Teachers' perception of spelling patterns and

Download PDF
0 downloads 0 Views 73KB Size Report
Comment
Mar 19, 2001 - test contained 32 target items with homophonous graphemes, ... morpho(phono)logical analysis easier to spell than Dutch-speaking children. ... typological features of the language being learned, especially by the degree of .... Orthographic systems often express morphological regularities in their units,.
Teachers' perception of spelling children's spelling errors: A cross-linguistic perspective

patterns

and

March 19, 2001

Dorit Ravid Tel Aviv University School of Education and Department of Communications Disorders Tel Aviv 69978 Israel tel.: +972 3 6408626 fax: +972 3 5360394 e-mail: [email protected]

Steven Gillis University of Antwerp – UIA Center for Dutch Language and Speech –CNTS Department of Linguistics – GER Universiteitsplein 1 B-2610 Wilrijk Belgium tel.: +32 3 8202766 fax.: +32 3 8202762 e-mail: [email protected]

Acknowledgements Preparation of this paper was partly sponsored by the Flemish Government and the University of Antwerp through a GOA-grant (contract G98/3). We thank Chaviva Zavda for her help with data collection.

2 Abstract

This paper is a cross-linguistic examination of teachers’ perception of morphologically-mediated spelling patterns, compared with children’s actual spelling performance on these same patterns. The study was conducted on two languages differing greatly in their typologies: Hebrew and Dutch. The research design of this study consisted of two spelling tests, one for Hebrew and one for Dutch. Each spelling test contained 32 target items with homophonous graphemes, equally divided into four groups of conditions, differing in degree of morphological and morphophonological cues. These spelling tests were first administered to 192 Israeli and 192 Belgian gradeschool children. Hebrew-speaking children generally scored higher than Dutch-speaking

children,

and

found

the

conditions

which

required

morpho(phono)logical analysis easier to spell than Dutch-speaking children. 40 Israeli and 40 Belgian college- and university-trained L1 teachers were given the 32 items of these respective spelling tests and asked (i) to identify the pairs of items that made up the test; and (ii) to motivate their identification. On both identification and motivation, Dutch teachers generally scored higher than Hebrew teachers, and they did better exactly on those conditions that Hebrew-speaking children found easier than Dutch-speaking children. One conclusion of this study is that the underpinnings provided by the language structure determine learning patterns in spelling development. Another conclusion is that patterns of spelling regularity, easily perceived by language teachers, may not be so easily grasped by their learners. We hypothesize that the explicit metalinguistic formulation of consistencies underlying spelling patterns operates differently than natural information processing in language use.

3 1.0 Introduction Languages often challenge their speaker/writers with phonologically homophonous segments with alternative spellings. This opacity may lead to spelling errors. At the same time, homophonous spelling may encode morphological units, which may help in the decision on letter choice. For example, the English adjective suffix has three different phonetic values in , , and ([k], [s], and [S] respectively), but is spelled consistently with the same letters , formally representing its unity as a morphological construct. Previous research has shown that learning patterns of homophonous segments are mediated by typological features of the language being learned, especially by the degree of morphological complexity of the language (Gillis & Ravid, 2000; Pacton, Perruchet, Fayol & Cleeremans, in press). This paper is a cross-linguistic examination of teachers’ perception of morphologically-mediated spelling patterns, compared with children’s actual spelling performance on items spelled according to these same patterns. The study was conducted on two languages differing greatly in their typologies: Hebrew, a Semitic language with a highly synthetic morphology, and Dutch, a Germanic language with a sparse morphology. The study focuses on teachers’ explicit knowledge of the role of morphological and morpho-phonological cues in spelling homophonous graphemes in Hebrew and Dutch, with alternative spellings for the same sound, e.g., [t] spelled by or in Dutch, [t] spelled by TAF or TET in Hebrew. In both languages, homophonous graphemes may have morphological function, e.g., signifying stem or affix, verb tense or lexical category, which serve as spelling cues. Spelling may also be recoverable through morpho-phonological cues, e.g., by pluralizing Dutch words to retrieve the [t] or [d] pronunciation, or by checking if stops and spirants alternate in

4 Hebrew to determine the appropriate grapheme. Our study of spelling development in Hebrew- and Dutch-speaking children indicates that they make differential use of these cues in their spelling, in line with the inherent typologies of their respective languages (Gillis & Ravid, 2000, in press). In the current study we investigate Hebrew-

and

Dutch-speaking

teachers’

metalinguistic

ability

to

detect

morpho(phono)logical cues in the spelling tests administered to children and to motivate them in linguistic terms. Awareness of spelling patterns is treated in this paper from two points of view. One provides the background perspective: to what extent do Israeli and Belgian children make use of morpho(phono)logical cues in spelling homophonous segments in the two targeted languages and what factors govern this usage. A second point of view, constituting the main or focal perspective of our paper, seeks to determine (i) to what extent children’s knowledge and use of morphology in spelling Hebrew and Dutch is matched by their teachers’ ability to explicitly identify and explain the underlying motivation for such spelling patterns; and (ii) whether Israeli and Belgian teachers receiving their training in teachers’ colleges differ in their ability to identify and explain spelling motivation from teachers being trained at universities. In the following sections we discuss two major themes, each relating to the two perspectives of this study. One theme (section 1.1 below) relates to the difference between implicit language knowledge and its usage versus explicit and analytical language awareness (or metalanguage). Another theme (section 1.2 below) discusses the components of linguistic knowledge relevant to correct spelling. 1.1 Language usage and language awareness Language knowledge, like knowledge in many other domains such as face recognition or geometry, is essentially implicit. This complex system is typically used

5 rather than addressed as a separate body of knowledge. In the natural context of discourse, speakers normally focus on maintaining or changing the discourse topic and their role as speaker or addressee, rather than on the linguistic form. The purpose of a linguistic transaction is usually informative, and so language users focus on content to achieve their communicative goals. Therefore, while performing any ‘natural’ and authentic linguistic act where language is used rather than analyzed, linguistic knowledge is applied holistically, to construct (or comprehend) a totality that integrates phonology, morphology and lexicon, syntax and semantics in a given context (Ravid & Tolchinsky, 2000). Side-by-side with the development of implicit language knowledge, and with increasing experience in different linguistic contexts, language users develop another linguistic facet of explicit and analytic awareness (Gombert, 1992; Karmiloff-Smith, Grant, Sims, Jones & Cuckle, 1996). This alternative mode treats language as a formal problem space, focusing analytically on its components as a cognitive goal in its own right. Metalanguage requires the ability to introspect on the linguistic components that blend together naturally in language usage - phonemes, morphemes, words, syntactic structures, and discourse types. Thus it involves an analytical perception of units of language, the ability to represent each unit separately, disassociating form from semantic content, and conscious monitoring of one’s own linguistic knowledge (Bialystok, 1986). Much of the metalinguistic research has typically focused on the onset and development of phonological awareness in preschoolers (Goswami, 1999). More recently, researchers have begun to look at morphological awareness, a knowledge domain that involves introspecting about the morphemic structure of the word (Carlisle & Nomanbhoy, 1993; Ravid & Malenky, in press; Wysocki & Jenkins,

