Portuguese spelling in primary grades: complexity, length and lexicality effects

This study investigates spelling abilities of 189 second, third, and fourth graders using a word and pseudoword dictation task in European Portuguese. We analyzed the effect of orthographic complexity on spelling accuracy and the moderating role of length (two vs. three syllables), lexicality (words vs. pseudowords), and grade (second, third, and fourth). Each item represented one of the following orthographic complexity categories: digraph, contextual consistency, position consistency, consonant cluster, stress mark, inconsistency, and silent letter ‹h›. Digraphs and position consistencies reached high levels of accuracy already in Grade 2, but stress marks, inconsistencies, and the silent letter ‹h› were not yet fully mastered by the end of primary school. Performance across complexities was more discrepant in Grade 2 than in Grades 3 and 4. Moreover, within each complexity, there were larger differences between Grades 2 and 3 than between Grades 3 and 4. Words were better spelled than pseudowords in position consistency and stress mark categories, and a shorter length improved accuracy only in consonant clusters and stress marks. These findings underline the importance of applying learning and teaching strategies in early education adapted to the properties of the writing system to be learned.


3
Portuguese spelling in primary grades: complexity, length… compared spelling skills of fifth graders with dyslexia to those of typical readers from Grade 5 and Grade 3 (matched in reading level). Results from typical readers showed that children in Grade 5 had more correct word spellings than those in Grade 3, particularly in context-dependent and irregular spellings. Finally, a study by Almeida, Guerreiro, and Mata (1998) showed that Portuguese second graders struggled more in justifying the spelling of inconsistent than consistent words. Other studies have turned to pre-school children and addressed the impact of invented spelling programs in reading and spelling development (e.g., Martins, Albuquerque, Salvador, & Silva, 2013;Martins, Salvador, Albuquerque, & Silva, 2016). On the whole, then, an overall picture of how spelling develops across the primary school grades is lacking. Thus, the present study aimed to analyze how the complex and inconsistent features of the Portuguese orthography affect spelling acquisition beyond the first grade and how length and lexicality impact on this acquisition.

Characterization of the Portuguese spelling system
Portuguese is a Romance language from the Indo-European family. According to the Ethnologue database (Eberhard, Simons, & Fennig, 2019), it is the native language of more than 220 million people widespread across more than 10 countries. Here, we focus on European Portuguese, whose spelling and phonology do not fully overlap with other varieties such as Brazilian Portuguese (e.g., Sucena, Castro, & Seymour, 2009). Core features of the Portuguese spelling system are presented below, with phonemes indicated between slashes //, graphemes between angled brackets ‹›, and translations in italics.
Concerning the orthographic depth of sound-to-print correspondences, Portuguese has 37 phonemes to 67 simple and complex graphemes (Gomes, 2001). This proportion might suggest that one phoneme matches about two graphemes, but the actual speech-to-print distribution is not homogeneous. Some phonemes have a univocal correspondence to print (one phoneme to one grapheme), such as /p/ (‹pedra› stone /′pɛ.dɾɐ/) or /t/ (‹tela› screen /′tɛ.lɐ/). To correctly spell these sounds, children can rely on the simple rule that /p/ is always written ‹p› and /t/ is always written ‹t›. However, other phonemes have multiple correspondences (one phoneme to two, three, or four graphemes; see Moutinho, 2000), such as /i/, that can be represented by ‹e› (‹efeito› effect /i.′fɐj.tu/), ‹i› (‹bico› beak /′bi.ku/) or ‹í› (‹íris› iris /′i.ɾiʃ/), and these can be of two types: complex consistencies and inconsistencies (Schmalz et al., 2015). Complex consistencies occur when the use of correct graphemes is predictable but involves a multi-letter representation and/or takes into account contextdependent rules. These rules rely on additional information about the target phoneme, such as its phonemic context (e.g., subsequent vowel) or its position within the word (e.g., syllable-or word-initial or final positions). Inconsistencies occur when no explicitly known rules indicate the correct spelling of the word.
In spelling, children may also process information at the level of morphemes, the smaller chunks of meaning. The Portuguese language includes lexical morphemes (radicals), as well as grammatical morphemes that allow the formation of new words originating in the radical (e.g., variations in gender, such as -o / -a in ‹gato› cat and ‹gata› cat, or number, such as -s in ‹gatos› cats; see Cunha & Cintra, 2014). Because words with a common radical share a common part, children may use this knowledge as a cue to spell unknown words (e.g., if the child knows that ‹humana› human is spelled with an initial silent ‹h›, he/she can infer that the word ‹humani-dade› humanity begins also with an ‹h›).
In what follows we present features of the Portuguese spelling system that children need to master in order to become fluent readers and writers (for a description of expected landmarks in basic education for achieving proficient use of Portuguese as a native language, see, e.g., Sim-Sim, Duarte, & Ferraz, 1997).

