Context-sensitive rules and word naming in Italian children

Abstract The present study examines the role of orthographic complexity onItalian children’s word reading. Two experiments are reported in which ele-mentary school children (3rd and 5th graders) read aloud words containingsimple or contextual letter-sound conversion rules. In Experiment 1, bothgroups of participants read words containing contextual rules more slowly andless accurately than words containing simple, non-contextual rules. Experiment2 showed that the effect of rule complexity held for low but not high frequencywords, on both reading speed and accuracy. No interactions with grade werefound. This pattern is in line with previous findings on the effects of rule con-textuality on adult performance [Burani, C. Barca, L. & Ellis, A. W. (2006).Psychonomic Bulletin & Review, 13, 346–352]. Despite the regularity of theItalian orthography, the presence of complex rules influences both reading speedand accuracy of young readers. Moreover, the reading system of readers of ashallow orthography seems efficient from the first years of reading instruction.

Keywords Italian language Æ Young readers Æ Orthographic complexity ÆWord reading aloud

Introduction

Alphabetic writing systems represent information about the pronunciationand sound of words. In English, words vary in the degree of consistency

L. Barca (&) Æ A. W. EllisDepartment of Psychology, University of York, Heslington, York YO10 5DD, UKe-mail: [email protected]

L. Barca Æ C. BuraniInstitute for Cognitive Sciences and Technologies, ISTC-CNR, Rome, Italy

123

Read Writ (2007) 20:495–509DOI 10.1007/s11145-006-9040-z

ORI GI N A L P A PE R

Context-sensitive rules and word naming in Italianchildren

Laura Barca Æ Andrew W. Ellis Æ Cristina Burani

Received: 2 January 2006 / Accepted: 18 October 2006 / Published online: 28 November 2006� Springer Science+Business Media B.V. 2006

between orthographic transcription and pronunciation. Words with regular,consistent spelling-sound correspondences, such as dog, are read faster thanwords with irregular, inconsistent spelling-sound correspondences, such aspint. This so-called regularity (or consistency) effect has been investigated inmany studies researching the role of spelling-sound transparency in visualword recognition. Several studies have reported an interaction between reg-ularity effects and word frequency on the reading performance of adult skilledreaders (e.g., Andrews, 1982; Monaghan & Ellis, 2002; Seidenberg, Waters,Barnes, & Tanenhaus, 1984; Waters, Seidenberg, & Bruck, 1984) and typicallydeveloping children (Masterson, Laxon, & Stuart, 1992; Mayall, 2002), with asmall effect of regularity for high frequency words but faster naming of low-frequency regular words than of low-frequency irregular words.

Languages with a more transparent (or shallow) orthography, such asItalian, have less opportunity to show regularity effects because they havequite predictable spelling-to-sound correspondences. The majority of Italianletters have only one pronunciation with a one-to-one mapping betweengrapheme and phoneme. Words like domani (tomorrow) and tamburo (drum)could be read unerringly by a letter-sound conversion procedure whichoperated one letter at a time in a strictly serial, left-to-right fashion. Thepronunciation of the letters c and g, however, is determined by context-sen-sitive rules and depends upon the other letters that follow. The letter c ispronounced as /k/ when it is followed by any letter apart from e or i, in whichcases is pronounced /tS/. Thus, the letter c in the words clima [climate] andcasa [house] is pronounced /k/, whereas in the word cinque [five] is pro-nounced as a soft /tS/. When c is followed by h, the combination ch constitutesone grapheme which is pronounced /k/ (e.g., chiave [key]). The letters ci alsoconstitute one grapheme (pronounced /tS/), but only when followed by an-other vowel (e.g., cielo [sky]). The pronunciations corresponding to the dif-ferent sequences in which the letter g is involved show similar variation (for adetailed description of context-sensitive graphemes see Burani et al., 2006,Table 1; Lepschy & Lepschy, 1991).

The presence of complex graphemes has been reported to affect theprocessing of English non-word reading (i.e., the ‘whammy effect’, Rastle& Coltheart, 1998) and, recently, of word reading in experiments done onFrench (Rey & Schiller, 2005) and Italian (Barca, Burani, Di Filippo, &Zoccolotti, 2006; Burani et al., 2006). Burani et al. (2006) reported thatcollege students are faster at reading aloud words containing only simplegrapheme to phoneme (G-P) conversion rules than words containingcontext-sensitive G-P rules. Notably, the effects of grapheme contextualitywere confined to low frequency words. These results were interpreted inthe framework of the dual-route model of visual word recognition (Colt-heart, Curtis, Atkins, & Haller, 1993; Coltheart, Rastle, Perry, & Ziegler,2001). Such a model assumes lexical and non-lexical mechanisms convert-ing print to sound. The lexical route uses a parallel activation process; itoperates by accessing a word’s representation in an orthographic inputlexicon followed by the retrieval of the word’s spoken form from a pho-