6 1987). Language awareness is not a uniform phenomenon. It increases in explicitness and concurrently involves representational reorganization into more coherent, denser and more accessible forms (Karmiloff-Smith, 1992). For example, perception of the consonantal root elements in Hebrew emerges early on, but becomes more explicit in older subjects (Ashkenazi & Ravid, 1998; Ravid, in press a). Tasks requiring controlled, analytical, explicit verbalization of linguistic processes and constructs are beyond the capacities of young children, and may not be fully achieved before adolescence (Ashkenazi & Ravid, 1998; Nippold, 1998). Moreover, metalinguistic insights reflect different perceptions of language at different ages (Nippold, Uhden & Schwarz, 1997). For example, in becoming efficient readers and writers, the most important morphological aptitude in English learners is the growing ability to segment, extract and discuss stems and affixes (Henry, 1993; Lewis & Windsor, 1996; Smith, 1998; Wysocki & Jenkins, 1987), and Israeli gradeschoolers have been shown to benefit from morphological awareness in text comprehension and prosodic reading (Appel-Mashraki, 2001). The changing nature of linguistic awareness is a combined result of development, language experience, and school instruction. Being able to represent and access linguistic form and meaning at will is the result of a complex, unified, coherent body of linguistic knowledge is thus possessed only by linguistically literate adults (Ravid & Tolchinsky, 2000). Despite its enormous importance in developing literacy, linguistic awareness is not the primary target in teaching gradeschool and highschool students, but rather an indirect means of getting a handle on a solution to problems in coming to grips with written language as both a discourse style and a notational system; and as a factor that explains children’s choice of strategy in approaching language problems.

7 School children normally cannot and are not expected to provide a comprehensive explanation of the linguistic underpinnings of language tasks such as spelling homophonous graphemes. Rather, they are expected to spell correctly. Teachers are a different matter: L1 language teachers are indeed expected to understand the linguistic structure of their native tongue, to have a systematic knowledge of its linguistic categories and concepts, and to be able to relate and analyze them explicitly, using conventional linguistic terminology. This meta-knowledge about language is a component of what makes them teachers and presumably enables them to understand what processes children go through in their literate development, to evaluate this development and to provide remedial treatment when necessary (Wilson, Shulman & Richert, 1987). 1.1.1 Spelling and language awareness Spelling, though belonging to the written domain and related to school-type activities, is an authentic linguistic act which usually addresses the spelled word as a whole when conducted in the natural course of writing. When attending to a spelling test, spellers may need to pay more attention to specific sites in the word which are likely to cause trouble. Thus the task of spelling requires a certain degree of language awareness, but this does not go beyond identifying the set of possible graphemes to be inserted in the word. Reconstructing a spelling test on linguistic grounds, and providing the motivation for the selection of spelling items, is a very different matter. It calls for the ability to represent a number of linguistic systems, examine them from phonological, morphological and syntactic points of view, and map linguistic categories onto a set of graphemes in accordance with specific conventions. This requires a highly analytical ability, access to a broad array of linguistic structures, systematic knowledge of the language domains involved, and the ability to relate

8 domains in novel ways as well as verbalize the process. This is the type of language awareness that Karmiloff-Smith terms E2/E3: levels where linguistic knowledge is recoded so as to make conscious access and verbal report possible, which have not been investigated empirically so far (1992: 22-23). Section 1.2 below discusses the language domains which are relevant to spelling, on the one hand, and which need to be accessed and integrated in order to understand and represent the structure of a spelling test with a linguistic agenda, on the other. 1.2 Knowledge domains relevant to spelling Four knowledge domains are relevant to spelling alphabetical orthographic systems develops in four domains: Mapping phonology onto graphemic segments; learning about internal conventions of the orthographic system; learning about the reflection of morphological regularities in the spelling system; and mapping morphophonological segments onto written representation (Gillis & de Schutter, 1996; Nunes, Bryant, & Bindman, 1997; Ravid, in press a; Totereau, Theverin, & Fayol, 1997). 1.2.1 Phonology Homophonous graphemes, which provide alternative spellings for the same phoneme frequently occur in many orthographies. Therefore, plotting out the spelling of specific words and learning to spell systematically also means learning to overcome opacity in mapping phonemes onto graphemes (Goswami, 1999). The two languages investigated in this study differ in their “orthographic depth”, yet both have homophonous graphemes, e.g., the segment [k] and the graphemes and in Hebrew; the segments [e], [E] and schwa and the grapheme in Dutch. 1.2.2 Internal orthographic conventions

9 Alphabetic orthographies are governed by internal principles and internal consistencies that have to be figured out by learners (Treiman & Cassar, 1997). One such issue is vowel representation: Dutch, like all languages using Latin and Cyrillic script, has a vocalized orthography. This means that both consonants and vowels are represented consistently using graphemic units from the same domain, namely, letters. Hebrew, in contrast, has two orthographic versions: a vocalized orthography representing consonants by letters and vowels by diacritic marks as well as by letters; and a nonvocalized orthography, representing all consonants, with vowels partially and ambiguously represented by the letters . This is the default version of written Hebrew which we used in the Hebrews spelling test described in this paper. Knowledge in this domain also includes precise information about sites of word segmentation and how to segment content and function words. In Hebrew, some of the function letters designate syntactic constructs – conjunctions such as ‘and’, ‘that’, the definite article, and four prepositions - which are written attached to the next word, e.g., the written string [vehabajit] ‘and the house’ (Koriat, Greenberg & Goldshmid, 1991; Ravid, in press a). 1.2.3 Morphology Orthographic systems often express morphological regularities in their units, which can be assumed to exist in the linguistic cognition of mature spellers and to facilitate conventional spelling despite the disrupted phoneme-to-grapheme mapping (Jones, 1991; Treiman & Cassar, 1996, 1997; Treiman, Zukowski & RichmondWelty, 1995). Morphology is crucially important to the current study, since the two languages under investigation differ in the degree of their morphological syntheticity, which may affect spelling patterning in development (Ravid & Tolchinsky, 2000).

10 In Hebrew, affix letters – inflectional, derivational and clitic morphemes - are all spelled regularly and consistently. For example, Hebrew [t] has two alternative spellings: and (the latter representing a neutralized historically emphatic coronal stop). However, affixal t, as in past-tense suffix –ti, is always spelled . Previous work has shown that Hebrew speakers make use of this information from early on, and that affix letters are spelled correctly earlier than root letters in Hebrew (Ravid, in press a). A morphological principle guides the spelling of Dutch words: The "principle of similarity", entails that a word, stem, prefix or affix is always written in the same way; and "the principle of resemblance" holds that words that are formed in the same way are written in the same way. In the orthography this results in highly morphologically transparent word forms. For instance, the simple present, third person singular is formed by adding the suffix to the stem of the verb (except when the stem ends with a ). Thus 'he plays' is written in Dutch as ( + ); 'he answers' is spelled ( + ); and the exception is 'he eats', which is not written with a final geminate: (instead of ). In other words, Dutch orthography abstracts away from the effects of phonological rules such as final devoicing, voice assimilation, and other rules of connected speech, in order to preserve the identity of morphemes (Instituut voor Nederlandse Lexicologie, 1997).

