingdisability. As was elaborated in the introduction,although several studies reported that poorreaders are deficient in syntactic comprehension(see Vellutino, 1979), others could not find solidevidence to support this hypothesis (e.g., Glass &Perna, 1986). If disabled readers are less aware ofthe syntactic structure of the sentence (as part oftheir general linguistic handicap), or do not usesyntactic information as efficiently as goodreaders, syntactic context effects should be weakerin disabled than in good readers. Consequently,the effect of syntactic congruity on correctidentification of sentences should be smaller indisabled than in good readers.

A second prediction concerns the nature oferrors made by good and disabled readers in theidentification of words presented in syntacticallyincorrect sentences. The auditory mask probablyinduces some degree of uncertainty in theauditory input. If listeners are aware of thesentence context, they may attempt to use it tocomplement the information that is missing in theauditory stream. Such a strategy would causeerrors in the identification of words that violatethe syntactical structure because in thosesentences the target does not conform to theexpected syntactic rules. Therefore, errors inidentification that are induced by syntacticawareness should reflect the use of correctsyntactic forms. In an English example, if thesentence were "I would like to have many child"and the word "child" was masked, the subject mayerroneously identify the target as "children." Onthe other hand, if the subjects are not aware of thesyntactic structure or not bothered by itsviolations, their errors in the identification ofmasked words should not be related to thesyntactically correct form of the target. In thiscase, the response may be a randomly selectedword, or may relate to the acoustical form - forexample, substituting "mild" for the target "child"in the above example. If good readers are moreaware of the syntactic structure of the sentencethan disabled readers, the percentage of errors ofthe first type-"syntactic corrections," and of thesecond type-"random errors"-should vary withreading ability. In an extreme case, we should findmore "syntactic correction" errors than "random"in good readers and vice versa in disabled readers.

Analyses of the errors made by each readinggroup would be a first step towards understandingthe cause of inter-group differences in syntacticcontext effects, if they exist. However, in order toassess syntactic awareness as a metalinguistic

Syntactic Competence and Reading Ability in Children 185

ability rather than automatic use of syntacticstructures for word identification, a more directmeasure had to be employed. In a recent study,Fowler (1988) compared good and poor readers'ability to detect and to correct violations of syntaxin orally presented sentences. In that study, theability to judge sentences as correct or incorrect inthe "judgment" task was not associated withreading ability. In contrast, good readersperformed syntactically better than poor readersin the "correction" task. Fowler concluded thatpoor readers do not differ from good readers insyntactic knowledge but that they may be inferiorin manipulating verbal material in short-termmemory (see also Shankweiler & Crain, 1986). Weused Fowler's technique to supplement our studyof syntactic context effects on the identification oforally presented words. If, as in Fowler's study, adifference emerges only for the correctioncondition, then syntactic awareness is not at fault.Rather, one would ascribe the differences tosyntactic processing difficulties that prevent thedisabled readers from using their syntacticknowledge productively.

Method

Tests and Materials

A. Reading Tests. We were interested intesting two kinds of reading: the ability to decodethe phonology from print and the ability to use thesentence context in reading without vowels.Because all the standard reading tests in Hebrewprimarily test reading comprehension, weconstructed two new reading tests for ourpurposes. The first was a test of decoding ability.It contained a set of 24 meaningless three- or four­letter strings (pseudowords) presented with vowelmarks. The vowels were chosen according toHebrew morphophonemic rules, and included alllawful combinations. Each pseudoword wasprinted individually on a white, 9 cm X 12 cmcardboard. The size of each letter was 0.5 em. Thesubject was instructed to read each pseudowordexactly as it was written. The accuracy andnaming onset time were measured. The subject'sscore on this test consisted of the percentage ofaccurately read pseudowords and the meanlatency of naming onset time.

The second test was designed to test the abilityto read Hebrew without vowel marks andparticularly to use the sentence context todetermine the reading of unvoweled Hebrewwords that were both phonologically andsemantically ambiguous. This test contained 48

four- or five-word sentences printed on whitecardbord using the same fonts as for thepseudowords in the former test. The last word ineach sentence was the target word. In the absenceof vowel marks, 32 out of the 48 targets werephonologically ambiguous, Le., they could havebeen assigned at least two sets of vowels to formtwo different words. Thus, correct reading of thosetargets could be determined only by apprehendingthe meaning of the sentence. The 32 ambiguoustargets were 16 pairs of identical letter stringseach representing a different word in therespective sentence. Eight of these 16 ambiguoustargets represented two words of equal frequency.The words represented by each of the remainingambiguous words differed in frequency such thatone member of each pair was a high-frequencyword while the other member was a low-frequencyword. In a previous study Bentin and Frost (1987)reported that when undergraduates werepresented with isolated ambiguous words in anaming task, they tended to choose the mostfrequent phonological alternative. We assumedthat, without context, the children would tend tochoose the same. Therefore, insensitivity to thecontext of the sentence should increase thenumber of errors in reading the targets,particularly when the correct response requiresthe use of the less frequent phonologicalalternative. The remaining 16 targets were wordsthat without the vowel marks could have beenmeaningfully read in only one manner. Eight ofthose sixteen targets were high"frequency wordsand the other eight targets were low-frequencywords. The subjects were instructed to read eachsentence aloud. The time that elapsed from themoment the sentence was exposed until thesubject finished reading it was measured to thenearest millisecond. The score on this test was theaverage percentage of errors and the average timeto read a sentence.