Digraphs
A digraph refers to the use of two letters to represent a single phoneme. Digraphs do not exist in some languages, such as Hebrew (Kahn-Horwitz, Schwartz, & Share, 2011), though they are very common in others, such as French (Fernandes et al., 2008). Portuguese is somewhere in the middle: it includes two digraphs that establish biunivocal correspondences with phonemes, ‹nh› and ‹lh› (/ɲ/ and /λ/, respectively), as well as digraphs that are inconsistent, such as ‹ch›. The digraph ‹ch› represents the phoneme /ʃ/, but depending on the context /ʃ/ can also be spelled with ‹x›, ‹s›, or ‹z›. Portuguese also includes the digraphs ‹rr› and ‹ss›, which represent /ʀ/ and /s/ in intervocalic position; ‹qu› and ‹gu›, which are context-dependent digraphs (see next paragraph); and combinations such as ‹am› or ‹an› (i.e., a vowel plus ‹m› or ‹n›), which represent nasal vowels depending on context. Learning that two letters represent one sound can pose difficulties in the initial school grades. However, as learners develop their spelling ability, these letter combinations turn out to be processed as a unit (Tainturier & Rapp, 2004).

Contextual and position consistencies
In Portuguese, some phoneme-grapheme correspondences are context-dependent. Here, we use the term contextual consistencies to distinguish cases where a phoneme is represented by a specific grapheme depending on the vowel that immediately follows it, and position consistencies to refer to cases where the selection of the grapheme is determined by the position of the phoneme within the word (a particular case of context-dependent consistencies). Examples of contextual consistencies involve /k/ and /g/. If these phonemes are followed by the front vowels /e/, /ɛ/, /i/, /e/, or /ĩ/, the semivowel /j/, or schwa (/ə/), the graphemic representations ‹qu› and ‹gu› should be used. For example, /kɐ.′fɛ/ (‹café› coffee) and /′gɔ.lɐ/ (‹gola› collar) are spelled with ‹c› and ‹g›, whereas /′kɛ.dɐ/ and /′gi.zu/ are spelled ‹queda› fall and ‹guizo› rattle. If ‹c› and ‹g› were used in the latter examples, they would be read as /s/, /′sɛ.dɐ/, and /ʒ/, /′ʒi.zu/. Position consistencies occur in the representation of the phonemes /ʀ/ and /s/. In word initial position, the graphemes ‹r› (‹rato› mouse /′ʀa.tu/) and ‹s› (‹saco› bag /′sa.ku/) are used. However, the same graphemes in coda position would be pronounced /ɾ/ (‹corda› rope /′kɔɾ.dɐ/) and /ʃ/ (‹pasta› folder /′paʃ.tɐ/). In intervocalic position, /ʀ/ and /s/ have to be represented with double letters (‹rr›, ‹ss›). Another position consistency relates to the nasalization of the vowels: the grapheme ‹m› should be used before /p/ and /b/, and the grapheme ‹n› before all other consonants. Overall, both contextual and positional consistencies require children to go beyond the single phoneme and take into account its neighborhood. In Spanish, which shares with Portuguese similar constraints regarding the spelling rules described above, children from Grades 1-4 achieved similar accuracy in contextual and position consistencies (Defior, Jiménez-Fernández, & Serrano, 2006. However, the acquisition of this type of complex consistencies seems to be influenced by the depth of the orthographic learning system. For example, when asked to spell words containing context-dependent consistencies, second graders learning a shallow orthography (Spanish) outperformed their peers learning a deeper orthography (French; Carrillo, Alegría, & Marín, 2013).