1.2.4 Morpho-phonology In

addition

to

encoding

meaning-carrying

morphemes,

alphabetic

orthographies also express morpho-phonological information in systematic ways. This information may be used to recover underlying phonological distinctions by

11 extracting

morpho-phonological

segments,

analyzing

them

and

comparing

morphologically related words. Recovering neutralized phonological distinctions in Hebrew may involve comparing stop/spirant pairs [p,b,k] / [f,v,X] as in [miXtav] ‘letter’ / [ktuba] ‘marriage contract’, both from root k-t-b. Some of these stop/spirant alternants are homophonous with other phonological segments. For example, [X] can derive from either a spirantized /k/, spelled , or from /Ì/ (pharyngeal fricative) neutralized to [X], spelled . Being able to juggle word forms in one’s mind to see if the spirant in the word alternates with a stop may help in selecting the correct letter in spelling. Since the [X] in [miXtav] alternates with [k] in words from the same morphological family sharing the same root such as [ktav] ‘writing’, [ktovet] ‘address’, [ktuba] ‘marriage contract’, it can be assumed that it is spelled with . In contrast, the [X] in [maXs&ev] ‘computer’ never alternates with [k] in related words, e.g., [Xis&ev] ‘computed’, [Xas&av] ‘thought’, [hitXas&vut] ‘consideration’, therefore it should be spelled with . Another type of morpho-phonological clue in learning to spell in Hebrew involves vowel lowering. Though pharyngeals and glottals are no longer pronounced in mainstream Israeli Hebrew, they nonetheless continue to operate at the morphophonological level, mainly by attracting low vowels in their environment. For example, [dereX] ‘road’ and [keraX] ‘ice’ share the same pattern CéCeC (in which Cs stand for root consonants) as well as a final segment [X]. This segment derives from a spirantized /k/ in [dereX], which accounts for the spelling ; and from a neutralized pharyngeal fricative /Ì/ in [keraX], which accounts for both the deviant phonological form CéCaC with the low vowel [a], as well as for the spelling . The association of low vowels, especially [a], with one of the possible letters, may aid in choosing the correct spelling.

12 One of the main morphophonological clues that can be used in Dutch involves 'undoing' the final devoicing of voiced segments in auslaut: The voiced segment surfaces when it is pronounced in intervocalic position. For instance, the final devoiced [d] in {av´nt}, written as ('evening'), surfaces in the plural {av´nd´}, and the final devoiced [d] in the verbform {Antwort} ('answer') surfaces in the simple past form {Antword@´}. Given this background, our study investigates the domain of spelling acquisition in two languages with differing typologies: Hebrew, a Semitic language with a rich morphology and a “deep” orthography, and Dutch, a Germanic language with a sparse morphology and a “shallow” orthography. The goal of this paper is to find out to what extent Dutch- and Hebrew- teachers with different training backgrounds (teachers’ colleges versus universities) are able to detect and explain the linguistic motivation underlying a spelling test in their native tongue. 2.0 The children’s spelling study (Gillis & Ravid, 2000, in press) Despite the typological and orthographic differences between Hebrew and Dutch, they share the same phenomenon: homophonous graphemes. For different reasons, certain phonological distinctions in both Dutch and Hebrew are neutralized, yet these segments are mapped onto distinct graphemes. Such opaque or “deep” phonology-to-orthography mapping constitutes an obstacle to the acquisition of orthographic, or conventional, spelling. For example, the two Dutch words (/arEnt/, ‘eagle’) and (/aVEnt/, ‘officer’) share a final [t] in speech due to final devoicing, however written Dutch retains the / distinction in the spelling. Similarly, Hebrew [tarim] (/tarim/ ‘you,SgMasc-will-lift’) and [ta’im] (/ta9im/ ‘tasty’) share an initial [t] due to historical neutralization processes, however written Hebrew makes a distinction between the spellings of and . The spelling

13 study which forms the basis for the current teachers’ study was concerned with the ways Dutch and Hebrew learners employed morphological and morpho-phonological cues in order to spell phonologically neutralized segments. In order to compare similar, though not identical phenomena in two typologically divergent languages with different orthographies, we created identical test conditions (see below). 2.1 Subjects and procedure The study population consisted of 192 Israeli and 192 Belgian monolingual schoolchildren with a middle-high socio-economic background from grades 1-6, all native speakers of Hebrew and Dutch respectively. Subjects were presented with a spelling test containing neutralized phonological segments. They were asked to spell the target words, which were given in a sentential context to ensure clear and nonambiguous understanding. Each target word contained one target grapheme. 2.2 Materials Condition I: Morphological and morpho-phonological cues In both Hebrew and Dutch, Condition I of the spelling tests contained 8 homophonous target segments recoverable through both morphological and morphophonological cues. In Dutch, Condition I consisted of pairs of verbs in present tense and in past participle ending with surface [t] due to final devoicing, e.g., {b´tov´rt} spelled ‘bewitch, present tense’ / ‘bewitch, past participle’ with and respectively. There are two ways to recover the difference in the spelling:

(1) through morphology, that is present tense spelled with , past

participle spelled with ; (2) through morpho-phonology, by converting the past participle forms to an adjective or to the simple past, {b´tov´rd´} both spelled , thus recovering the spelling.

14 In Hebrew, Condition I consisted of pairs of words containing the same surface segment [v] which may be spelled either by or , following the historical form of the word, and according to its morphological role as either function or root letter. For example, in the form [vair] ‘and-city’, the [v] designates a function letter ‘and’, spelled . In [uvair] ‘and-bright’ [v] is a root letter spelled with . The different spellings may be recovered either through morphology - [v] as function letter is always spelled , whereas [v] as root letter may be spelled either as or ; or through morpho-phonology: always represents a spirant, whereas represents an alternating pair of stop and spirant, which can be detected by morphological conversions. Condition II: Morpho-phonological cues Condition II contained 8 homophonous items with a morpho-phonological, but not a morphological, conversion cue for each language. In Dutch, Condition II consisted of pairs of nouns ending with surface [t] due to final devoicing, e.g., ‘eagle’/ ‘officer’. The final segment is part of the stem, with no separate morphological value, however it is morpho-phonologically recoverable by preventing final devoicing through pluralization: singular [arEnt] ‘eagle’ is pluralized to [arEnd´] ‘eagles’, spelled ; whereas singular [aVEnt] ‘officer’ is pluralized to [aVEnt´] ‘officers’, spelled . In Hebrew, Condition II consisted of pairs of words in the same morphological pattern, containing a surface [X]. This identical segment is a root letter in both cases, which is morpho-phonologically recoverable through the low vowel associated with [X] deriving from low guttural /Ì/ and spelled , e.g., [kerax] ‘ice’ . Condition III: Morphological cues

