Two types of developmental surface dysgraphia: to bee but not to bea

ABSTRACT

We report on two types of developmental surface dysgraphia. One type, exhibited by 8 participants, is orthographic lexicon surface dysgraphia, which involves an impairment in the orthographic output lexicon, leading to nonword phonologically-plausible misspellings. The other type, shown by 3 participants, is disconnection surface dysgraphia. In this type, the orthographic output lexicon is disconnected from the semantic system and from the phonological input lexicon, but still contributes to spelling via support to the orthographic output buffer, resulting in mainly lexical phonologically-plausible misspellings (writing be as “bee” but not “bea”).

The specific localization of the impairment in spelling, in the lexicon or in its connections, allowed us to examine the question of one or two orthographic lexicons; four participants who had a deficit in the orthographic output lexicon itself in writing had intact orthographic-input-lexicon in reading. They made surface errors in writing but not in reading the same words, supporting separate input and output orthographic lexicons.

KEYWORDS:

• Surface dysgraphia
• orthographic lexicon
• Hebrew
• developmental dyslexia
• surface dyslexia

1. Introduction

Surface dysgraphia is a spelling deficit that stems from an impairment in the lexical route for spelling, resulting in spelling via the phoneme-to-grapheme conversion (PGC) route. This may cause incorrect spelling when the orthographic form of the target word is not fully predictable from its phonological form. Such unpredictability can occur, for example, when a word can be spelled via phoneme-to-grapheme conversion in more than a single way because it contains a phoneme that can be converted to one of two letters (for example, the phoneme /s/ can be spelled in English with c or s), or when the word is irregular (i.e., spelled in a way that does not conform to the PGC rule, such as words with silent letters or geminates with no corresponding phonological doubling, such as listen and tomorrow) (Brunsdon et al., 2005; de Partz et al., 1992; Ellis, 1993; Kohnen, Nickels, Brunsdon, et al., 2008, 2008; Krajenbrink et al., 2015; McCloskey & Rapp, 2017; Romani et al., 1999; Tainturier & Rapp, 2001; Temple, 1985; Weekes & Coltheart, 1996).

The deficit in the lexical route in surface dysgraphia is traditionally conceived as a deficit in the orthographic lexicon (e.g., de Partz et al., 1992; Ellis, 1993; Hatfield & Patterson, 1983; Romani et al., 1999). However, a deficit in the lexical route can also result from disconnection – inability to access the orthographic lexicon from the semantic system and the phonological input lexicon.

In the current study we present individuals with these two different subtypes of surface dysgraphia, in the orthographic lexicon or in the connections to it, each resulting from a deficit in the lexical route in one of the two loci.

The participants in our study have developmental surface dysgraphia. For the sake of clarity, in this paper we use the terms “impairment” and “deficit” to describe the current status of cognitive components, both for acquired and for developmental dysgraphia (or dyslexia). In the context of developmental cases this end state may result from impaired development of the components. This deficit affecting development may be similar to that witnessed in acquired cases, with which the children are born (e.g., impaired neural substrates for an orthographic lexicon).

In the first experimental part of this article, we describe the subtypes of surface dysgraphia and their different error patterns. In the second part we proceed to explore the relation between subtypes of surface dysgraphia and subtypes of surface dyslexia, in order to shed light on the question of whether there is a single orthographic lexicon or two separate orthographic lexicons for input and output. After we have identified the locus of impairment in the lexical route in the orthographic lexicon or in the access to it, we examined whether we can identify individuals with a deficit in the orthographic lexicon in writing but not in reading or vice versa.

In the following section we briefly describe the model we assume for spelling and describe surface dysgraphia. We then present the ways to distinguish between the subtypes of surface dysgraphia that are predicted by the spelling model.

1.1. The spelling model

The spelling model, portrayed in Figure 1 based on much cognitive neuropsychological work, mainly on acquired dyslexias (Ellis, 1993; Ellis & Young, 1996; Miceli & Capasso, 2006; Rapp, 2002; Shallice, 1988; Tainturier & Rapp, 2001, 2003), includes two routes to spell a word, a lexical route and a sublexical route. The lexical route uses the orthographic output lexicon, which is a store of orthographic representations of words (or word bases and roots). This lexicon can be accessed from the semantic system, in free writing, and from the phonological input lexicon, in writing to dictation. The graphemes selected from the lexicon and their relative positions are held for a short time in the orthographic output buffer, an orthographic short term memory store, and are then transmitted to the allographic stage, where the abstract letter forms are selected, and then to the motor execution of writing, in which the specific motor patterns for the specific letters are activated in order to perform the hand movements required for writing (these later stages are not depicted in Figure 1). Only items that exist in the orthographic output lexicon can be spelled via this lexical route, and hence nonwords and new words cannot be spelled through this route.