In addition to these two special purpose tests,each subject was tested for reading comprehensionby a standard test. The nation-wide average scoreon this test for fourth graders is 70% with a SD of12%.

B. Intelligence tests. The IQ of each subjectwas obtained either using the WISC (Full Scale)(whenever those data were available) or testingthe children on the Raven Colored Matrices andtransforming their performance into IQ scores.

C. Syntactic awareness test. Syntacticawareness was assessed by testing identificationof auditorily presented words masked by whitenoise as in Experiment 1. On the basis of a pilot

186 Bentin et al.

GoodReaders 8.1% 1.6 sec 4.1% 2.9 sec 813%

DisabledReaders 38.8% 2.6 sec 19.6% 6.3 sec 55.7%

Table 2. Reading performance of the severely disabledand good readers.

They were within the normal intelligence range(Mean IQ (FS)=104.83, ranging between 85 and130). The disabled readers selected for the presentstudy were chosen because they not only showedpoor decoding ability, as compared to goodreaders, but also performed badly in the test ofreading without vowels, thus suggesting specialproblems in dealing with context. Table 2 presentsthe reading performance of the good and deficientreaders as revealed by our reading tests.

Readingunvoweled ReadingSentences Comprehension

% of Time per %errors sentence correct

Readingvoweled

Nonwords% of Timepererrors item

Children in both groups were all nativespeakers of Hebrew without known motor,sensory, or emotional disorders. All children hadbeen tested for nonnal hearing.

ResultsThe overall percentage of correct identification

of masked targets was similar in good (44.2%) anddisabled readers (48.3%) (F(I,32)=2.52, MSe=112,p > .12). However, the percentages of syntacticallycorrect and incorrect sentences that were correctlyidentified were different in the two groups (Table3). Although syntactically correct sentences wereidentified better than syntactically incorrectsentences in both groups, the effect of thesyntactic context was smaller in the disabledreaders than in the good readers.

This observation was supported by a mixed­model two-factors analysis of variance. Thebetween-subjects factor was reading ability, andthe within-subjects factor was syntactic context(correct, incorrect). The syntactic context effectwas highly significant across groups(F(I,32)=784.47, MSe=50, p < .0001). A moreinteresting result, however, was the significantinteraction that revealed that the syntacticcontext effect was greater in good readers than indisabled readers (F(I,32)=11.90, MSe=50, p <.002).

study with children, in order to keep the overallcorrect identification of targets around 50%, weincreased the signal-to-noise ratio in Experiment2 from 1:2.75 to 1:2.25.

ProcedureEach child was tested individually in three

sessions. During the first session, readingperfonnance and intelligence were tested. Readingperfonnance was recorded on tape for subsequenterror analysis and off-line measuring of time. Atthe end of the test of reading without vowels, theexperimenter verified whether the subject knewthe meaning of the targets that had been readincorrectly. In the very few doubtful cases, thesentence was excluded and a substitute sentenceof the same type was given. The children who hadbeen selected for this study (see below) wereinvited to a second session during which theauditory word identification test was given. Theprocedures for the word identification test wereidentical to those of Experiment 1. In addition, atthe end of the second session, the children weretested for the ability to repeat from memory thesentences presented during the auditory test. Thiswas done by presenting the children with 16sentences selected from the same pool of sentencesfrom which the test set was selected. Eight ofthese 16 sentences were syntactically correct andthe other eight syntactically incorrect. In therepetition test the sentences were presentedwithout the masking noise. Finally, during a thirdsession (three months later), all 104 sentenceswere presented to each child without the maskingnoise. Following the presentation of each sentencethe child was asked whether "this is the way itshould be said in Hebrew" (Judgment Task).Whenever the answer was "no," the child wasasked to correct the sentence (Correction Task).

SubjectsThe good readers were 15 children (7 males)

selected from a population of fourth graders of twoelementary schools in Jerusalem. Their agesranged between 8.9 and 9.7 years (mean age 9.3years). The average IQ (FS) score (as assessed bytransforming the Raven score) was 102.5, rangingfrom 85 to 122.5. They were selected to matchpoor readers from the same school on decodingability and IQ. The precise selection criteria willbe elaborated in Experiment 3.

The disabled readers were 19 children (12males), aged from 9.7 to 14 years, (mean age 11.6),selected from a population of 32 children withsevere reading disorders who had been referredfor special supplementary training in reading.

Syntactic Competence and Reading Ability in Children 187

Table 4. Percentage of errors of each type (out of thetotal number of responses) made by the disabled andgood readers in the auditory-identification task.

Table 3. Percentage of correct identification ofsyntactically correct and incorrect sentences in thereading disabled and good readers.

Post hoc analysis revealed that good anddisabled readers performed equally well withcorrect sentences. However, the percentage ofcorrect identifications of words embedded inincorrect sentences was higher in disabled than ingood readers (p < .01).