15 Condition III contained 8 homophonous items with a morphological, but not morpho-phonological, conversion cue for each language. In Dutch, Condition III consisted of pairs of verbs containing a surface [t], which may be spelled either as a single or a geminate . There is no morpho-phonological conversion rule, however the members of each pair have distinct morphological functions: [t] spelled as occurs in adjectives, e.g., ‘required, adjective’, whereas [t] spelled as occurs in simple past, e.g., ‘required, simple past’. In Hebrew, Condition III consisted of pairs of words containing a parallel segment [t] which may be spelled as either or . There is no morphophonological conversion rule in Modern Hebrew to recover the neutralized phonological segments /t/ and /t/2 respectively, however the members of each pair have distinct morphological functions: [t] standing for a function letter is always spelled , e.g., [kas&ot] ‘hard, Fm,Pl’ is spelled with a final feminine suffix; whereas [t] standing for a root letter may be spelled as either or , e.g., [mas&ot] ‘oar’ is spelled with a final root letter . Condition IV: No Cues Condition IV consisted of 8 homophonous segments with two possible spellings with no recoverability through either morphological or morpho-phonological cues. In Dutch, the test items were pairs of words containing the diphthong [Eê], which can be spelled as either or , for historical reasons3, e.g., ‘pigs’ / ‘trains’. This is how minimal orthographic pairs like 'guide' and 'suffer' are created. In Hebrew, the test items were pairs of words containing the vowel [i], which may be either spelled by or else not represented at all in non-vocalized Hebrew spelling, e.g., [min] ‘from’ spelled compared with [min] ‘gender’ spelled . The linguistic conditions under which these two

16 spellings occur are either arbitrary or available only to specialists in historical Hebrew phonology. Table 1 presents the overall schema of the spelling test, the set of testwords can be found in the Appendix. INSERT TABLE 1 ABOUT HERE 2.3 Predictions Our straightforward prediction was that we could expect better results, that is, fewer spelling errors, in both languages in the more motivated conditions with morpho-phonological and morphological cues. Thus we expected the children to make the least errors in items from Condition 1 and most errors on items from Condition 4. 2.4 Results On the whole, Hebrew- speaking children did better on the spelling test than their Dutch-speaking counterparts. In Figure 1 the success scores are plotted out per grade and per language. INSERT FIGURE 1 ABOUT HERE Moreover, our predictions seemed to work very well for Hebrew and to predict the learning behavior of Israeli children, but did not work well for Dutch. In Figure 2 we present a general picture of the results: for each condition the percentage of correct responses in each language is plotted out. The Hebrew results show that, indeed, the motivated condition (Condition 1, C1) led to the least results, and the least motivated condition (C4) led to the most errors in the spelling test. In Dutch however, children made most errors in Conditions 1 and 3 (C1, C3) and least errors in Conditions 2 and 4 (C1, C4). INSERT FIGURE 2 ABOUT HERE

17 Full details and a discussion of the children’s spelling study are given in Gillis & Ravid (2000, in press). The most relevant finding of the spelling study, the one that will also be investigated in the teachers’ study, is the patterning of the experimental conditions. Figure 1 shows that in Dutch there are two clear clusters: Dutch-speaking children score relatively low on the morphologically motivated conditions (Conditions 1 and 3) and high on the morphologically unmotivated conditions (Conditions 2 and 4). The scores of the Hebrew-speaking children are much closer, and, more importantly they achieve their highest score on the morphologically motivated conditions (Conditions C1 and C3) and the lowest scores on the morphologically unmotivated conditions (Conditions C2 and C4). Morphophonological recoverability is not prominent for Dutch children while it is a good cue provider for Hebrew-speaking children. 3.0 The teachers’ study In Gillis & Ravid (2000, in press) the difference between the Dutch-speaking and the Hebrew-speaking children was attributed to typological differences in their mother tongue. Briefly summarized, we hypothesized that growing up in a morphologically rich language makes children more attuned towards the role of morphology, and hence it is easier for Hebrew-speaking children to deal with spelling puzzles that require morphological sensitivity (such as the correct spelling of homophonous graphemes that have a morphological function). Dutch-speaking children acquire a morphologically sparse language. Their attention is drawn much less to morphology and hence they find it much more difficult to deal with homophonous graphemes the correct spelling of which depends crucially on morphological analysis.

18 However, there is an alternative explanation that does not put the burden of explanation on the children’s linguistic functioning, but on teachers’ knowledge of spelling difficulties and on their knowledge of spelling rules. Starting from the observation that in general Hebrew-speaking children were better spellers than Dutchspeaking children, it can be hypothesized that teachers of Hebrew as L1 are more aware than teachers of Dutch as L1 of the spelling patterns that cause problems for the children and/or that they are more knowledgeable of the spelling rules and morphological underpinnings of the difficult cases. This hypothesis leads to two specific predictions that were tested in the experiment that we will report on in this section. The first prediction is that if Hebrew teachers in Israel are more sensitive than teachers of Dutch in Belgium of the specific difficult spelling patterns for their pupils, they would be more successful in identifying the word-pairs in the children’s dictation task. In other words, if we ask the teachers to identify the pairs of words that were the target items in the children’s spelling test, we would expect Hebrew-speaking teachers to attain a higher score than Dutch-speaking teachers. The second prediction is that when asked to formulate the spelling rules governing the pairs of target items, Hebrew-speaking teachers are expected to exhibit a better mastery or higher degree of proficiency in spelling out the rules than their Dutch-speaking colleagues. 3.1 Population 40 Hebrew-speaking and 40 Dutch-speaking L1 novice teachers participated in this study. Half of the population consisted of students of a teacher training college (henceforth, TTC) and the other half were university students enrolled in a teaching qualification program (henceforth, UNI).

19 The Israeli TTC students were in the 4th and last year of their professional training. From their first year, TTC trainees had had extensive practical training and experience in teaching in gradeschool and junior highschool under the supervision of teacher trainers and training teachers. From a curricular point of view, the Hebrew department at the teacher training college we sampled is affiliated to the Hebrew language department at Tel Aviv University, and the syllabus is similar. Their college degree allows them to teach Hebrew in gradeschool and junior highschool. The UNI students were graduates of the Hebrew Language Department and enrolled in the teacher training program at the School of Education where they take courses on psychology, pedagogy and language learning and undergo a training program, mainly in junior highschool and highschool, similar in principle to that of TTC trainees but less extensive. Their teacher qualification allows them to teach Hebrew junior highschool and all levels of highschool. The Belgian population consisted of students of a teacher training college (“pedagogisch hoger onderwijs”). They were all in their third and final year of professional training. From their first year of training onwards, these students have had extensive practical training and experience in teaching in gradeschool. Almost their entire third year of training is spent in gradeschools teaching under the supervision of a training teacher and a teacher trainer. The university students come from the department of Germanic languages where they are enrolled in a full curriculum in literature and linguistics. In addition these students take a teaching qualification in the pedagogical department of the university. During two years they attend classes on various theoretical aspects of their future profession (such as developmental psychology, fundamental and applied didactics, etc.). Their practical training is far more restricted in comparison with the