Figure 1. The single word spelling model (presented within a broader lexical model also for naming and reading).

represents the locus of impairment in orthographic lexicon surface dysgraphia;

represents the loci of impairment in disconnection surface dysgraphia

The other route, the sublexical route, uses phoneme-to-grapheme conversion rules to convert phonological sequences in the phonological input buffer to orthographic sequences in the orthographic output buffer. The orthographic products of the phoneme-to-grapheme conversion are transmitted from the orthographic output buffer to the allographic and motor stages. This sublexical route is the only possible route for spelling new words and nonwords – letter sequences that do not exist in the orthographic lexicon.

1.2. Surface dysgraphia: one source of deficit or more?

Various types of dysgraphia exist, each type resulting from a selective impairment to a different part of the spelling process (Ellis, 1993; McCloskey & Rapp, 2017; Miceli & Capasso, 2006; Tainturier & Rapp, 2001; Yachini & Friedmann, 2022). A deficit to the lexical route results in surface dysgraphia (Angelelli et al., 2004, 2010; Beauvois & Dérouesné, 1981; Brunsdon et al., 2005; Coltheart et al., 1983; Douklias et al., 2009; Ellis, 1993; Goodman & Caramazza, 1986; Goulandris & Snowling, 1991; Hanley et al., 1992; Hanley & Gard, 1995; Kohnen, Nickels, Brunsdon, et al., 2008, 2008; Romani et al., 1999; Tainturier & Rapp, 2001; Temple, 1984, 1985; Weekes & Coltheart, 1996).

Individuals with surface dysgraphia are forced to spell through the sublexical route when the lexical route is unavailable to them. Spelling words through the sublexical route may result in incorrect spelling of unpredictable words, including words that have more than a single possible conversion from phonemes to graphemes, and words that do not obey standard phoneme-to-grapheme conversion rules (Baxter & Warrington, 1987). This results in phonologically-plausible misspellings – which follow the phoneme-to-grapheme conversion rules but do not correspond to the target lexical item, including regularization errors (writing lisn instead of “listen”, and nife for “knife”), and homophonic letter substitution errors (writing jem instead of “gem”). These errors may result in nonwords, like jem and lisn, or they may create other existing words, homophones (writing no instead of “know”) or potentiophones (i.e., words that are not homophones but, when spelled sublexically, yield other existing words such as writing none instead of “known”¹). The sublexical conversion route uses phoneme-to-grapheme conversion rules and no lexicon. Errors resulting from sublexical spelling may create existing words either by accident, when the result of the conversion happens to be identical to an existing word, or using later support from the orthographic output lexicon, as we describe below.

These phonologically-plausible misspellings are also found in the early spelling of typically-developing non-dyslexic children, when trying to spell words that are not stored in their orthographic output lexicon yet (for example, Yachini & Friedmann, 2022, reported on 281 typically-developing Hebrew-writing children, the 3rd graders made such errors on 25% of the words they wrote (SD = 13.3%), and the 4th graders made 18% of the words (SD = 11.7%)).

Surface dysgraphia is traditionally ascribed to a deficit in the orthographic output lexicon, marked in Figure 1 (Caramazza et al., 1987; Goodman & Caramazza, 1986; Rapp et al., 2016). However, a look at the spelling process (depicted in the model (2) in Figure 1) suggests that an impairment in the orthographic lexicon is not the only impairment that can cause spelling words through the sublexical route. A deficit in the lexical route can also result from disconnection (arrows marked in Figure 1)– inability to access the intact orthographic lexicon from the semantic system and the phonological input lexicon (see discussions of possible loci of impairment in the lexical route giving rise to surface dysgraphia in Hepner et al., 2017; Kohnen, Nickels, Brunsdon, et al., 2008, 2008; Krajenbrink et al., 2015; Hanley & Sotripoulos, 2018). In this case, the orthographic lexicon itself is intact, so spelling can proceed via a bypass route that would enjoy the support of the entries in the orthographic lexicon: either through a connection from the phonological output lexicon to the orthographic lexicon, or by feedback: spelling through the phoneme-grapheme conversion route, ending up with a string of letters in the orthographic output buffer, which will receive support from the intact orthographic output lexicon (see McCloskey et al., 2006 for discussion of bidirectional activation between the grapheme level – the output buffer, and the lexeme level – the orthographic output lexicon).