Stress marks
In some languages, diacritics can be used to convey word stress. In Portuguese, three diacritics mark stress and indicate whether the vowel is open or closed. The acute accent /´/ is used in words stressed in the antepenultimate syllable and in the penultimate syllable of words ending with ‹i› or ‹l›. The circumflex accent /^/ is used in closed vowels, such as ‹â›, ‹ê›, and ‹ô›. The tilde /~/ nasalizes the vowels ‹a› and ‹o› (e.g., ‹cão› dog /′kɐw/, ‹ações› actions /′ɐ.sõjʃ/) and it is typically used in stressed syllables. The most frequent stress mark in Portuguese is the acute accent, followed by the tilde and the circumflex accent (Gomes, 2001). In Spanish, stress marks are one of the most difficult orthographic complexities to acquire (Defior et al., 2006(Defior et al., , 2009; in Brazilian Portuguese, teachers also expect proficiency in stress marks to occur only in later phases of spelling development (Pinheiro, 1995).

Inconsistencies
Inconsistent phoneme-grapheme correspondences are not regulated by rules. In Portuguese, inconsistent mappings occur mostly in the representation of the phonemes /ʒ/ and /ʃ/. The sound /ʒ/ followed by the front vowels /e/, /ɛ/, /i/, /e/, or /ĩ/, the semivowel /j/, or schwa (/ə/) can be accurately spelled with ‹j› or ‹g› (plus ‹e› or ‹i›, depending on the vowel). In reading, there is a rule determining that the letter ‹g› followed by ‹i› or ‹e› is always read /ʒ/; examples are ‹girassol› sunflower / ʒi.ɾɐ.′sɔɫ/, and ‹tigela› bowl /ti.′ʒɛ.lɐ/, respectively. But in spelling the two possibilities are allowed. Therefore, children need to invoke lexical or morphological knowledge to correctly spell inconsistent words. For example, if children know the word ‹lojista› shopkeeper, their lexical knowledge in enough to spell the word. If children are not familiar with the word, but they are aware of the morpheme ‹loj-› (from ‹loja› shop), they can use this knowledge to correctly spell the derived form ‹lojista› (where ‹-ista› indicates the profession). In case of /ʃ/, there is also no rule to determine the grapheme that should be selected by the writer, namely, ‹ch›, ‹x›, ‹s›, or ‹z›. Again, children will only produce correct spellings if they rely on the lexicon or on morphological cues (e.g., ‹coxo› lame /′ko.ʃu/, attending to the stem ‹cox-›, which is common to related words such as ‹coxear› to limp /ku.′ʃjaɾ/). Because phonological knowledge and orthographic rules cannot be applied, Portuguese children are expected to have more difficulty in mastering inconsistent over consistent correspondences, a finding already observed in their Brazilian Portuguese-speaking peers (Pinheiro, 1995).

Silent letter ‹h›
Silent letters are graphemes that do not have a phonemic correspondence. Silent letters can occupy different positions within the word and can be always silent or only in some cases. In Portuguese, the grapheme ‹h› is always silent. For example, ‹hera› ivy and ‹era› was/era are both pronounced /′ɛ.ɾɐ/. In Spanish also, words written with an initial silent ‹h› do not follow any consistent rule, and children struggle to spell them correctly in the early grades (Defior et al., 2006(Defior et al., , 2009. To successfully spell these words, children need to rely on lexical or morphological knowledge.

Length and lexicality effects
The difficulty of spelling a word or pseudoword is not uniquely dependent on the presence of orthographic complexities, such as those previously presented. One feature that can create additional demands on spelling is length. It is harder for children to spell longer items, regardless of length being assessed with number of phonemes or of letters (Treiman, 1993). Longer items can make spelling even more demanding for individuals with acquired or developmental spelling difficulties. For example, in adults with acquired dysgraphia longer words elicited a higher number of spelling errors (Buchwald & Rapp, 2003;Tainturier & Caramazza, 1996). Furthermore, it was recently observed that length had an even more detrimental effect in spelling in children and adolescents with dyslexia compared to typically developing peers (Juul & Petersen, 2017). Because stimulus length can influence spelling accuracy, considering length effects in the study of spelling acquisition allows a more refined characterization of the development of each orthographic complexity.
Another feature that impacts spelling development is lexical status. A lexicality effect in spelling occurs when writers show better performance in writing words than non-lexical stimuli such as pseudowords (e.g., Defior et al., 2009). Lexicality effects provide important information for the study of spelling acquisition. On the one hand, they suggest that spelling strategies progress from a general phonemeto-grapheme conversion procedure to the use of specific knowledge about words. In Portuguese, children from Grade 1 showed a lexicality effect in spelling simple words vs. simple pseudowords at the end of the school year, suggesting incipient acquisition of lexical knowledge (Fernandes et al., 2008). On the other hand, comparing word and pseudoword performance may inform about the consolidation of rules underlying orthographic complexities. For example, Defior et al. (2009) found that the advantage of word over pseudoword spelling in primary grades was more pronounced for digraphs, position consistencies, and stress marks than for contextual consistencies. In general, the study of orthographic consistencies in pseudowords is a useful indicator of whether children have generative knowledge of spelling rules and apply it to novel stimuli or if they just use them in words they memorized.