The errors that children made were distributedamong four error types: Type 1 errors were"syntactical corrections," that is, errors that weremade in attempt to use the correct syntacticstructure of a syntactically incorrect sentence.Type 2 errors were "random errors"­misidentifications that made no sense whatsoeveror reflected acoustical confusions. Type 3 errorswere "logical substitutions," that is, substitutionsof the masked words with other words that gavethe sentence a logical meaning. Type 4 were "1don't know" responses, which were not encouragedbut were accepted. The percentage of errors ofeach type (out of the total number of responses) ineach group is presented in Table 4.

GoodReaders 13 5.4(SEm) 0.4 0.9

ReadingDisabled 69 32.1(SEm) 19 45

Older readingDisabled 0.3 5.6

Task"Judgment" "Correction"

Table 5. Percentage of errors made by disabled andgood readers in the judgment and correction tasks.

Because we had clear predictions only forsyntactic corrections and random errors, weanalyzed the distribution of these two error typesin each group by a mixed-model (reading group Xerror type) analysis of variance. This analysisshowed that, across the two types of error, thegood readers made fewer errors than the disabledreaders (F(1,32)=5.11, MSe=18, p < .031). Acrossgroups, the percentage of errors of each type wassimilar (F(1,32)=3.01, MSe=16, p > .09). Mostinteresting, the interaction between readingability and error type was highly significant(F(1,32)=21.54, MSe=16, p < .0001). Post hocanalysis (Tukey-A) revealed that more randomerrors were made by disabled readers than bygood readers, whereas syntactic corrections weremore frequent in good than in disabled readers.All the children were able to repeat verbatim allsixteen sentences that they heard without themasking noise.

Good readers were better than disabled readerson both the judgment and the correction tests.Among the disabled readers, however, a secondarydistinction was evident between four children whowere 13-14 years old and those who were younger.The mean percentage of errors made by eachgroup in each task is presented in Table 5.

27.022

6951.9

DisabledReaders

17.024

71.422

GoodReaders

SyntacticallyIncorrect

Mean(SEm)

SyntacticallyCorrectMean(SEm)

Type of errorLogical "I don't

Corrections Random Substitutions know"

ReadingDisabled 3.9 10.2 21.9 15.7(SEm) 0.4 15 1.7 22

GoodReaders 62 33 17.4 28.8(SEm) 0.8 0.7 1.7 28

Because the number of older disabled readerswas too small to form a reliable independent levelin a factorial design but, on the other hand,clearly formed a distinct group, they wereexcluded from the statistical evaluation. Thus, thepercentage of errors in each task was comparedonly for good and disabled readers who were moresimilar in chronological age. As before, a mixed­model analysis of variance was employed wherethe between-subjects factor was reading groupand the within-subject factor was the test

188 Bentin et al.

Gudgment or correction). The analysis of varianceshowed that good readers made significantly fewererrors than disabled readers (F(1,24)=26.58,MSe=127, p < .0001) and that more errors weremade in the correction than in the judgment task(F(1,24)=79.25, MSe=35, p < .0001). A significantinteraction suggested that the task affected thepercentage of errors made by disabled readersmore than it affected the good readers(F(l,24)=41.0, MSe=35, p < .0001). Thisinteraction supports Fowler's results byemphasizing the difference between the judgmentand correction tests. However, in contrast to herresults, post hoc Tukey-A tests revealed that thegood readers made significantly fewer errors thandisabled readers not only in the correction task,but also in the judgment task. The inclusion of thefour older disabled readers in the analysis did notchange the pattern of results, although these fourchildren clearly performed better than the otherdisabled readers.

DiscussionFor children with a severe reading disability,

the syntactic context effect on the identification ofspoken words was smaller than for good readers.One explanation for the results might be thatdisabled readers are worse at identifyingauditorily masked words than good readers(Brady, Shankweiler, & Mann, 1983). Such anhypothesis, however, is not supported by thepresent data. If the disabled readers in thepresent study had been handicapped in theidentification of masked words, any manipulationthat increased the difficulty of identifying thewords should have had a greater effect on disabledthan on good readers. Therefore, we should haveobserved a stronger rather than a weakersyntactic context effect in disabled readers.Further, if masking had a more deleterious effecton identification of words by disabled relative togood readers, the overall identificationperformance in the disabled group should havebeen lower. In fact, the overall correctidentification percentage in disabled readers wasslightly higher than in the good reader group.

A second account of the results might be thatthe smaller syntactic context effect in poor readersreflects a more general problem, such as disordersof short-term or working memory. There is indeedample evidence that disabled readers haveproblems with verbal short-term memory (Mann,Liberman, & Shankweiler, 1980; for a review seeBrady, 1986). Therefore, memory disorders mightexplain why their performance is affected by

sentence context less than that of good readerseven when decoding difficulties are eliminated;they simply do not remember the sentence wellenough. However, a simply reduced short-termmemory span cannot· easily account for thepresent results because the children in bothreading groups could accurately repeat sentencessimilar to those used in the identification taskwithout any difficulty. It is still possible, however,that more complex working memory problemscould have contributed to the disabled readers'pattern of performance. We will return to thishypothesis in the General Discussion.