496 Read Writ (2007) 20:495–509

123

nological output lexicon. The time required for these representations toachieve full activation is a function of word frequency. In contrast, thenon-lexical route assembles a pronunciation for a letter string by applyinga set of G-P correspondence rules. This assembly occurs serially, one lettera time, from left to right. The time taken for the non-lexical translation isnot affected by word frequency but by other variables such as word length.Regular words are read equally well by both routes; irregular words areonly read correctly by the lexical route, and non-words can only be readby the non-lexical route.

The dual-route interpretation of the contextuality by frequency interac-tion follows from the idea that the speed with which the lexical procedureoperates depends on word frequency. The lexical processing of high fre-quency words is sufficiently rapid that the lexical procedure produces thecorrect pronunciations for both words with simple graphemes and wordswith context-sensitive graphemes before the non-lexical route can computethem. The lexical processing of low-frequency words, however, is sufficientlyslow to allow the effects of the non-lexical processing to show. Processing aletter sequence that incorporates context-sensitive graphemes provokes acompetition between phonemes that have been activated by the contextualgraphemes. The competition among phonemes slows the rise of activationfor the correct phoneme until the subsequent context has been processed,and produces a processing cost relative to simple graphemes, even for skilledreaders.

A number of studies have reported that the rate of development ofdecoding skills is greater in languages with a transparent orthographic sys-tem, as demonstrated by greater accuracy in word and non-word reading inchildren learning a transparent orthography than in children learning moreopaque orthography (Cossu, Shankweiler, Lieberman, & Gugliotta, 1995;Goswami, Gombert, & De Barrera, 1998; Landerl, 2000; Seymour, Aro,& Erskine, 2003; Thorstad, 1991; Wimmer & Goswami, 1994). More re-cently, the grain size theory proposed by Ziegler and Goswami (2005) placesspecial emphasis on the development and use of different grain sizes andrecoding strategies across orthographies: learning to read a regular orthog-raphy relies to a greater extent on smaller grain size units such as the G-Psize unit.

Some authors have suggested that Italian children use a systematic pho-nological strategy when learning to read, with direct, orthographic access tothe lexicon developing much later than in languages with a deeper orthog-raphy such as English (e.g., Paulesu et al., 2000; Thorstad, 1991). Theassumption is that if orthography is highly transparent, the G-P correspon-dences should be easier to detect and use. In less transparent orthographies,on the other hand, it may be more adaptive initially to learn spelling pat-terns for individual words and then to use various strategies such as analogyto read new words. Further support for this view comes from Orsolini, Fa-nari, Tosi, De Nigris, and Carrieri’s (2005) longitudinal study, which tracesthe development of reading abilities of Italian children, highlighting the

Read Writ (2007) 20:495–509 497

123

transition from an initial phase based on phonological recoding to a lexicalphase characterized by the analysis of the word as a whole. Nevertheless,lexical effects (e.g., word frequency) have been reported to affect the per-formance of children with average reading ability since the first years ofreading instruction (Burani, Marcolini, & Stella, 2002; Mazzotta, Barca,Marcolini, Stella, & Burani, 2005). The standard interpretation of wordfrequency effects would imply that lexical procedures are therefore operat-ing alongside sublexical decoding procedures, even in beginning readers ofItalian.

In Experiment 1 of the present paper we tested for an effect of rulecontextuality in young Italian readers in the third and fifth year ofschooling. We expected that, similarly to adult performance (Burani et al.,2006), children from both grades would be affected by rules contextualityshowing a latency cost in reading words made up of context-sensitivegraphemes relative to words made up of simple graphemes. Moreover, if inthe early phases of learning to read Italian the highly transparent orthog-raphy encourages word recognition via G-P conversion, then youngerreaders (3rd graders) should have more difficulty in reading words withcontext-sensitive graphemes than older readers (5th graders). Less skilledreaders’ prevalent use of the non-lexical procedure would be evidenced byslower naming latencies and more errors in reading words with context-sensitive graphemes than words with simple graphemes. More skilledreaders, on the other hand, should be less affected by rules contextuality asthey rely more on direct reading. If, on the other hand, beginning readersdevelop efficient lexical as well as sublexical procedures, there may be nodifferences between age groups in the nature and magnitude of the rulecontextuality effects.