The present study
Our main goal was to examine how the complexities of the Portuguese orthography affect spelling acquisition of beginning writers in primary schools, taking into account stimulus length and lexical status. For this purpose, we assessed the spelling abilities of typically developing Portuguese children from Grades 2, 3, and 4 using a spelling-to-dictation task. This assessment included words and pseudowords requiring explicit or implicit knowledge of selected features of the Portuguese orthography. Our primary focus concerned the acquisition of the orthographic complexities of the Portuguese spelling system. Which features are developed earlier and possibly mastered at the end of primary school? We hypothesized that consistent features (digraph and contextual and position consistencies) and consonant clusters would be developed earlier than inconsistent features (inconsistency and silent letter ‹h› categories) and the stress mark. Because consistent features are predictable, we expected them to be learned before inconsistencies, as inconsistent features require word-specific knowledge and do not rely on general rules. In addition, the stress mark was expected to be a challenging feature. Data from Spanish children in Grades 1-4 (Defior et al., 2009) revealed that stress assignment was one of the most demanding features, not yet mastered at the end of primary school (fourth graders achieved only about 50% accuracy in stress-marked words, whereas in the remaining complexities the minimum score was 63%). Our second focus was the moderating role of lexicality and length on the acquisition of orthographic complexities across grades. Specifically, we aimed to test whether complexities became more demanding when embedded in longer words. This would suggest a possible cumulative effect of length and complexity. Regarding lexicality, we aimed to explore if, and at which point of primary education, children have consolidated rule-based knowledge in order to apply it to novel stimuli (pseudowords).
Studying the effect of different sources of complexity in spelling acquisition within a language is important for different reasons. First, the specific prevalence of each complexity depends on language. Second, the contribution of each complexity to predict spelling performance may vary across grades (Willson, Rupley, Rodriguez, & Mergen, 1999). Finally, understanding how children acquire each complexity contributes to inform teaching practices (Defior et al., 2006). From a crosslinguistic perspective, the study of an orthography of intermediate depth such as Portuguese can improve our understanding of intermediate systems. In reading, for example, Portuguese elicited the use of flexible (task-dependent) decoding strategies (Lima & Castro, 2010), an adaptive pattern that may not be so present in orthographies with marked shallow or deep features.

Participants
Participants were 189 children, native speakers of European Portuguese, from three primary schools in the Aveiro area (center of Portugal): 71 second graders (M = 94.66 months, SD = 3.90; 38 girls), 49 third graders (M = 105.43, SD = 5.07; 26 girls), and 69 fourth graders (M = 118.45, SD = 4.32; 32 girls). Children came from public schools in urban middle-class neighborhoods, and none had known hearing, visual, intellectual, or behavioral disorders. Data collection took place between January and April (the second school term in Portuguese schools). All children were individually assessed for intellectual capacity measured with the Raven's Colored Progressive Matrices (Portuguese norms from Simões, 2000), reading proficiency measured with a reading age test (Teste de Idade de Leitura, TIL; Sucena & Castro, 2008; how many correct choices within 5 min for 36 sentences where the last word is missing and has to be chosen from five alternatives), and vocabulary skills measured with the Vocabulary subtest from the Wechsler Intelligence Scale for Children-III (WISC-III; Portuguese version from Simões, Rocha, & Ferreira, 2003). Averages by grade are provided in Table 1. None of the children showed impairments in these tests.

Instructional setting
During primary school years, foundational reading and spelling competencies are part of the Portuguese language school curriculum (Buescu, Morais, Rocha, & Magalhães, 2015). In Grades 1-2, the teaching of reading and spelling is greatly focused on the development of basic phonological and orthographic skills through different activities (e.g., letter, grapheme and word/pseudoword naming and writing; phoneme deletion; syllable counting; writing-by-dictation of short texts). In Grades 3-4, teaching is more focused on developing complex reading and writing skills, particularly at the text level. Vocabulary development is fostered along the 4 years. Children contact with texts of different genres (e.g., narration, poetry) and learn to analyze their key components (e.g., main ideas, information about time, space, and characters' attitudes).