We are left with the most direct hypothesis thatinferior syntactic awareness is the reason for therelatively poor use of syntactic context by thereading disabled children of Experiment 2. Thispossibility is supported by the results of the erroranalysis. The percentage of "syntactic correction"errors made by good readers was almost twice asgreat as that made by disabled readers. "Syntacticcorrection" errors could have only been madewhen the subject knew what the correct structureof the sentence should have been and expected it.In those circumstances, when the physicalstimulus was degraded the good readers appliedsyntactic rules and misidentified the target. In thesame situation, getting only partial informationfrom degraded stimuli, disabled readers oftenapplied a random guessing strategy disregardingthe sentence context completely. Indeed, thepercentage of "random" errors was three timesgreater in the disabled readers group than in goodreaders.

An additional question examined in the presentexperiment was whether the disabled readers hadmastered the correct syntactic structures but didnot use them properly, or had problems with basicsyntactic knowledge. We examined this questionby testing the ability of both groups to detectviolations of syntactic structure and to correct thedetected violations. The good readers performedbetter than the disabled readers in both tasks.Although the difference between the groups wasgreater for the correction task, disabled readerswere significantly inferior to good readers in thejudgment task as well. This latter resultcontradicts the results reported by Fowler (1988)and suggests that this group of disabled readerswere inferior to good readers in their awareness ofbasic syntactic structures.

The discrepancy between the present resultsand Fowler's (1988) results, as well as thedisagreement between our conclusion regardingthe syntactic awareness of disabled readers and

Syntactic Competence and Reading Ability in Children 189

previous assertions in the literature that basicphonological disability and deficient use ofworking memory mechanisms underlies thesyntactic inferiority observed in poor readers (e.g.,Shankweiler & Crain, 1986; Shankweiler et al., inpress), can be explained in two ways. One possibleexplanation is that different types of mechanismsunderly reading deficiencies in differentlanguages. Recall that we selected our deficientreader group to emphasize problems of usingcontext while reading without vowel marks. Indoing so, we may have selected a group of childrenwho were poor in syntactic processing. A secondexplanation is that we have examined childrenwith a reading disability that was considerablymore severe than that of the poor readersexamined by Fowler. Experiment 3 thereforeattempted to generalize the results of the presentexperiment to poor readers selected, as in Fowler'sstudy, from the normal student population ofregular elementary schools.

EXPERIMENT 3Any attempt to generalize about the

characteristics of reading disability or to predictthe performance of children with readingdisorders is impeded by the heterogeneity of thispopulation. Indeed, reading disorders can appearas the most conspicuous symptom in children whosuffer from attentional disorders or generallearning disability; they can be the main symptom(but rarely the only symptom) of developmentaldyslexia and, at the other extreme, they maycharacterize the performance of otherwise normalstudents who happen to be at the lower end of anormal distribution of reading ability. It ispossible, therefore, that the prior selection ofdifferent types of reading disorders underlies mostdisagreements about this important handicapamong educators and scientific investigators.

In Experiment 2, the reading disabled childrenwere selected from a population of children withsevere reading disorders. Although they were atleast in the fourth grade, had normal IQ's, andhad no documented neurological symptoms, someof those children could hardly read single wordswith or without vowel marks. We found that theywere inferior to good readers in syntacticknowledge and in using syntactic context to helpidentify spoken words. In Experiment 3, our aimwas to extend these findings to another readergroup. Thus, we compared good readers withchildren in regular classes who, when formallytested, were inferior to good readers in reading

performance. In particular, we wished to comparethe good readers with a group of relatively poorreaders who were equal to the good readers inbasic decoding ability (as revealed by theirperformance on reading voweled pseudowords) butwere poorer at reading without vowel marks. Weassumed that the relatively poor readingperformance of this group primarily reflectsinefficient use of the sentence context, andexpected to be able to measure this relativedisability by our auditory test of syntactic contexteffects. In addition, the good and poor readers inthe present experiment were tested for ability todetect and to correct syntactic violations.

MethodSubjects

The subjects were 30 children selected from apopulation of 167 fourth graders in two publicelementary schools. The selection was based onperformance on the test of decoding ability andthe test of reading without vowels which weredescribed in Experiment 2. Two reading groupswere assembled. The poor reader group included15 children (9 males); their ages ranged between8.8 and 9.6 years (mean age 9.1 years). Theiraverage IQ (FS) score (as assessed bytransforming the Raven score) was 102.5, rangingfrom 85 to 122.5. Each of the those poor readersmade no more than four errors (16.6%) in the testof decoding voweled pseudowords but at leasttwice as many errors as the good readers whilereading meaningful sentences without vowels. Onthe basis of the assumption that the relativelypoor reading performance of those childrenreflected problems with processing of contextualinformation, we will label this group the "Poorcontext" group. The good readers were the same15 children who were described in Experiment 2.Each child in this group was selected to match onechild in the poor context group on the ability todecode voweled pseudowords and on IQ. However,the good readers performance on the unvoweledsentences was at least twice as good as that of hisor her matched subject in the poor context group.The average scores of the two reading groups onthe reading tests are presented in Table 6.