Experiment 1

Method

Participants

Sixty-seven Italian-speaking elementary school children participated in thisstudy. They were all natives of the area in and around Lecce, with normalreading abilities. Reading level was examined with a standard readingachievement test (Sartori, Job, & Tressoldi, 1985). In this test, children arerequired to read aloud 4 lists of 28 words each (2 lists of high frequency wordsand 2 lists of low frequency words), and speed and correctness are scored.Participants in this study scored within 1 standard deviation from the mean ofthe grade of reference in both speed and correctness (Tressoldi, 1996). Thefinal sample included 34 children from 3rd grade (mean age 8 years, half maleand half female) and 33 children from 5th grade (mean age 10 years, half maleand half female).

498 Read Writ (2007) 20:495–509

123

Materials

Two sets of 28 morphologically simple Italian nouns were used (see Buraniet al., 2006).1 Words in the complex rules set contained the letters c and g, whichrequire the following context to be taken into account if they are to bepronounced correctly. In contrast, words in the simple rules set were composedexclusively of letters than can be translated into sounds on a simple one letter/one phoneme basis. The words were 6–9 letters long, and were taken from a low-to-medium frequency range on both adults and child written frequency measure(lexvar database, Barca, Burani, & Arduino, 2002, http://www.istc.cnr.it/mate-rial/database). The word sets were matched on initial phoneme characteristics(voicing and manner), position of the lexical stress, number of double letters,2

and on several lexical and sublexical variables (such as imageability,orthographic neighborhood size and bigram frequency). Means on the match-ing variables for the four lists of words are shown in Table 1.

Apparatus and procedure

A reading aloud task was used. Children were tested individually in a quietroom in the school and were asked to read aloud, as quickly and as accuratelyas possible, a series of words that would appear on the computer screen infront of them. A voice key connected to the computer, within the SuperLab

Table 1 Mean values on the matching variables for simple and complex rules words

Simple rules Complex rules

Word frequency Mean 52.68 59.79SD 76.00 141.03

Imageability Mean 4.78 5.39SD 1.26 1.19

Orthographic neighborhood size (N) Mean 0.06 0.06SD 0.13 0.13

Bigram frequency Mean 10.81 10.79SD 0.33 0.23

Letter length Mean 7.50 7.82SD 0.84 0.90

Note: Frequency is on 1 million tokens; imageability is given as 7-point subjective rating; bigramfrequency values are transformed on the basis of the natural logarithm. Values, drawn from Barcaet al. (2002), are available at http://www.istc.cnr.it/material/database

1 Burani, Barca, and Ellis (2006). The materials of the experiments may be accessed through thePsychonomic Society’s Norms, Stimuli, and Data Archive, http://www.psychonomic.org/archive.To access the file, search the archive for this article using the journal title (Psychonomic Bulletin &Review), and the first author’s name (Burani).2 In Italian, all consonants except ‘h’ can be doubled. Consonant geminates (double letters) havea lengthening effect on pronunciation of the corresponding phoneme (i.e., the phoneme is pro-nounced much more forcefully than a single consonant). In the absence of any empirical evidencethat consonant geminates affect vocal reaction times, we decided to balance their presence acrossthe word sets.

Read Writ (2007) 20:495–509 499

123

Pro 2.0 experimental control shell, collected reaction times from word pre-sentation to the beginning of each vocalization (onset). The stimuli werepresented in upper case preceded by a fixation point (500 ms). The stimulistayed on the screen for a maximum of 3000 ms and disappeared as soon asthe participant made a vocal response. No feedback was provided. There wasa 1200 ms interval before presentation of the next sequence (fixation point,stimulus, etc.). The 56 experimental items were combined with 44 filler wordsinvolving simple letter-sound correspondences (included in the list to avoid ahigh density of context-sensitive graphemes). The list was presented to eachchild in a different random order, divided into two blocks. The order of blockpresentation was randomized across children. There was a pause after eachblock. The experiment was preceded by a brief practice session in order tofamiliarize participants with the task.

Results and discussion

Individual reaction times of less than 250 ms (0.3% of total data points in 3rdgrade and 0.2% in 5th grade) were treated as missing data. Pronunciationerrors and false responses (e.g., invalid trials due to external noise) wereexcluded from the analyses on reaction times and accounted for, respectively,19% and 3.4% of total data points in 3rd grade and 11.2% and 1.5% in 5thgrade. Vocal reaction times (RT) for correctly named items and percent errorscores are presented in Table 2.

Two-way ANOVAs with Grade (3rd versus 5th) as a between-groups factorand contextuality (contextual rules versus non-contextual rules) as a repeatedmeasures factor were carried out on mean naming latencies for correct re-sponses and number of errors (F1). The generalizability over items was alsoexamined with items as cases (F2).