Tasks and measures
Based on the study by Defior et al. (2009), we designed a spelling-to-dictation task composed of two parts, first words and then pseudowords.

Word dictation task
This task is composed of 56 words, evenly divided into seven orthographic complexity categories with four two-and three-syllable words each (7 × 4 × 2): digraph, contextual consistency, position consistency, consonant cluster, stress mark, inconsistency, and silent letter ‹h›. The words were selected from Portulex (Teixeira & Castro, 2006), a lexical database containing words from Portuguese textbooks used in Grades 1-4. We selected words containing the most common syllable types in Portuguese (CV, CVC, V; Gomes, 2001;Vigário & Falé, 1994; see "Appendix"). Within each category, word absolute frequency ranged from low to medium values as described next: 1-36 for digraphs, 1-18 for contextual consistency, 1-30 for position consistency, 1-26 for consonant clusters, 2-50 for stress marks, 3-26 for inconsistencies, and 1-25 for silent ‹h›. Below we describe and provide examples of each complexity category.

Digraph
Digraphs are two letters representing one sound. We included the digraphs ‹lh› (/λ/), ‹nh› (/ɲ/), and ‹rr› (/ʀ/), each represented by three, three, and two words, respectively. All the selected phoneme-grapheme correspondences in this category are consistent, that is, only the target digraph could be used to correctly spell them. Examples are ‹talho› butchery /′tɐ.ʎu/, ‹banho› bath /′bɐ.ɲu/, and ‹derrota› defeat /də.′ʀɔ.tɐ/. As noted in Introduction, the case of /ʀ/ spelled as ‹rr› could also be framed in position consistencies (/ʀ/ is ‹r› in word initial position, and ‹rr› between vowels). However, we included ‹rr› in the digraph category as it seems to be its most salient property, determining specific rules to be followed by children (e.g., separating the two ‹r› in translineation). This is also reflected in the Portuguese language school curriculum, where ‹rr› is specified as a digraph (see Buescu et al., 2015).

Consonant cluster
A consonant cluster refers to consecutive consonants at the onset of a syllable (CCV syllable). In this category, the second consonant of the cluster was /ɾ/ or /l/. We included the following target spellings: ‹dr›, ‹fr›, ‹gr›, ‹pr›, ‹tr›, ‹bl›, ‹cl›, and ‹pl›. The cluster appeared in the beginning or in middle of the word (three and five items for each case, respectively).

Silent letter ‹h›
In Portuguese, the letter ‹h› as a single grapheme is used only in word initial position and has no phonemic value. Examples of our stimuli are: ‹hino› anthem /′i.nu/ and ‹humana› human /u.′mɐ.nɐ/.

Pseudoword dictation task
This task is composed of 40 pseudowords organized into five orthographic complexity categories, all but inconsistency and silent letter ‹h›. Each category included four two-syllable plus four three-syllable pseudowords. The pseudowords were formed by replacing one to two phonemes of the words in the word dictation task by another phoneme(s) of the same phonetic category (i.e., manner and/or place of articulation). For example, the word ‹talho› butchery /′tɐ.ʎu/, from the digraph category, resulted in the pseudoword ‹dalho› /′dɐ.ʎu/ by changing the first phoneme. In this case, both manner and place of articulation were kept, and the voiceless dental stop /t/ was replaced by the voiced homorganic /d/. Another example is the pseudoword ‹jagrado› /ʒɐ.′gɾa.du/, which was built from the word ‹sagrado› sacred /sɐ.′gɾa.du/ by changing place of articulation from alveolar to postalveolar. Using this procedure, we built pseudowords with the same syllable structure, length, and orthographic category as the corresponding words. To keep scoring criteria identical between words and pseudowords-only one correct spelling-we did not create pseudowords for the inconsistent categories (e.g., the pseudoword /′to.ʃu/ can be spelled as ‹toxo› or ‹tocho›, because both spellings represent correctly the phonological form).

Procedure
After gathering parental written consent and children's verbal assent, word and then pseudoword dictation tasks were administered to children in their classrooms. Each word and pseudoword was read once by the experimenter, who gave enough time for all children to spell it in a paper sheet. If a child did not understand the target word, it was repeated once. Each child was tested individually on the background measures (Raven CPM, reading age and WISC-III Vocabulary) in a separate session.