Tests and MaterialsThe reading tests and the auditory word

identification test were identical to thosedescribed in Experiment 2. The IQ scores wereestimated by testing the children with the RavenColored Matrices test.

190 Bentin et al.

Table 6. Reading performance of the good and poorcontext readers.

Table 7. Percentage of correct identification ofsyntactically correct and incorrect sentences in goodand poor context readers.

ResultsThe percentage of total correct identifications in

the "poor context" group (40.4%) was notsignificantly different from that of the goodreaders (44.2%) (F(l,28)=1.03, MSe=209, p > .31).

The percentages of syntactically correct andincorrect sentences that were correctly identifiedin each group are presented in Table 7.

ProcedureThe procedure was similar to that employed in

Experiment 2. The children were tested in twosessions. The first session was dedicated to theselection of subjects for this study. All 167 fourth­graders were tested for reading ability and IQ.During the second session only the selectedchildren were tested on the auditory wordidentification test, and during a third session,their ability to detect and correct the syntacticviolations. During the third session the sentenceswere presented without any masking noise.Sessions one and two were held close to thebeginning of the academic year. Session three wasthree months latter.

24.83.5

28.82.8

24.52.9

17.41.7

6.91.2

3.30.7

3.50.6

6.20.8

Poor ContextReadersMean(SEm)

Good ReadersMean(SEm)

Our a priori predictions concerned only errors ofType 1 (correction) and Type 2 (random errors). Amixed-model analysis of variance showed nosignifican t main effects but a significantinteraction between the type of error and readinggroup (F(1,28)=10.63, MSe=14, p < .003). Post hoccomparisons (Tukey-A) showed that thepercentage of syntactical correction errors washigher in good readers than in the poor contextgroup, whereas the percentage of random errorswas higher in the poor context group than in goodreaders.

The average percentages of errors in thejudgment and correction tasks for each group arepresented in Table 9.

A mixed-model analysis of variance as inExperiment 2 was used to analyze these data.Across groups, there were more errors in thecorrection task than in the judgment test(F(1,24)=39.16, MSe=16, p < .0001). Overall, thegood readers made fewer errors than poor contextreaders (F(1,24)=5.24, MSe=25, p < .035). The

Type of errorLogical "I don't

Corrections Random Substitutions know"

Table 8. Percentage of errors of each type made bygood and poor context readers (out of the total numberofresponses) in the auditory identification task.

These data were analyzed by a mixed-modelanalysis of variance as in Experiment 2. As before,the syntactic context was highly significant(F(1,28)=505.56, MSe=81, p < .0001). However, incontrast to the findings of Experiment 2, theinteraction between the syntactic context effectand reading group was not significant(F(1,28)=0.96).

As in Experiment 2, the errors made by eachgroup were categorized into four types. Type 1were "syntactical corrections," Type 2 were"random errors," Type 3 were "logicalsubstitutions," and Type 4 were "I don't know."The distribution of errors in each of the tworeading groups (out of the total number ofresponses) is presented in Table 8.

72.0%,

813%

6533.9

15.43.6

Poor ContextReaders

Readingunvoweled ReadingSentences Comprehension

% of Time per %errors sentence correct

71.422

17.02.4

GoodReaders

Readingvoweled

Nonwords% of Timepererrors item

SyntacticallyCorrectMean(SEm)

SyntacticallyIncorrect

Mean(SEm)

PoorContext 7.8% 1.7 sec 18.2% 4.7 secReaders

GoodReaders 8.1% 1.6 sec 4.1% 2.9 sec

Syntactic Competence and Reading Ability in Children 191

Table 9. Percentage of errors made by good and poorcontext readers in the judgment and correction tasks.

interaction between the test and the group factorswas significant (F(1,24)=6.32, MSe=16, p < .020).Replicating Fowler's (1988) results, post hocanalysis revealed that good and poor contextreaders did not differ in the judgment test,whereas in the correction test good readers madesignificantly fewer errors than poor contextreaders.

Task"Judgment" "Correction"

All children were able to repeat verbatim all the16 sentences presented to them in absence ofmasking white noise.

DiscussionExperiment 3 sought to generalize the results of

Experiment 2 to groups of relatively poor readersselected from the normal distribution of fourthgraders. Unlike in Experiment 2, the magnitude ofthe syntactic context effect was similar in the goodand in the poor context readers.