Naming latencies

The analyses on RT revealed significant main effects of grade (F1 (1,65) = 24.9,MSE = 88876.1, p < .001; F2 (1,58) = 352.8, MSE = 6660.7, p < .001) and con-textuality (F1 (1,65) = 94.7, MSE = 96.7, p < .001; F2 (1,58) = 11.2,MSE = 41679.6, p < .001), with slower responses for 3rd than for 5th gradechildren and for words with contextual graphemes than for words with simplegraphemes. The grade by contextuality interaction was not significant (bothFs < 1).

Table 2 Mean vocal reaction times (in ms) and percentage errors (Experiment 1)

Simple rules Complex rules

3rd RT 1104 1223%ERR 15 22

5th RT 847 965%ERR 8 14

500 Read Writ (2007) 20:495–509

123

Accuracy

The analyses on accuracy mirrored those on RT. Significant main effects ofgrade (F1 (1,65) = 17.7, MSE = 9.1, p < .001; F2 (1,58) = 21.5, MSE = 9.3,p < .001) and rule contextuality (F1 (1,65) = 30.4, MSE = 4.7, p < .001;F2 (1,58) = 4.7, MSE = 34.1, p < .05) were observed, reflecting greater accuracyin older readers and in words with simple graphemes. The grade by contex-tuality interaction was not significant (both Fs < 1).

The effect of contextuality on both vocal reaction times and accuracy, withthe lack of a grade · contextuality interaction, indicate that children at allreading ages are sensitive to the degree of transparency between graphemesand phonemes, being faster and more accurate in reading words with simplegraphemes than words with context-sensitive graphemes.

In adult readers of Italian, the contextuality effects are particularly strongfor low frequency words, where the non-lexical procedure is presumed to bemore involved in the reading process (Burani et al., 2006). In the nextexperiment the contextuality by frequency interaction was investigated inchildren’s reading.

Experiment 2

In this experiment two groups of children who were all proficient readers readwords that varied factorially on rule contextuality and frequency. An effect offrequency (better performance in reading high than low frequency words) hasbeen reported in Italian children’s reading (Burani et al., 2002; Mazzottaet al., 2005). If beginning readers of Italian read using a G-P conversionprocedure with lexical procedure being developed later, then the effect ofword frequency should be stronger in older than in younger children, and theinteraction between frequency and rule complexity should be stronger in olderchildren.

Method

Participants

Participants were the same as in the first experiment, a sample of 67 children(34 from 3rd grade and 33 from 5th grade).

Materials

The materials of Burani et al.’s (2006) Experiment 2 were used. The experi-mental condition resulted from a factorial combination of word frequency(high versus low) and contextuality (contextual graphemes versus simplegraphemes). Four sets of 24 morphologically simple nouns each were used.The word sets were matched on several lexical and sublexical variables,

Read Writ (2007) 20:495–509 501

123

namely imageability, orthographic neighborhood size, mean bigram fre-quency, length in letters and initial phoneme, based on Barca et al.’s (2002)database. All words had regular stress on the penultimate syllable. Means onthe matching variables for the four lists of words are shown in Table 3.

Apparatus and procedure

A standard reading aloud experiment was used. The 96 experimental itemswere presented in random order in a single block. No fillers were included inthe list because the words with contextual graphemes had a lower density ofcomplex letter clusters than those in Experiment 1. Other details of theapparatus and procedure were the same as in Experiment 1.

Results and discussion

Individual RTs of less than 250 ms (1.3% of 3rd grade children’s responsesand 1.4% of 5th grade children’s responses) were treated as missing data.Pronunciation errors and false responses were excluded form the analyses of

Table 3 Mean values on matching variables for simple and complex rules words of high or lowfrequency

High frequency Low frequency

Simplerules

Complexrules

Simplerules

Complexrules

Word frequency Mean 314.00 305.80 10.10 13.00SD 298.60 346.00 10.66 9.95

Imageability Mean 5.22 5.25 5.29 4.83SD 1.08 1.13 0.85 1.07

Orthographicneighborhood size (N)

Mean 1.08 1.21 0.75 0.92SD 1.10 1.79 1.29 1.28

Bigram frequency Mean 10.82 10.75 10.67 10.78SD 0.43 0.32 0.46 0.39

Letter length Mean 5.71 5.92 6.08 6.17SD 1.20 1.32 1.25 1.20

Note: Frequency is on 1 million tokens; imageability is given as a 7-point subjective rating; bigramfrequency values are transformed on the basis of the natural logarithm. The values are drawn fromBarca et al.’s (2002) database, available at http://www.istc.cnr.it/material/database