Complexity, length, lexicality, and grade effects
To examine the effects of complexity, length, lexicality, and grade on the spelling performance of primary grade children, we conducted a 5 × 2 × 2 × 3 (Complexity Category [digraph, contextual consistency, position consistency, consonant cluster, stress mark] × Length [two-, three-syllables] × Lexicality [word, pseudoword] × Grade [second, third, fourth]) Analysis of Variance (ANOVA), with repeated measures on the first three factors. As this analysis included multiple comparisons, we adopted a conservative approach by setting the alpha level to .001 to reduce the probability of Type I errors. The average percentage of correctly spelled two-and three-syllable words and pseudowords by complexity category and grade are presented in Table 2.
We found significant main effects of Complexity, Length, Lexicality, and Grade (see Table 3). Moreover, there were significant two-way interactions between Complexity and Grade, between Complexity and Lexicality, and between Complexity and Length. No significant three-or four-way interactions were found. The significant two-way interactions were decomposed with simple effect analyses, further detailed below.

Differences between complexity categories
There were differences in children's performance according to complexity for both two-syllable, F(4, 183) = 253.08, p < .001, η 2 p = 0.85, and three-syllable stimuli, F(4, 183) = 407.82, p < .001, η 2 p = 0.90. Pairwise comparisons are detailed in Table 4. With the exceptions of position consistencies and consonant clusters, the pattern of differences across complexities was similar between two-and three-syllable stimuli. For both lengths, digraphs had the highest-accuracy categories (94% and 92% in two-and three-syllable stimuli, respectively), and stress marks the lowest and most discrepant scores in comparison to the remaining categories (44% and 36% in twoand three-syllable stimuli, respectively).
Portuguese spelling in primary grades: complexity, length… Table 2 Descriptive statistics for the percentage of correctly spelled words and pseudowords by complexity category, length, and grade Inconsistency and silent ‹h› categories do not apply to pseudowords Complexities

Differences between words and pseudowords
Children were more accurate when writing words than pseudowords in the position consistency and stress mark categories (lexicality effect of about 5% for both categories) , F(1, 186)

3
Portuguese spelling in primary grades: complexity, length…

Complexity, length, and grade effects
Because the inconsistency and silent letter ‹h› categories were only represented in words, these categories were not considered in the previously presented analyses. Thus, to examine whether grade and length affected performance in these word categories, and whether this performance differed from that in the other complexity categories, we conducted an additional 7 × 2 × 3 ANOVA (Complexity [digraph, contextual consistency, position consistency, consonant cluster, stress mark, inconsistency, silent letter ‹h›] × Length × Grade), with repeated measures on the first two factors. In this new analysis, we removed the Lexicality factor and added two extra levels in the Complexity factor. Following the previous analysis, we set the alpha level to .001. As depicted in Table 5, results revealed significant main effects of Grade and Complexity. Furthermore, there were significant two-way interactions between Grade and Complexity, and between Length and Complexity, which were decomposed with simple effects analyses. No other significant interactions were found. To avoid repeating the results from the previous ANOVA, grade and length effects are only presented for the inconsistency and silent letter categories.

Differences between two-and three-syllable words
Results showed a length effect for words in the silent letter ‹h›, F(1, 186) = 34.07, p < .001, η 2 p = 0.15 (accuracy about 11% higher for three-syllable words), but not for the inconsistencies (F < 1).

Differences between complexity categories
There was an effect of complexity category in Grade 2, F(6, 181) = 172.58, p < .001, η 2 p = 0.85, in Grade 3, F(6, 181) = 93.19, p < .001, η 2 p = 0.76, and in Grade 4, F(6, 181) = 89.04, p < .001, η 2 p = 0.75. Across grades, children's accuracy was significantly lower in the inconsistency and silent letter ‹h› categories than in digraphs, contextual and position consistencies, and consonant clusters. Additionally, accuracy in the silent letter ‹h› was similar to the stress mark and inconsistency categories in Grades 3 and 4, but significantly lower than those two categories in Grade 2. Also, second-and third-graders produced more errors in stress marks than in inconsistencies, but fourth-graders performed similarly in the two categories.