The syntactic ability of the children in the poorcontext group was not, however, entirelyequivalent to that of the good readers. Analysis ofthe identification errors made by each grouprevealed that the proportion of errors thatreflected an attempt to correct the syntacticviolation was lower in children who had relativelymore difficulties in reading unvoweled words thanin good readers who were matched with them forphonological decoding ability. In contrast, the

.proportion of misidentifications that reflectedtotal ignorance of the sentence's context (eithersyntactic or semantic) was lower in good readersthan in the poor context group. This pattern oferrors might suggest that although wordidentification was similarly affected by syntacticcontext in both reading groups, the good readerswere more aware of the syntactic structure of thesentence than were the children in the poorcontext group. The results of the judgment test,

GoodReaders

Mean(SEm)

Poor ContextReaders

Mean(SEm)

130.4

1.70.5

5A0.9

11.42.3

however, did not support this hypothesis. As itturned out, both groups were equally sensitive toviolations of syntactic structures. It is possiblethough, that part of this result reflected a ceilingeffect in that task. The groups differed, however,

. in their ability to correct those violations.The common aspect of both the "syntactical

correction" errors and the test of correctingsyntactic violations is that both measures reflectthe child's ability to actively generate correctsyntactic structures. This ability is not requiredby the judgment test and may not be reflected inidentification performance. Therefore, the presentdata suggest that although the good and therelatively poorer readers did not differ in theirsyntactic awareness-that is, in the sensitivity toand knowledge of basic syntactic structures-thegood readers had a superior ability to use theirsyntactic knowledge, and a tendency to do so.

GENERAL DISCUSSIONIn the present study we examined the relation

between reading ability and syntactic competenceas it is reflected in the ability to use syntacticcontext for word identification and to detect andcorrect syntactic violations. In contrast to thegreat majority of studies of context effects in goodand poor readers, we used auditory rather thanprinted word identification. Auditory presentationwas used to circumvent a bias that might havebeen induced by the reading disorder itself. Thus,we were better able to assess differences insyntactic processing ability that might relate toreading achievement. The sensitivity of ourauditory test to syntactic context was verified byshowing that undergraduate students, fluentreaders of Hebrew, identified target words maskedby white noise significantly more accurately if thetargets were syntactically congruent with thesentence in which they appeared than if theyviolated the syntactic structure.

A syntactic effect similar to that found inundergraduates was obtained when the same testwas given to fourth graders. However, thedifference between the correct identification ofsyntactically correct and syntactically incorrectsentences was smaller in a group of children witha severe reading disability than in either goodreaders or relatively poor readers selected fromthe normal distribution of fourth-grade students(the poor context group). The good readers and thepoor context group did not differ in the auditoryidentification test.

A second difference between the severelyreading disabled and the children in the poor

192 Bentin et al.

context group was observed in the judgment task.Children in the poor context· group detectedsentences that contained an error as well as goodreaders. In contrast, the reading disabled wereworse in this test than either the good readers orthe poor context readers.

The relative inferiority of the severely disabledreaders can not be accounted for only by a simplereduction of their short-term memory span. Incontrast to the complex sentences and complexsyntactic structures typically used in otherstudies, we used only very short and simplesentences (three or four words). When formallytested, all the children were able to repeat thesentences verbatim without any problem. Holdinga sentence in working memory for syntacticanalysis probably requires more mental effort andretention of the whole sentence for a longer timethan required by immediate repetition. As waspreviously reported, the factor of delay influencesthe memory ability of poor readers more than thatof good readers (Liberman, Shankweiler,Liberman, Fowler, & Fisher, 1977). However,rather than requiring the manipulation of moresubtle syntactic aspects, the syntactic violationswhich we have used in the present study were, aswe have said, straightforward corruptions of thebasic syntactic relationship between subject andpredicate or a word order that clearly violated thesyntactic structure of the sentence. Therefore, weagree with Byrne (1981) in doubting that deficientuse of verbal memory mechanisms by disabledreaders, at least .as this deficiency could berevealed by simple repetition, was a major causefor the deficient use of syntactic context in thepresent study. Instead, we are inclined to believethat the reduced syntactic ability suggested by theperformance of disabled readers reflected agenuine deficiency of linguistic endowment (in thesyntactic and phonological domains) rather thanreduced general cognitive ability or poormetalinguistic insight.

Although the syntactic context effect on theidentification task was equal in the good readersand the poor context readers, the syntacticcompetence of these two groups was not entirelyequivalent. In particular, the good readers madesignificantly more syntactical correction errorsthan the poor context readers. The differencebetween the two groups was even moreconspicuous in the correction task. Similar to theresults reported by Fowler (1988) for Americanpoor readers, the ability of poor context children tocorrect syntactic violations was significantlyinferior to that of good readers. This result is

particularly interesting because, as was notedearlier, the syntactic violations used in thepresent study were much simpler and more directthan those used by Fowler. Moreover, our samplesof good and poor context readers were matched fortheir ability to decode and read voweled nonwords.Therefore, just as for the disabled readers, thedifference between the ability of good and poorcontext readers to correct syntactically incorrectsentences cannot be easily accounted for only byassuming differences between the poor and thegood readers in general cognitive skills. Rather,we suggest that, at least for the specificallyselected group of poor readers whose readingerrors reflected reduced ability to analyzecontextual information, both the correction testand the pattern of errors in the identification testsuggest a specific impairment in the ability to usetheir syntactic knowledge in a productive way.

Although both the reading disabled and the poorcontext readers are inferior to good readers insyntactic competence, these two groups differ fromone another. In comparison to good readers, thedisabled readers showed a weaker syntacticcontext effect in the word identification task, aninferior ability to detect syntactical aberrations inspoken sentences, and an inferior ability to correctdetected syntactically incorrect sentences. Thepoor context children were equal to good readersin the syntactic context effect on wordidentification, were equally able to detectsyntactic aberrations, but were inferior to goodreaders in the ability to correct the detectederrors. This pattern of results suggest that thedifferent tasks tap different aspects of syntacticcompetence which might develop at differentrates.