Table 4 Mean vocal reactiontimes (in ms) and percentageerrors (Experiment 2)

Note: HF, high frequencywords; LF, low frequencywords

3rd 5th

Simplerules

Complexrules

Simplerules

Complexrules

HF RT 918 944 741 779%ERR 5 5 1 2

LF RT 1019 1094 796 847%ERR 13 17 5 7

502 Read Writ (2007) 20:495–509

123

reaction times, accounting for 10% and 2.1% of total data points in 3rd gradeand 5% and 1.7% in 5th grade, respectively. Vocal RTs for correctly nameditems and percent error scores are presented in Table 4.

Three-way ANOVAs with grade (3rd versus 5th) as between-groups factor,and contextuality (contextual rules versus non-contextual rules) and fre-quency (high frequency words versus low frequency words) as repeatedmeasures factors, were carried out on mean naming latencies for correctresponses and number of errors (F1). Generalizability over items was alsoexamined with items as cases (F2).

Naming latency

The latency data exhibited significant effects of grade (F1 (1,65) = 29.1,MSE = 94768.7, p < .001; F2 (1,92) = 386.4, MSE = 5113.8, p < .001), frequency(F1 (1,65) = 104.8, MSE = 5646.9, p < .001; F2 (1,92) = 18.1, MSE = 23392.0,p < .001) and contextuality (F1 (1,65) = 22.9, MSE = 2859.6, p < .001;F2 (1,92) = 4.6, MSE = 23392.0, p < .05). This pattern of main effects indicatesthat 3rd graders were slower than 5th grade children, that high frequency wordswere read faster than low frequency words, and that words with contextualgraphemes were read slower than words with simple graphemes. The grade byfrequency interaction was significant (F1 (1,65) = 6.8, MSE = 5646.7, p < .05;F2 (1,92) = 9.7, MSE = 5113.8, p < .001). A posteriori comparisons (Duncantest) indicated that the frequency effect was reliable in both groups (bothp < .001), though inspection of Fig. 1 indicates a larger frequency effect in 3rdgraders.

The frequency by contextuality interaction was significant in the subject-based analysis (F1 (1,65) = 6.6, MSE = 3542.5, p < .05; F2 < 1):3 a difference

600

700

800

900

1000

1100

3rd grade 5th gradeR

eact

ion

times

High frequency

Low frequency

Fig. 1 Frequency by gradeinteraction on naminglatencies (ms)

3 The lack of significance of the frequency by contextuality interaction in the analysis by itemsmay have two sources. First, the overall power of the contextuality factor might be attenuated inExperiment 2 due to the 2 · 2 factorial manipulation, which resulted in a list composed of a smallproportion of words that included more than one complex sequence relative to Experiment 1.Second, in experiments with factorial designs (in which items are not sampled completelyrandomly) the item analysis may be too conservative and bias against a significant result(Raaijmakers, 2003).

Read Writ (2007) 20:495–509 503

123

between simple and complex items was present for low (p < .001) but not forhigh frequency words (see Fig. 2). No other interactions were significant.

Accuracy

The error data mirrored the latency data with significant effects of grade(F1 (1,65) = 20.0, MSE = 5.8, p < .001; F2 (1,92) = 386.4, MSE = 5113.8,p < .001), frequency (F1 (1,65) = 84.3, MSE = 2.7, p < .001; F2 (1,92) = 18.1,MSE = 23392.0, p < .001) and contextuality (F1 (1,65) = 5.3, MSE = 1.7,p < .05; F2 (1,92) = 4.6, MSE = 23392.0, p < .05). The grade by frequencyinteraction was significant (F1 (1,65) = 8.0, MSE = 2.6, p < .001; F2 (1,92) = 9.7,MSE = 5113.8, p < .001), and the Duncan test revealed that the frequencyeffect was reliable for both groups (p < .001). However, Fig. 3 shows that thefrequency effect is larger in 3rd graders.

The frequency by contextuality interaction was significant by subjects only(F1 (1,65) = 4.5, MSE = 1.4, p < .05; F2 < 1), with contextuality effect restrictedto low frequency items (p < .001) (see also Fig. 4). As with the latency data,the group by contextuality interaction failed to reach significance (both F1 andF2 < 1).