Discussion
The present study was designed to analyze how orthographic complexities influence learning to spell in Portuguese, an orthography of intermediate depth. We examined the spelling accuracy of 189 children from Grades 2, 3, and 4, who performed dictation tasks composed of words and pseudowords. Words were chosen from a database of stimuli from Portuguese primary textbooks (Grades 1 to 4), and pseudowords were derived from these words. Each stimulus represented an orthographic complexity, assigned to one of seven categories: digraph, contextual consistency, position consistency, consonant cluster, stress mark, inconsistency, and the silent letter ‹h›. By focusing on a larger range of orthographic complexities and primary grade levels, we aimed to fill the gap in existing findings about spelling acquisition in Portuguese primary graders (e.g., Fernandes et al., 2008;Serrano et al., 2011). Specifically, we analyzed the effect of orthographic complexity on spelling accuracy and the moderating role of length (two vs. three-syllable stimuli), lexicality (words vs. pseudowords), and grade (second, third, and fourth). In general, results showed that there were differences in accuracy according to the complexity categories, and those differences were modulated by item length, item lexical status, and grade.
The comparison between different complexities revealed that children's best performance was for digraphs and position consistencies in words, and for digraphs in pseudowords. This result is not surprising because these categories represented entirely consistent orthographic features of Portuguese. Children can rely on unambiguous phoneme/grapheme mappings to correctly spell the phonological forms, no access to lexical knowledge required. However, it should be noted that a lexicality effect occurred for position consistencies (better performance in words than pseudowords). This result suggests that, in addition to knowledge on position-related regularities, children might have used word knowledge to spell accurately. In any case, regardless of the spelling strategy digraphs and position consistencies seem to be mastered in the very initial phases of learning: second graders showed high accuracy rates in both categories (about 90% for digraphs and 85% for position consistencies, words and pseudowords lumped together). These results complement well those observed by Fernandes et al. (2008) with first graders: children were more accurate in spelling simple or complex regular words (containing one-letter graphemes or digraphs, respectively) than irregular words.
Regarding inconsistencies, the silent letter ‹h› and stress marks, results showed that these orthographic conditions were the most difficult for both beginning and advanced primary graders when compared to the remaining categories. The advantage of consistent over inconsistent categories replicates the regularity effect observed in first graders by Fernandes et al. (2008), and shows that it persists until Grade 4. This advantage is also congruent with instructional practices: consistencies can be taught explicitly through rules that may be memorized and trained by children, but inconsistencies do not. Thus, it is not surprising that children may face difficulties in explaining the underlying principles of inconsistent spellings (Almeida et al., 1998). Fostering reading practices to increment lexical knowledge can help children to master inconsistencies, which are essentially acquired by learning new words. It is therefore possible that examining children's reading habits, either in class or at home, could explain some variability in the inconsistency category. Concerning stress marks, these have already been shown to be a source of difficulty for children speaking other languages, such as Spanish (Defior et al., 2006, 2009) and Greek (Protopapas, Fakou, Drakopoulou, Skaloumbakas, & Mouzaki, 2013).
The examination of our results by item length has showed that length significantly modulated spelling performance in only two categories, consonant clusters and stress marks, where shorter stimuli (two-syllable) were spelled more accurately than longer ones (three-syllable). These results show that the impact of length on spelling is not homogeneous. In consonant clusters, the disadvantage of longer items can be due to the cumulative difficulty imposed by the cluster itself (the phonological representation) and the more complex trisyllabic structure. By adding an epenthetic vowel or omitting one of the consonants of the cluster (common errors), children break down the complex syllable structure turning it into a simpler sequence of CV syllables. In spelling longer (pseudo)words, this syllable simplification could be a useful strategy to handle stimuli with a higher load of information to process (more phonemes and syllables). Poor performance with the stress marks, especially in longer items, may be related not only to failing orthographic knowledge, but also to difficulties in dealing with the prosodic aspects of the language and in the awareness of lexical stress.
The analysis of differences from a grade-level perspective revealed that third and fourth graders performed similarly, and better than second graders. In other words, relevant improvements in spelling abilities occurred mostly from the second to the third grade. The exception was the stress mark category, where a significant gain in performance was also observed from Grades 3 to 4. The growth in spelling abilities beyond Grade 2 was also observed in another study with Portuguese children, where spelling accuracy in a text-to-dictation task improved from Grades 3 to 4 (Horta & Martins, 2004). Nevertheless, our data suggest that the middle stage of primary school is an important landmark for the development of spelling skills, though important learning steps are still to be achieved after the third grade. Specifically, at the end of primary school, the most difficult orthographic complexities have yet to be mastered, as revealed by the mean accuracy levels of fourth graders when writing stimuli with stress marks (about 52%, words and pseudowords together), and the silent letter ‹h› and inconsistencies (about 49% and 58%, respectively). As noted before, stress marks rely on the articulation between orthographic and prosodic representations, which may be harder to acquire. Teaching stress marks pose additional demands to teachers who need to focus on non-orthographic aspects (e.g., prosody and awareness of lexical stress) and deal with the difficulties exhibited by children in mapping orthography and prosody. These additional demands may contribute to explain the slower acquisition of stress marks along primary school. Even though some teachers may consider adopting individualized approaches that could facilitate learning, these may consume extra time and may not always be feasible.