20 TTC students: on the whole they spend one week in a secondary school actually teaching. Their teacher qualification allows them to teach Dutch, English and/or German in the upper three years of secondary schools. 3.2 Materials and procedure The 32 items of the Gillis & Ravid spelling test were randomized and presented in a written list, stripped of the sentential context in which they had appeared. Participants were told that these were items which comprised a spelling test for gradeschool, and were asked: (i) to identify the pairs that the spelling test originally consisted of (pairing task); (ii) to justify and motivate their choice in writing (motivation task). Examples were provided. 3.3 Scoring Pairing (identification) task. For each correct identification, participants received 1 point, and for each incorrect identification they received 0. Motivation task. In order to evaluate participants’ responses for Hebrew and for Dutch, we mapped out the linguistic phenomena involved in the correct spelling of the four conditions. A scale of 0-3 was then constructed to compare participants’ responses against this grid. Thus, across the two languages under consideration, the scales reflect the internal motivation of the original categories: Condition I requires the discussion of two cues (morphological and morphophonological) for a maximal score; Condition II requires the discussion of one cue (morpho-phonological) for a maximal score; Condition III requires the discussion of one cue (morphological) for a maximal score; Condition IV requires an explicit statement of the arbitrary spelling of the pair. In general, the motivation for assigning 0-3 scores across the two languages was as follows:

21 •

0 points. Incorrect or no identification, totally incorrect and irrelevant motivation.

•

1 point. Identification of the relevant graphemes or phonemes, or a vague nonspecific generalization, e.g., that’s how you spell verbs.

•

2 points. Identification of the relevant graphemes or phonemes, and a partial, incomplete, or convoluted motivation.

•

3 points. Identification of the relevant graphemes or phonemes and a full, explicit verbalization of the phenomena involved, though not necessarily matching the degree of linguistic explicitness provided above. 3.4 Predictions Given the finding of our previous study that Hebrew speaking children are

better spellers than Dutch speaking children, we expected children's performance to be reflected in the teachers' results in both tasks. The underlying hypothesis was that the Hebrew-speaking children scored better than the Dutch-speaking children on the spelling test because their teachers were better tuned to where the spelling problems are and had a better mastery of the spelling rules, and hence they were more efficient in overcoming those problems. More specifically, on the identification task, we expected target pairs to be more readily identified by the Hebrew-speaking teachers, and less well identified by the Dutch-speaking teachers. On the motivation task, we expected Hebrew-speaking teachers to be better in formulating the spelling rules than the Dutch-speaking teachers. We also expected the recoverability cues to be more readily stated by the Hebrew-speaking teachers than by their Dutch-speaking colleagues, since the Hebrew-speaking children scored much better for the recoverable items than for the non-recoverable ones.

22 As to the difference between the TTC and the UNI students, we expected a major difference between the two languages. In the Belgian population we expected a large difference between the TTC and the UNI teachers. Given the assumption that TTC teachers would have difficulties in identifying the pairs and in formulating the spelling rules, UNI teachers were predicted to score higher on both tasks since they would benefit from their comprehensive training in linguistics. In the Israeli population, on the other hand, we did not expect a major difference between the TTC teachers and their UNI colleagues: The UNI teachers would benefit from their training in abstract thinking about (their) language as their Belgian colleagues. But since the TTC teachers were expected to perform well on the identification and the motivation task, their performance should hardly differ from the UNI teachers. Given the finding that Dutch children's results show a clustering of conditions and that the Hebrew children's results do not show a similar clustering, we expected to find similar clusterings in the teachers' results. More specifically: Dutch speaking children made significantly more errors in Conditions 1 and 3 (their performance was around 50%) than in Conditions 2 and 4. We expected to find the same clustering in the Dutch teachers' results. Hebrew speaking children's results showed no similar clustering, and hence we did not expect a significant difference between the teachers' performance in the pairing task. In other words, we expected to find the children's spelling performance to be reflected in the teachers' results. 4.0 Results We first present the results on the pairing task and then on its motivation. 4.1 Identification (pairing) task We first present the results on the pairing task, i.e., identifying the original pairs of the spelling test. A 2 (Language) x 2 (Type of Teacher: UNI-TTC) x 4

23 (Condition: the four experimental conditions) Wald Chi Square effect test was performed on the pooled data. The Language effect was significant (X2 = 23.36, p < 0.0001), as well as the Type of Teacher (X2 = 8.18, p < 0.0042), the Condition effect was not significant: X2 = 3.32, p < 0.3445). There were significant interactions between Language and Type of Teacher (X2 = 10.37, p < 0.0013), and Language and Condition (X2 = 21.0, p < 0.0001). All other interactions turned out to be insignificant. The test results indicate that language is a significant main effect. The relevant data are displayed in Table 2. But contrary to our expectations - we expected Israeli teachers to score higher than the Belgian teachers - the Dutch-speaking subjects were significantly more successful in identifying the pairs of test words than the Hebrewspeaking subjects. Thus although the Dutch-speaking children scored lower on the spelling test than the Hebrew-speaking children, the Dutch-speaking teachers were significantly better in identifying the word pairs than their Hebrew–speaking colleagues. INSERT TABLE 2 ABOUT HERE We expected a significant interaction between Language and Condition. The effect test revealed that there was indeed such an interaction between those two variables. The relevant data are displayed in Table 3. INSERT TABLE 3 ABOUT HERE The marked difference between the conditions in the Dutch-speaking population was expected, but contrary to our expectations there is also a marked difference in the Hebrew population. Recall that our expectations were based on the results of the children’s tests and we expected teachers to show a similar picture Thus for Dutch the morphologically motivated conditions (Conditions 1 and 3) should

24 show a lower success score than the morphologically unmotivated conditions (Conditions 2 and 4). For Hebrew the opposite pattern was expected to show up, judging from the results of the children’s tests. In Table 4 the results for the morphologically motivated conditions and the morphologically unmotivated ones are shown for both languages. These data are graphically represented in Figure 3, and for the sake of clarity we also show the results of the children’s tests. INSERT TABLE 4 ABOUT HERE INSERT FIGURE 3 ABOUT HERE The teachers’ results (Table 4, Figure 3, bottom) are almost an exact mirror image of the results of the children (Figure 3, top). Table 4 shows that for the morphologically unmotivated conditions (Conditions 2 and 4), Hebrew and Dutch teachers score approximately at the same level. The main difference lies in the morphologically motivated conditions (Conditions 1 and 3). Contrary to what was expected on the basis of the children’s performance of the spelling tests, the Dutchspeaking teachers score very high on these conditions (95%) while the children scored at approximately chance level (see also Figure 3, top). The score of the Hebrewspeaking teachers is around 75%, while the children almost reached a ceiling score for those conditions at the end of gradeschool. The population consisted of TTC and UNI students. In Table 5 the percentage of correct responses for the two subgroups of subjects in the two languages is provided. INSERT TABLE 5 ABOUT HERE The effect test showed a significant interaction between Language and Type of Teacher. In the Dutch population there is virtually no difference between UNI and TTC. In Hebrew TTC students score markedly lower than UNI students. Both results