Analysis of errors

In this analysis the reading errors made by 3rd and 5th graders in Experiments1 and 2 were combined. The reading errors of the two grades were very

600

700

800

900

1000

1100

High frequency Low frequencyR

eact

ion

times

Simple rules

Complex rules

Fig. 2 Frequency bycontextuality interaction onnaming latencies (ms)

0

5

10

15

20

25

3rd grade 5th grade

% e

rror

s

High frequency

Low frequency

Fig. 3 Frequency by gradeinteraction on percentage oferrors

504 Read Writ (2007) 20:495–509

123

similar. The majority of the pronunciation errors could be classified asorthographic errors (e.g., single letter insertions, substitutions or omissions),which preserved at least 50% of the stimulus letters in the response. Theseerrors, usually resulting in a non-word (e.g., mondo fi monda), occurredequally often in 3rd and 5th grades (52% and 57%, respectively). Less com-mon were stress errors (failure in the assignment of the correct lexical stress,25% in 3rd grade and 18% in 5th grade), again resulting in a non-word (e.g.,pantera fi pantera). Only a small percentage of errors included a failure inapplying the correct grapheme-to-phoneme correspondence rule for a contextsensitive grapheme (5% and 7% in 3rd and 5th grade, respectively). Theseerrors also resulted in non-words (e.g., unghia fi undZia). Hence, asreported for other transparent orthographies (see Ellis et al., 2004), the greatmajority of reading errors were non-word mispronunciations. Whole-wordsubstitutions accounted for only 18% of total errors in both grades.

General discussion

In this paper the effects of orthographic complexity on reading perfor-mances of Italian children proficient readers were investigated. Two groupsof elementary school children (3rd graders and 5th graders), took part intwo reading aloud experiments. In Experiment 1 children read wordscontaining context-sensitive graphemes more slowly and less accurately thanwords with a one-to-one mapping between graphemes and phonemes.Younger children were slower and less accurate than older children but theeffect of rule contextuality was remarkably similar in the two groups interms of both reading time and accuracy. When rule contextuality wasinvestigated in combination with word frequency, its effect was restricted tolow frequency words for both grades (Experiment 2). This pattern of resultsis very similar to the one for Italian adults, although in adults the con-textuality effect and its interaction with word frequency were restricted tothe latency data (Burani et al., 2006) because pronunciation errors werevirtually non-existent.

Despite the regularity of the Italian orthography, then, the application ofcomplex orthographic rules (which involve context-sensitive graphemes) slows

0

5

10

15

20

25

High frequency Low frequency

% e

rror

s

Simple rules

Complex rules

Fig. 4 Frequency bycontextuality interaction onpercentage of errors

Read Writ (2007) 20:495–509 505

123

down the reading speed of skilled and younger readers as well as the readingaccuracy of younger readers, but the size of the effect on reading speed doesnot appear to change with age.

According to the dual-route model (Coltheart et al., 2001, Rastle &Coltheart, 1998), the locus of the graphemic complexity effect is locatedwithin the non-lexical route. The assignment of the correct pronunciation toa context-sensitive grapheme by the non-lexical route suffers a latency costas a result of the initial activation of two or more corresponding phonemes.These spurious phoneme activations slow the rise of activation for thecorrect phoneme until the following context is processed. The presence ofcontextual graphemes is not relevant for high frequency words becausetheir reading is accomplished mainly through a whole word reading pro-cedure (namely, the lexical procedure) and the phonological representationcan be quickly accessed through lexical retrieval. In contrast, the repre-sentations of low frequency words are accessed less rapidly in the lexicalroute, and their processing is slowed by the competing phonemes activatedin the non-lexical route by context-sensitive graphemes. Is it also possiblethat the present findings could be accounted for with other computationalmodels such as parallel distributed processing (PDP) models of reading. Ininteractive PDP models (e.g., Harm & Seidenberg, 2001), orthographicrepresentations are not parsed into graphemes, and the phonological outputis computed directly from orthographic input. The contextuality effectswould arise because of competition between multiple orthography-to-pho-nology mappings. The present results do not arbitrate between thesealternative accounts.

One widespread view is that children learn to read using low-level spelling-to-sound correspondence rules and only later in development make use ofmappings involving larger orthographic units (e.g., Brown & Deavers, 1999;Coltheart & Leahy, 1992; Ehri, 1992). Brown and Deavers (1999) reportedthat less skilled English readers (mean age of 8 years), though able to makeuse of rime-based spelling-to-sound correspondences, predominantly usedsimple grapheme-to-phoneme correspondences in reading isolated unfamiliaritems whereas skilled readers (mean age of 11 years) were more likely toadopt an analogy strategy. Consistent with this view, we hypothesized that ouryounger readers to be more sensitive than older readers to the contextuality ofthe rules. However, our results do not support this hypothesis as applied toItalian: 8-years-old children were slower and less accurate than 10-years-oldchildren, but they were similarly affected by frequency and rule complexity.The reading system of young Italian readers seems to be efficient from the firstyears of reading instruction (see also Burani et al., 2002; Mazzotta et al.,2005).