Some insight into the nature of the syntactic·disability reflected by the word identification taskcomes from the observation that the significant

. interaction between the reading group and thesyntactic context effect was not caused by asymmetrical effect of reading group on bothsyntactically correct and incorrect sentences.Rather, it seems that syntactically correctsentences were identified equally well by bothreading disabled and good readers; however, goodreaders were affected more than reading disabledby violations of the syntactic structure ofsentences. A possible interpretation that issupported by these data is that automaticsyntactic processing was equivalent in bothgroups, but that the disabled readers were lessaware of the syntactic structure and did not useidentification strategies that were based upon it.

Syntactic Competence and Reading Ability in Children 193

A more definite interpretation obviously requiresa neutral condition in the identification task,which was absent in this study. However, theseresults strongly suggest that the identificationtest is sensitive more to strategic differences andsyntactic awareness than to (automatic) syntacticprocessing.

It is noteworthy to recall in this connection thefour older disabled readers: Although they weresimilar to other disabled readers in the auditoryidentification task, their performance in thejudgment and correction tasks was similar to thatof good readers. These older children may havehad a higher level of syntactic competence so thatwhen their attention was intentionally directed tothe structure of the sentence (as was the case inthe judgment and correction tasks in contrast tothe identification task), the additional knowledgeenabled them to use their syntactic knowledgeproductively.

In conclusion, the results of the present studysuggest that syntactic factors are directly relatedto reading disabilities, at least in Hebrew. Twodistinct populations of poor readers have beenidentified. One group was formed of children whoin absence of a better term were labeled readingdisabled. These children were probably able to usebasic syntactic structures, as was evident in theireveryday speech ability and in their identificationof syntactically correct sentences. However, theywere not explicitly aware of the syntacticstructures, and therefore were not inhibited bysemantic incongruity in the identification test;they were less able than good readers to detectsyntactically incorrect sentences, and they wereless able to correct those errors that had beendetected. The second group of poor readers weregood decoders but were relatively weak inanalyzing the context of the sentence in reading.The performance of these children in theidentification judgment tests suggested that theywere aware ofbasic syntactic structures and coulduse them for perception of speech. However, theywere inferior to good readers in using thosestructures productively as suggested by theirrelatively worse performance in the correctiontest. Thus, our data set limits to previousassertions that poor reading is not related tosyntactical impairment (Gleitman & Rozin, 1977;Liberman, 1971; Mattingly, 1972; Shankweiler &Crain, 1986).

Of course, we do not claim to have found acausal relationship between syntactic ability andreading disorders. What we have seen is that, atleast in Hebrew, there are poor readers of normal

intelligence who are good decoders. Theirperformance suggests that there are aspects ofpoor reading that are not accounted for bydeficient phonological processing. Moreover, wehave shown that this impairment is associatedwith deficiencies in linguistic ability, hereexemplified in the syntactic domain.

REFERENCESBecker, C. A., &r Killion, T. H. (1977). Interaction of visual and

cognitive effects in word recognition. Journal of ExperimentalPsychology: Human Perception and Performance, 3,389-401.

Bentin, S., &r Frost, R. (1987). Processing lexical ambiguity andvisual word recognition in a deep orthography. Memory &Cognition, 15, 13-24.

Brady, S. A. (1986). Short-term memory, processing and readingability. Annals of Dyslexia, 36, 138-153.

Brady, S. A., Shankweiler, D., &r Mann, V. A. (1983). Speechperception and memory coding in relation to reading ability.Journal of Experimental Child Psychology, 35, 345-367.

Brittain, M. M. (1970). Inflectional performance and early readingachievement. Reading Research Quarterly, 6, 34-48.

Byrne, B. (1981). Deficient syntactic control in poor readers: Is aweak phonetic memory code responsible? AppliedPsycholinguistics, 2, 201-212.

Chomsky, N. (1969). Acquisition of syntax in children from 5 to 10.Cambridge, MA: MIT Press.

Cromer, W., &r Wiener, M. (1966). Idiosyncratic response patternsamong good and poor readers. Journal of Consulting Psychology,30,1-10.

Fowler, A. E. (1988). Grammaticality judgment and reading skillin grade 2. Annals of Dyslexia, 38, 73-94.

Glass, A. L., &r Perna, J. (1986). The role of syntax in readingdisability. Journal of Learning Disabilities, 19, 354-359.

Gleitman, L. R., &r Rozin, P. (1977). The structure and acquisitionof reading: Relation between orthography and the structuredlanguage. In A. S. Reber &r D .L. Scarborough (Eds.), Towards aPsychology of Reading: The Proceedings of the CUNY Conference(pp. 1-54). Hillsdale, NJ: Lawrence Erlbaum Associates.

Goldman, S. R. (1976). Reading skill and the minimum distanceprinciple. A comparison of listening and readingcomprehension. Journal of Experimental Child Psychology, 22,123­142.