25 are contrary to our expectations: since Dutch-speaking children scored much lower than Hebrew-speaking children, we hypothesized that the Belgian teachers would find it much more difficult to identify the pairs of target words than their Israeli colleagues. However, the contrary is the case: although the children are poorer spellers, the Dutch-speaking teachers had no difficulty at all to detect the relevant pairs. On the other hand, the Hebrew-speaking teachers from TTC score remarkably low: they were not able to identify 30% of the pairs of test words. As predicted, UNI students scored at the same level: both Hebrew- and Dutch-speaking attain around 90% correct pairings. However the décalage between the Israeli TTC and UNI teachers as well the almost equal results for Belgian TTC and UNI students were both unexpected. 4.2 Motivation task The aim of the motivation task was to find out if the teachers could reconstruct the reasons why the pairs of items featured in the tests. This meant that they had to be aware of the locus of the spelling difficulty (what are the target graphemes?), formulate in some detail the spelling rules (if any) involved and indicate the recoverablility conditions (if any). The answers were scored on a four-point scale, and each consecutive point on the scale indicates a growing sophistication of the answer. Similar to our predictions for the pairing task, we expected Hebrew-speaking subjects to outscore the Dutch-speaking subjects. In order to test our predictions a 2 (Language, Hebrew vs. Dutch) x 2 (Type of Teacher, TTC vs. UNI) x 4 (Condition, the four experimental conditions) Wald Chi Square effect test was performed on the pooled data. Language yielded a significant effect (Wald Chi Square = 61.22, p < 0.0001) as well as Type of Teacher (Wald Chi

26 Square = 19.70, p < 0.0002) and Condition (Wald Chi Square = 163.18, p < 0.0001), but these effects are less important for the present study. More interestingly there was a significant interaction between Language and Type of Teacher (Wald Chi Square = 14.05, p < 0.0028), as well as between Language and Condition (Wald Chi Square = 71.36, p < 0.0001) and between Type of Teacher and Condition (Wald Chi Square = 19.22, p < 0.0234). There was no significant three way interaction between Language, Type of Teacher and Condition (Wald Chi Square = 12.05, p < 0.2106). We will first give an overview of the scores per language. INSERT TABLE 6 ABOUT HERE First of all, the mean score for the Dutch-speaking subjects is 1.69 (SD = 1.03) and for the Hebrew-speaking subjects the mean score is 1.21 (SD = 1.16). Thus contrary to our expectations, Dutch-speaking subjects attain a better overall score than Hebrew-speaking subjects. This difference also appears when we look at the distribution of the scores. Table 6 shows per language what percentage (N = 640) of the answers was assigned the respective scores. Hebrew-speaking subjects have more than twice as many 0-scores, which means that either they did not identify a correct pair of words, or they did not identify the correct pair of target graphemes or sounds. For the identification of the target graphemes or sounds (score 1), Hebrew-speaking and Dutch-speaking subjects score almost identically: in almost 30% of the answers, only the target letters or sounds were identified by the subjects. Score 2 accounts for 30% of the answers of the Belgian teachers and only 12% of the answers of the Israeli teachers. This means that in one answer out of three, the Dutch-speaking teachers identify the target graphemes or sounds and they point at the morphological (or morphosyntactic) regularity that underlies the correct spelling of the target grapheme.

27 In the case of the Hebrew subjects, this type of answer occurs much less often. A complete answer, viz. one in which the target graphemes or sounds are correctly identified, the morphosyntactic regularity is correctly phrased and the recoverability (if any) is indicated, occurs almost as frequently in both populations. The children’s results on the spelling test revealed a striking difference between the two languages with respect to the morphological function of the target letters (see Figure2): when morphological function was involved, Dutch-speaking children scored much lower than when no morphological function was involved. Moreover, the Dutch-speaking children’s scores were significantly lower than the Hebrew-speaking children’s score for those conditions. We predicted that the same picture would occur for the teachers. This prediction was not confirmed: the mean scores in Table 7a show that Dutch-speaking teachers score consistently higher than their Hebrew-speaking colleagues. When grouped according to the morphological function of the target letters (Table 7b), the results show even more clearly that (1) in Dutch the morphologically motivated conditions score much better than the unmotivated ones; and (2) while Dutch-speaking and Hebrew-speaking teachers show the same pattern, the mean scores for Dutch are considerably higher than the ones for Hebrew. INSERT TABLE 7 ABOUT HERE When we split up the results according to the type of teacher (TTC versus UNI) in each language (see Table 8), it appears that the Dutch-speaking subjects score in a highly comparable way. The mean scores of the TTC and the UNI students are almost identical (respectively 1.68 and 1.69), and the distribution over the four scores is also very similar. The results for the Hebrew-speaking teachers show a clearly different picture: the TTC mean score is far below that of the university students: 0.94

28 versus 1.48. And the distribution over the scores is also very different: almost 50% of the answers of the Hebrew-speaking TTC was scored as ‘0’, which means that either they failed on the word pairing or they did not identify the correct target graphemes or sounds. UNI students performed much better in this respect. In almost one fourth of the cases, the Hebrew-speaking TTC teachers only identified the correct target graphemes or sounds. The scoring categories ‘2’ and ‘3’ which imply that also the morphological regularities were identified and / or the recoverability conditions were clearly stated occurred much less often for the Hebrew-speaking TTC teachers. INSERT TABLE 8 ABOUT HERE 5.0 Discussion This study examined spelling performance and knowledge in two groups of subjects: Hebrew- and Dutch-speaking gradeschool children, who took spelling tests containing homophonous letters in their respective languages; and Hebrew- and Dutch-speaking language teachers, whose task it was to analyze the same spelling items linguistically so as to identify the test pairs and motivate their linguistic relationship. The children’s spelling results revealed higher scores in Hebrew than in Dutch spellers, despite the fact that Dutch orthography is more transparent than Hebrew orthography and does not encode as many complex morphological relationships. Moreover, young Hebrew spellers did especially well in the conditions that required morphological and morpho-phonological knowledge, and worse on those conditions that did not require such knowledge. Young Dutch spellers did not find morphological and morphophonological cues helpful, and did especially well on those conditions that did not provide such cues.

29 Two kinds of explanations may account for these results. The typological account claims that children’s spelling is motivated by knowledge patterns and strategies deriving from their native tongue. Hebrew, a synthetic Semitic language, encodes many syntactic and semantic notions in a wide array of morphological devices, and therefore Israeli children treat both spoken and written words as complex entities, looking for morphological regularities. Dutch, a morphologically sparser language guides children to look for cues in the lexical and syntactic domains rather than

in

morphology.