Word frequency effects were found in both grades, with high frequencywords named faster (and more accurately) than low frequency words. Con-trary to our hypothesis, 3rd graders showed greater, not smaller, frequencyeffects. However, it may be that the apparently greater effect among youngerreaders is a reflection of their slower overall RTs (i.e., due to an over-addi-

506 Read Writ (2007) 20:495–509

123

tivity effect; see Faust, Balota, Spieler, & Ferraro, 1999). An alternativeexplanation is that for younger children, some of the words have a lowerfrequency (including the possibility that a few were for them non-words) thanfor 5th graders. The similarities between the performances of children in thetwo grades are confirmed by the high correlation between their readinglatencies. An item-based analysis was carried out to evaluate response con-sistency in the two groups of children: the reaction times of 3rd and 5thgraders were closely correlated (Pearson’s r = .82, p < .001). The reactiontimes of the children were also correlated with the reaction times of the adultreaders4 (r = .37 and .48, for 3rd and 5th graders, respectively), suggesting thatwords that are easier for 3rd graders are also easier to read for 5th graders aswell as for expert readers.

In sum, Italian elementary school children are sensitive to the contex-tuality of the rules in reading words containing complex (context-sensitive)letter-sound correspondences more slowly than words containing simpleletter-sound correspondences. The effects of rule contextuality were re-stricted to low-frequency words. The absence of any interactions withgrade suggests that Italian children in grade 3 have already developed anefficient reading system, including lexical processing of more familiarwords.

Acknowledgements We wish to thank Simone Mazzotta for help with collecting the data. Thiswork was supported by MIUR-FIRB Grant RBAU01LE9P to Cristina Burani. The authors aremembers of the Marie Curie Research and Training Network: Language and Brain (RTN:LAB)funded by the European Commission (MRTN-CT-2004-512141) as part of its Sixth FrameworkProgramme.

References

Andrews, S. (1982). Phonological recoding: Is the regularity effect consistent? Memory &Cognition, 10, 565–575.

Barca, L., Burani, C., & Arduino, L. S. (2002). Word naming times and psycholinguistic norms forItalian nouns. Behavior Research Methods, Instruments & Computers, 34, 424–434.

Barca, L., Burani, C., Di Filippo, G., & Zoccolotti, P. (2006). Italian developmental dyslexic andproficient readers: Where are the differences? Brain and Language, 98, 347–351.

Brown, G. D. A., & Deavers, R. P. (1999). Units of analysis in nonword reading: Evidence fromchildren and adults. Journal of Experimental Child Psychology, 73, 208–242.

Burani, C., Barca, L., & Ellis, A. W. (2006). Orthographic complexity and word naming in Italian:Some words are more transparent than others. Psychonomic Bulletin & Review, 13, 346–352.

Burani, C., Marcolini, S., & Stella, G. (2002). How early does morpho-lexical reading develop inreaders of a shallow orthography? Brain and Language, 81, 568–586.

Coltheart, M., Curtis, B., Atkins, P., & Haller, M. (1993). Models of reading aloud: Dual-route andparallel-distributed-processing approaches. Psychological Review, 100, 589–608.

Coltheart, M., Rastle, K., Perry, C., & Ziegler, J. (2001). DRC: A dual route cascaded model ofvisual word recognition and reading aloud. Psychological Review, 108, 204–256.

Coltheart, V., & Leahy, J. (1992). Children’s and adults’ reading of nonwords: Effects of regularityand consistency. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18,718–729.

4 Data taken from Burani et al. (2006).

Read Writ (2007) 20:495–509 507

123

Cossu, G., Shankweiler, D., Liberman, I. Y., & Gugliotta, M. (1995). Visual and phonologicaldeterminants of misreading in a transparent orthography. Reading and Writing, 7, 237–256.

Ehri, L. C. (1992). Reconceptualizing the development of sight word reading and its relationshipto recoding. In P. B. Gough, L. C. Ehri, & R. Treiman (Eds.), Reading acquisition (pp. 107–143). Hillsdale, NJ: Lawrence Erlbaum Associates Inc.

Ellis, N. C., Natsume, M., Stavropoulou, K., Hoxhallari, L., van Daal, V. H. P., Polyzoe, N., Tsipa,M., & Petalas, M. (2004). The effects of orthographic depth on learning to read alphabetic,syllabic, and logographic scripts. Reading Research Quarterly, 39, 438–468.