Limitations and future directions
The findings discussed above should be interpreted taking into account five limitations, which may be used to guide future research. First, as this study has a crosssectional design, it does not allow direct conclusions about children's development. Future research on spelling acquisition in Portuguese could adopt a longitudinal design across primary school years. Alternatively, a cross-sectional study with an extended range of school grades beyond the fourth grade would also be an important development, as the most difficult orthographic complexities are not yet learned after 4 years of instruction. Second, it was not possible to collect detailed data about the socioeconomic level of children and about the teaching methods adopted in each class. Future research into the acquisition and development of spelling would benefit from a refined characterization of the social, cultural and economic level of children's families, as well as of teaching methods. Information on these methods may help to explain some of the reported findings.
Third, as we assessed spelling abilities using dictation tasks with pre-defined items, the performance in those tasks may not fully resemble the spelling performance of children in spontaneous writing (e.g., a composition). Nevertheless, as the type of errors given in dictation tasks may overlap with those occurring in spontaneous writing (e.g., children's misspellings of consonant clusters in handwriting compositions; Treiman, 1991), our spelling-to-dictation task remains a valuable tool to assess the spelling abilities of Portuguese beginning writers. Moreover, although we employed a scoring method using a dual criterion (correct vs. incorrect) based on orthographic accuracy, it is possible to use other methods, such as those based on phonological accuracy. The use of more fine-grained analysis of children's spelling, either at the level of phonemes or letters, can also provide relevant information (see Treiman, Kessler, & Caravolas, 2019, for a recent analysis of the power of different scoring methods to predict spelling ability).
Fourth, due to practical reasons, we did not examine the moderating role of frequency, a variable known to impact spelling performance (Fernandes et al., 2008;Sprenger-Charolles, Siegel, & Bonnet, 1998). The inclusion of additional frequency conditions would increase task complexity, namely in number of stimuli, and this would not be feasible to include in children's assessments. Future studies targeting the systematic analysis of frequency and complexity effects together would provide a deeper characterization of the orthographic properties studied here. To keep assessment feasibility, this could be achieved either by planning a more complex task divided in two parts given to children in separate moments, or by not including length-related conditions.
A final limitation is that our spelling-to-dictation task does not cover all the possible complexities of the Portuguese orthography. For example, we did not directly assess the influence of morphological aspects in spelling. Also, the selected items do not provide an exhaustive representation of the different variations within each orthographic complexity. For instance, although we studied the most common consonant clusters in Portuguese (those with /ɾ/ or /l/ as the second consonant), there are other, less frequent forms that can also be studied (e.g., the clusters with /n/ or /s/ as the second consonant, as in ‹gnomo› gnome /′gno.mu/ and ‹psicólogo› psychologist /psi.′kɔ.lu.gu/, respectively; examples from Cunha & Cintra, 2014).
As explicit instruction is beneficial for the development of spelling skills (Graham & Santangelo, 2014), a concrete understanding of how these skills are developed in an orthography is relevant to bring light upon aspects that may merit special attention in teaching. In this regard, our study was an important step to track how children learn both easier and more difficult spelling features of an orthography of intermediate depth.

Conclusion
This study provides a cross-sectional perspective of spelling acquisition in Portuguese. We showed that complex consistencies (digraphs and context-dependent consistencies) and consonant clusters do not seem to pose difficulties for most beginning writers as early as Grade 2; across grades, these complexities were successfully mastered by at least four out of five children. Greater improvements occurred mostly from Grade 2 to Grade 3, but by the end of Grade 4 stress marks and inconsistencies were still in need of further learning and experience to be efficiently mastered. As students benefit from systematic instruction of spelling (Graham & Santangelo, 2014), our focus in how spelling abilities develop in primary grades in the context of Portuguese orthography may be informative for adapting learning and teaching practices in early school grades. Furthermore, our study is important for cross-linguistic research. By providing data on spelling acquisition in relation to grade and orthographic complexity in European Portuguese, an orthography of intermediate depth, the present study paves the way for comparisons with other orthographies varying in degree of consistency and inconsistency.