Hence

Israeli

children’s

better

ability

to

handle

morpho(phono)logical cues in spelling, which appeared in 3 out of 4 conditions. The teaching profile account claims that Hebrew teachers are better able to serve as mediators between young Hebrew spellers and their spelling system than Dutch teachers since they are themselves better aware of spelling patterns and their underlying motivations than their Belgian counterparts. The teaching profile account has not gained support by the results of this study. Dutch teachers were more successful on both identifying original pairs as well as on motivating their choice than Hebrew teachers. Thus children’s spelling knowledge, at least of the type pinpointed by our tests, does not directly reflect their teachers’ understanding of the underpinnings of spelling patterns in either language. Moreover, teachers show a different pattern of responses, in some cases the mirror image of what children’s responses reveal: On identifying the conditions without morphological cues both teacher populations perform equally well, while Hebrew teachers do worse than Dutch teachers on identifying morphologically conditioned items – the reverse picture of children’s spelling scores. And on motivating their choice in conditions with morphological cues, Dutch teachers again do better than Hebrew teachers – again the reverse picture of children’s spelling

30 scores. Thus it cannot be the case that teaching Hebrew and Dutch spelling is directly responsible for children’s spelling patterns. A third finding relates to the two teacher populations – TTC and UNI students, and it does not support the teaching profile account either. On both identifying the test pairs and motivating their linguistic underpinnings, the two Dutch-speaking populations are similar, but TTC Hebrew students do worse than do UNI students. Current literature indicates that TTC trainees in Israel are not equal in academic potential to UNI students (Ayalon & Yogev, 2000), which may explain their differential performance. Thus those Israeli teachers who are in contact with gradeschool children at the time when they are internalizing spelling patterns are those who are least able to detect and explain these spelling patterns, while those Belgian teachers who show high ability in detecting and motivating spelling patterns do not seem to affect poor spelling knowledge in children. Can the typological account provide an explanation for the current results in addition to explaining the differences in children’s spelling? At first glance it seems strange that teachers’ metalinguistic knowledge of spelling patterns is a mirror image of children’s performance. If Israeli children’s strategies are morphological, why is it the case that Israeli teachers do not form appropriate morphological hypotheses about Hebrew spelling or at least properly detect pairs of items linked through morphology; and why is it the case that Belgian teachers are so well-versed in the morphological underpinnings of Dutch spelling that they are so successful at both detecting spelling test pairs as well as explicitly stating in what ways they belong together? Three points need to be clarified here. One is the difference between the kinds of knowledge required from children and teachers in this study. A second point is the

31 mirror image relationship of the results. And a third relates to the discrepancy between explicit knowledge and its procedural implications. It seems that sensitivity to language typology is an implicit kind of knowledge that directs linguistic behavior in certain directions from early on and in a sense affects this behavior more than explicit teaching. A confluence of cues results in linguistic patterns that are internalized by children during language acquisition, and children find natural what is natural in their language, even when it looks complex under a metalinguistic analysis. But explaining these patterns is the linguist’s job, and requires both specialized knowledge about levels of representation in language – phonetics, abstract phonology, morphological and syntactic constructs – as well as the ability to identify linguistic regularities and systems. This kind of explicit analysis does not come naturally to teachers, even those with language training. Thus a spelling test with morphological cues may be easier for Hebrew-speaking children who have been dealing with complex morphology from their first words than for Dutch-speaking children for whom syntax is much more salient. But their teachers, acting as amateur linguists, find it easier to explicitly formulate the simpler spelling rules of Dutch than the complex spelling patterns of Hebrew. Moreover, Dutch spelling rules are explicitly taught to Dutch spellers at all levels of literacy, whereas the understanding of Hebrew spelling patterns requires specialized knowledge in abstract Hebrew phonology and historical Hebrew periods, which is restricted to theoretical discussions at university level alone. Finally, knowledge of spelling rules by teachers does not ensure successful performance by learners. Spelling, as we claimed in Gillis & Ravid (2000), is a linguistic ability which is guided by the same principles of natural mother tongue acquisition; and mother tongue is not taught at school nor is shaped by explicit rule

32 learning but rather evolves following its specific typology. Learning about spelling rules resembles second or foreign language learning which uses alternative mechanisms, drawing especially on people’s problem-solving capacities and characterized by a discrepancy between stating linguistic rules and implementing them. For example, adding the 3rd person singular in present-tense English verbs is a very simple rule to formulate, taught and re-taught to second language learners, yet non-native adults do not apply the rule even if they know about it (De Keyser, in press). The discrepancy between the cognitive apparatus used to implement spelling rules and the cognitive apparatus used to learn about them may explain why teachers’ levels of knowledge about spelling patterns are not causally related to students’ spelling performance.

33 Appendix

Test items used in Experiment 1 and 2

Dutch test items Condition 1: betovert ‘bewitches’ – betoverd ‘charmed’ versiert ‘decorates’ - versierd.’decorated’ betekent ‘means’ – betekend ‘meant’ vertoont ‘shows’ – vertoond ‘shown’

Condition 2: agent ‘policeman’ - arend ‘eagle’ tomaat ‘tomato’ – sieraad ‘ornament’ fazant ‘pheasant ‘ - verband ‘bandage’ taart ‘cake’ – paard ‘horse’

Condition 3: verplichte ‘obligatory’ - verplichtte ‘forced’ verwachte ‘expected’ - verwachtte ‘expected’ verroeste ‘rusty’ roestte ‘(got) rusty’ verlichte ‘lighted’ - verlichtte ‘(were) lit’

Condition 4: krijsen ‘scream’ - reizen ‘travel’

34 lijnen ‘lines’ - kleine ‘small’ zwijnen ‘pigs’ - treinen ‘trains’ pijn ‘hurts’ - refrein.’chorus’

Hebrew test items Condition 1: ve-ala ‘and-went-up’ - be- vehala ‘in-fright’ va-ir ‘and-town’ - u-vahir ‘and-light’ ve-red ‘and-go-down’ – véred ‘rose’ va-adas&a ‘and-(a)-lens’ – va’adat – kis&ut ‘committee-(for)-decoration’

Condition 2: dérex ‘road’ – kérax ‘ice’ oréxet ‘setting the table, Fm’ – oráxat ‘guest, Fm’ holex ‘walks’ - s&oléax ‘sends’ nix s&al ‘fails’ – nexs&av ‘is considered’

Condition 3: tarim ‘lift,Imp,2nd,Sg,Masc’ – ta’im ‘tasty’ kas&ot ‘hard, Fm,Pl’ – mas&ot ‘oar’ mehirut ‘speed’ – karut ‘cut down, PP’ tapil ‘drop,Imp,2nd,Sg,Masc’ – takin ‘in order’

Condition 4: hirgiz 'annoyed' - hirgiz 'annoyed'

35 min ‘from’ – min ‘gender’ lispor 'to count' - lipol 'to fall' migrash 'empty lot' - nigash 'came nearer'

36 References

Appel-Mashraki, Y. 2001. Prosodic reading, text comprehension and metamorphological ability in 4th graders. MA thesis, School of Education, Tel Aviv University, Tel Aviv. Ashkenazi, O., and D. Ravid. 1998. Children’s understanding of linguistic humor: an aspect of metalinguistic awareness. Current Psychology of Cognition 17:367387. Ayalon, H., and A. Yogev. 2000. A window to the academic dream: Stratification and diversity in the expansion of Israeli higher education. Paper read at Conference of the Research Committee on Social Stratification and Mobility of

the

International Sociological Association, August 2000, at Calgary, Canada. Bialystok, E. 1986. Factors in the growth of linguistic awareness. Child Development 57:498-510. Carlisle, J., and D. Nomanbhoy. 1993. Phonological and morphological awareness in first graders. Applied Psycholinguistics 14:177-195. De Keyser, R. in press. The robustness of critical period effects in second language acquisition. Studies in Second Language Acquisition. Gillis, S., and G. De Schutter. 1996. Intuitive syllabification: universals and languagespecific constraints. Journal of Child Language 23:487-514. Gillis, S., and D. Ravid. 2000. Effects of phonology and morphology in children’s orthographic systems: a cross-linguistic study of Hebrew and Dutch. In Proceedings of the 30th Annual Child Language Research Forum, edited by E. V. Clark. Stanford: CSLI.