Faust, M. E., Balota, D. A., Spieler, D. H., & Ferraro, F. R. (1999). Individual differences ininformation-processing rate and amount: Implications for group differences in response la-tency. Psychological Bulletin, 125, 777–799.

Goswami, U., Gombert, J. E., & De Barrera, L. F. (1998). Children’s orthographic representationsand linguistic transparency: Nonsense word reading in English, French, and Spanish. AppliedPsycholinguistics, 19, 19–52.

Harm, M. W., & Seidenberg, M. (2001). Are there orthographic impairments in phonologicaldyslexia? Cognitive Neuropsychology, 18, 71–92.

Landerl, K. (2000). Influences of orthographic consistency and reading instruction on the devel-opment of nonword reading skills. European Journal of Psychology and Education, 15, 239–257.

Lepschy, A. L., & Lepschy, G. C. (1991). The Italian language today. London: Routledge.Masterson, J., Laxon, V., & Stuart, M. (1992). Beginning reading with phonology. British Journal

of Psychology, 83, 1–12.Mayall, K. (2002). Case-mixing effects on children’s word recognition: Lexical feedback and

development. The Quarterly Journal of Experimental Psychology, 55A, 525–542.Mazzotta S., Barca, L., Marcolini, S., Stella, G., & Burani, C. (2005). Frequenza, immaginabilita ed

eta di acquisizione delle parole: in che misura queste variabili influenzano la lettura deibambini italiani? [How do frequency, imageability and word age-of-acquisition affect Italianchildren’s reading aloud?]. Psicologia Clinica dello Sviluppo, 9, 249–268.

Monaghan, J., & Ellis, A. W. (2002). What, exactly, interacts with spelling-sound consistency inword naming? Journal of Experimental Psychology: Learning, Memory, Cognition, 28, 183–206.

Orsolini, M., Fanari, R., Tosi, V., De Nigris, & Carrieri, R. (2005). From phonological recoding tolexical reading: A longitudinal study on reading development In Italian. Language andCognitive Processes (available online).

Paulesu, E., McCrory, E., Fazio, F., Menoncello, L., Brunswick, N., Cappa, S. F., Cotelli, M.,Cossu, G., Corte, F., Lorusso, M., Pesenti, S., Gallagher, A., Perani, D., Price, C., Frith, C. D.,& Frith, U. (2000). A cultural effect on brain function. Nature Neuroscience, 3, 91–96.

Raaijmakers, J. G. W. (2003). A further look at the ‘‘Language-as-fixed-effect fallacy’’. CanadianJournal of Experimental Psychology, 57, 141–151.

Rastle, K., & Coltheart, M. (1998). Whammies and double whammies: The length effect onnonword reading. Psychonomic Bulletin & Review, 5, 277–282.

Rey, A., & Schiller, N. O. (2005). Graphemic complexity and multiple print-to-sound associationsin visual word recognition. Memory and Cognition, 33, 76–85.

Sartori, G., Job, R., & Tressoldi, P. E. (1985). Batteria per la valutazione della dislessia e disor-tografia evolutiva [Battery for the diagnosis of developmental dyslexia and dysgraphia].Firenze: Organization Special.

Seidenberg, M. S., Waters, G. S., Barnes, M. A., & Tanenhaus, M. K. (1984). When does irregularspelling or pronunciation influence word recognition? Journal of Verbal Learning and VerbalBehaviour, 23, 383–404.

Seymour, P. H. K., Aro, M., & Erskine, J. M. (2003). Foundation literacy acquisition in Europeanorthographies. British Journal of Psychology, 94, 143–174.

Thorstad, G. (1991). The effect of orthography on the acquisition of literacy skills. British Journalof Psychology, 82, 527–537.

Tressoldi, P. E. (1996). L’evoluzione della lettura e della scrittura dalla II elementare alla IIImedia [The development of reading and writing from grade 2 to year 8]. Eta Evolutiva, 53, 43–55.

508 Read Writ (2007) 20:495–509

123

Waters, G. S., Seidenberg, M. S., & Bruck, M. (1984). Children’s and adults’ use of spelling-soundinformation in three reading tasks. Memory & Cognition, 12, 293–305.

Wimmer, H., & Goswami, U. (1994). The influence of orthographic consistency onreading development: Word recognition in English and German children. Cognition,51, 91–103.

Ziegler, J. C., & Goswami, U. C. (2005). Reading acquisition, developmental dyslexia and skilledreading across languages: A psycholinguistic grain size theory. Psychological Bulletin, 131,3–29.

Read Writ (2007) 20:495–509 509

123