Goodman, G. 0., McOelland, J. L., &r Gibbs, R. W. (1981). The roleof syntactic context in word recognition. Memory & Cognition 9,580-586.

Guthrie, J. T. (1973). Reading comprehension and syntacticresponse in good and poor readers. Journal of EducationalPsychology, 65, 294-299.

Lasnik, H., &r Crain 5, (1985). On the acquisition of pronominalreference. Lingua, 65, 135-154.

Uberman, I. Y. (1971). Basic research in speech and lateralizationof language: Some implications for reading disability. Bulletinof the Orton Society, 21, 71-87.

Liberman, I. Y., Shankweiler, D., Liberman, A. M., Fowler, c., &rFisher F. W. (1977). Phonetic segmentation and recording in thebeginning readers. In A. S. Reber &r D. L. Scarborough (Eds.),Towards a Psychology of Reading: The Proceedings of the CUNYConference (pp. 207-225). Hillsdale, NJ: Erlbaum Associates.

Lukatela, G., Kostic, A., Feldman, L. B, &r Turvey, M. (1983).Grammatical priming of inflected nouns. Memory & Cognition,11, 59-63.

Mann, V. A, Uberman, I. Y., &r Shankweiler, D. (1980). O1ildren'smemory for sentences and word strings in relation to readingability. Memory & Cognition, 8,329-335.

Massaro, D. W., Jones, R. D., Lipscomb, c., &r Scholz, R. (1978).Role of prior knowledge on naming and lexical decisions withgood and poor stimulus information. Journal of Experimental

.Psychology: Human Learning and Memory, 4,498-512.Mattingly, I. G. (1972). Reading, the lingUistic process, and

linguistic awareness. In J. F. Kavanagh &r I. G. Mattingly (Eds.),

194 Bentin et al.

Language by ear and by eye (pp. 133-148). Cambridge, MA: MITPress.

Meyer, D. E" Schvaneveldt, R. W., & Ruddy, M. G. (1975). Loci ofcontextual effects on word recognition. In P. M. A. Rabbitt & S.Domic (Eds.), Attention and performance V (pp. 98-118). NewYork: Academic Press.

Newcomer, P., & Magee, P. (1977). The performance of learning(reading) disabled children on a test of spoken language. TheReading Teacher, 30, 896-900.

Perfetti, C. A (1985). Reading ability. New York: Oxford UniversityPress.

Perfetti, C. A., Goldman, S., & Hogaboam, T. (1979). Reading skilland the identification of words in discourse context. Memory &Cognition, 7, 273-282.

Perfetti, C. A, & Roth, S. (1981). Some of the interactive processesin reading and their role in reading skill. In A. Lesgold & C. A.Perfetti (Eds.), Interacti'CIe processes in reading (pp. 269-298).Hillsdale, NJ: Lawrence Erlbaum Associates.

Rozin, P., & Gleitman, L. (1977). The structure and acquisition ofreading II: The reading process and the acquisition of thealphabetic principle. In A. S. Reber & D. L. Scarborough (Eds.),Towards a Psychology of Reading: The Proceedings of the CUNYConference (pp. 55-142). Hillsdale, NJ: Erlbaum Associates.

Schvaneveldt, R. W., Ackerman, B., & Semlear, T. (1977). Theeffect of semantic context on children's word recognition. ChildDe'CIelopment, 48, 612-616.

Schwantes, F. M., Boesl, S. L., & Ritz, E. G. (1980). Children's useof context in word recognition: A psycholinguistic guessinggame. Child De'CIelopment, 51, 730-736.

Shankweiler, D., & Crain S. (1986). Language mechanisms andreading disorders: A modular approach. Cognition, 24, 139-168.

Smith, F. (1971). Understanding reading. New York: Holt, Rinehart,and Winston.

Stanovitch, K. E., & West, R. F. (1981). The effect of sentencecontext on ongOing word recognition: Tests of a two processtheory. Journal of Experimental Psychology: Human Perception andPerformance,7,658-672.

Stanovitch, K. E., West, R. F., & Feeman, D. J. (1981). Alongitudinal study of sentence context effects in second-gradechildren: Tests of an interactive compensatory model. Journal ofExperimental Child Psychology, 32,185-199.

Vellutino, F. R. (1979). Dyslexia: Theory and research. Cambridge,MA: MIT Press.

Vogel, S. A (1974). Syntactic abilities in normal and dyslexicchildren. Journal of Learning Disabilities, 7,103-109.

West, R. F., & Stanovitch, K. E . (1978). Automatic contextualfacilitation in readers of three ages. Child De'CIelopment, 49, 717­727.

West, R. F., & Stanovitch, K. E. (1986). Robust effects of syntacticstructure on visual word processing. Memory & Cognition, 14,104-112.

Wright, B., & Garret, M. (1984). Lexical decision in sentences:Effects of syntactic structure. Memory & Cognition, 12,31- 45.

FOOTNOTES• Journal of Experimental Child Psychology, in press.tAlso Department of Psychology and School of Education, The

Hebrew University, Jerusalem, Israel.ttAlso Department of Education, University of Connecticut,

Storrs.