37 Gillis, S., and D. Ravid. in press. Language-specific effects on the development of written morphology. In Proceedings of the 9th International morphology Meeting 2000, edited by W. U. Dressler. Amsterdam: Benjamins. Gombert, J. 1992. Metalinguistic development. New York: Harvester Wheatsheaf. Goswami, U. 1999. The relationship between phonological awareness and orthographic representation in different orthographies. In Learning to read and write: A crosslinguistic perspective, edited by M. Harris and G. Hatano. Cambridge: Cambridge University Press. Henry, M. 1993. Morphological structure: Latin and Greek roots and affixes as upper grade code strategies. Reading and Writing 5:227-241. Jones, N. 1991. Development of morphophonemic segments in children’s mental representations of words. Applied Psycholinguistics 12:217-239. Karmiloff-Smith, A. 1992. Beyond modularity: A developmental perspective of cognitive science. Cambridge: MIT Press. Karmiloff-Smith, A., J. Grant, K. Sims, M. Jones, and P. Cuckle. 1996. Rethinking metalinguistic awareness: representing and accessing knowledge about what counts as a word. Cognition 58:197-219. Koriat, A., N. Greenberg, and Y. Goldshmid. 1991. The missing-letter effect in Hebrew: Word frequency or word function? Journal of Experimental Psychology: Learning, Memory and Cognition 17:66-80. Lewis, D., and J. Windsor. 1996. Children’s analysis of derivational suffix meanings. Journal of Speech and Hearing Research 39:209-216. Lexikologie, Instituut voor Nederlandse. 1997. Woordenlijst Nederlandse taal. Den Haag: Sdu Uitgevers.

38 Nippold, M. 1998. Later Language development: The school-age and adolescent years. 2nd Edition. Austin: PRO-ED. Nippold, M., L. Uhden, and I. Schwarz. 1997. Proverb explanation through the lifespan: A study of adolescence and adults. Journal of Speech, Language, and Hearing Research 40:245-253. Nunes, T., P. Bryant, and M. Bindman. 1997. Morphological spelling strategies: developmental stages and processes. Developmental Psychology 33 (637-649). Pacton, S., P. Perruchet, M. Fayol, and A. Cleeremans. in press. Implicit Learning out of the lab: The case of orthographic regularities. Journal of Experimental Psychology: General. Ravid, D. 1995. Language change in child and adult Hebrew: A psycholinguistic perspective. Oxford: Oxford University Press. Ravid, D. in press a. Learning to spell in Hebrew: Phonological and morphological factors. Reading and writing. Ravid, D. in press b. A developmental perspective on root perception in Hebrew and Palestinian Arabic. In The processing and acquisition of root-based morphology, edited by Y. Shimron. Amsterdam: Benjamins. Ravid, D., and A. Malenky. in press. Awareness of linear and nonlinear morphology in Hebrew: a developmental study. First Language. Ravid, D., and L. Tolchinsky. 2000. Developing linguistic literacy: A comprehensive model. In Working papers in "Developing literacy across genres, modalities, and languages", Vol. III., edited by M. Aparici, N. Algerich, J. Perera, E. Rosado and L. Tolchinsky. Barcelona: Institut de Ciències de l’Educació, Universitat de Barcelona.

39 Sandra, D., S. Frisson, and F. Daems. 1999. Why simple verb forms can be so difficult to spell: the influence of homophone frequency and distance in Dutch. Brain and Language 68:277-283. Smith, M. 1998. Sense and sensitivity: An investigation into fifth grade children’s knowledge of English derivational morphology and its relationship to vocabulary and reading abilit, Graduate School of Education, Harvard University. Totereau, C., M. Theverin, and M. Fayol. 1997. The development of the understanding of number morphology in written French. In Learning to spell, edited by C. Perfetti, M. Fayol and L. Rieben. Hillsdale: Erlbaum. Treiman, R. 1993. Beginning to spell: A study of first-grade children. New York: Oxford University Press. Treiman, R., and M. Cassar. 1996. Effects of morphology on children’s spelling of final consonant clusters. Journal of Experimental Child Psychology 63:141-170. Treiman, R., and M. Cassar. 1997. Spelling acquisition in English. In Learning to spell, edited by C. Perfetti, M. Fayol and L. Rieben. Hillsdale: Erlbaum. Treiman, R., M. Cassar, and A. Zukowski. 1994. What types of linguistic information do children use in spelling? The case of flaps. Child Development 65:13101329. Treiman, R., A. Zukowski, and E. Richmond-Welty. 1995. What happened to the "n" of sink? Children’s spelling of final consonant clusters. Cognition 55:1-38. Wilson, S., L. Shulman, and A. Richert. 1987. ‘150 different ways’ of knowing: Representation of knowledge in teaching. In Exploring teachers’ thinking, edited by J. Calderhead. London: Cassell Education.

40 Wysocki, K., and J. Jenkins. 1987. Deriving word meanings through morphological generalization. Reading Research Quarterly 22:66-81.

41 Figure 1: Percentage correct scores per grade on the Dutch and the Hebrew spelling tests

100

50

25 Dutch Hebrew

Grade6

Grade5

Grade4

Grade3

Grade2

0 Grade1

%Correct

75

42 Figure 2: Interaction of language and condition (C1..C4 stand for the experimental conditions, C1 = + Morphological function, +M, and + Morphophonological recoverable, +R), C2 = -M and +R, C3 = +M and –R, C4 = -M and –R)

100

75

C2 50 C3 C4 25

Hebrew

0 Dutch

% Correct

C1

Language

43 Figure 3: Percentage correct score in the children’s and the teachers’ tests (+M = target has morphological function, -M = target has no separate morphological function)

Children’s results 100

%Correct

75

+M 50 -M

25

Dutch

Hebrew

0

Language

Teachers’ results

100

+M 50 -M

25

Hebrew

0 Dutch

% Correct

75

Language

44

45 Notes

1

We represent the historical voiced Hebrew pharyngeal fricative AYIN (still

pronounced by some speakers of the Sephardi dialect) by the numeral 9. 2

Representing the historical emphatic coronal stop.

3

The diphthong /Eê/ is spelled as when it derives historically from Proto-

Germanic /Ai/ and as when it derives from long /i/.