1.2.1. Predictions for an impairment in the orthographic lexicon vs. disconnection of an intact orthographic lexicon

Spelling that involves an intact but disconnected orthographic lexicon is predicted to differ from spelling that does not involve an impaired orthographic lexicon, because when the lexicon is involved, lexical effects are expected. As a result three main differences are predicted between the two types of surface dysgraphia:

1. Lexicality of the erroneous responses. In the disconnection type of surface dysgraphia, the intact orthographic lexicon is available to consult spelling, even if it is not accessible from the semantic system. In this case, the erroneous responses are expected to be mainly existing words, even if they are not the target words but rather homophones or potentiophones of the target (writing be but not bea instead of bee). In contrast, spelling through the sublexical route without the support of the orthographic lexicon will lead to the production of non-lexical responses (as well as accidental lexical ones). Without feedback from the lexicon, “bee” may be spelled as be or as the non-existing word bea (it may also be spelled as bee, by accident or according to the frequency of the mapping rule).

2. Target words that result in surface dysgraphia errors. Individuals with surface dysgraphia whose orthographic output lexicon contributes to spelling should make fewer errors in target words for which no phonologically-plausible misspelling creates an existing word (e.g., car, supermarket). This is because in such words their orthographic output lexicon would only support spelling of real words so it would support the only possible lexical result, the target word. They are expected to make more surface errors in spelling target homophonic and potentiophonic words, for which a phonologically plausible misspelling may create other existing words, that the lexicon would support. This is in contrast to individuals with surface dysgraphia whose deficit is in the orthographic output lexicon itself, who are not expected to show differences between words with and without possibility for lexical surface error.

3. Frequency effect. If an intact but disconnected orthographic output lexicon participates in spelling, the effects that operate on it should affect spelling, and the more frequent word in the lexicon would support the string in the buffer. Therefore, in the case of disconnection surface dysgraphia, we expect an effect of the frequency of the target words on surface errors: they are expected to make fewer surface errors in high-frequency target words.

   In the case of impairment in the orthographic output lexicon itself, the prediction regarding frequency is unclear. Frequency may not have an effect on spelling, if the lexicon is not consulted in most cases, or it may have an effect, depending on the nature of the lexical impairment.² In previous cases of surface dysgraphia, indeed word frequency effect was not consistently found. Luzzatti et al. (1998) examined the effect of word frequency on spelling in 18 individuals with surface dysgraphia and found that fewer than half of them showed a word frequency effect. Some single case studies which, according to the non-lexical surface errors reported seem to be cases of impaired orthographic lexicon, showed no effect of frequency on spelling (Hanley et al., 1992), and some did (de Partz et al., 1992; Temple, 1985; and Romani et al., 1999). Therefore, our prediction regarding frequency effect on spelling in surface dysgraphia is asymmetric: we predict word frequency effect for cases with an intact but disconnected lexicon, and we do not specify a prediction for cases with an impaired lexicon.

1.2.2. Previous cases of developmental surface dysgraphia and discussion of their locus of deficit

A developmental case of MC, a 12-year-old boy reported by Brunsdon et al. (2005) may be a case of disconnection surface dysgraphia. On the basis of his irregular and homophone word spelling, Brunsdon et al. suggested that MC’s deficit lay in the access from the semantic system to the orthographic output lexicon, and either in the access from the phonological output lexicon to the orthographic output lexicon or in the orthographic output lexicon itself. Brunsdon et al. based this conclusion on the finding that although MC wrote many nonwords when he tried to write existing irregular and homophonic words, approximately a fifth of his errors were existing words (including clear indications of knowledge of the irregular spelling of some homophones). Although not a clear-cut case, because he made 80% non-lexical errors, Brunsdon’s case suggests that impairments to the lexical route do not need to be in the orthographic output lexicon itself.

Angelelli et al. (2010) also mention the possibility of a deficit in access to the orthographic lexicon leading to surface dysgraphia. Sotiropoulos and Hanley (2017) tried to look at this issue through the performance in reading under the assumption of a single orthographic lexicon for input and output. They suggested that individuals who have surface dysgraphia with good lexical decision of pseudohomophones (in reading), judging correctly that “kar” is not a word, may be impaired in the access to the orthographic output lexicon from semantics or from the phonological input lexicon. Hepner et al. (2017), when discussing the option of impaired access to the orthographic output lexicon rather than a deficit in the orthographic output lexicon itself, conclude that it is difficult to test an assumption that lexical-orthographic representations are intact in cases of impairment attributed to access deficits. In the current study we tackle exactly this challenging task.

1.3. Unpredictability in spelling Hebrew

We examine these predictions in Hebrew, a fantastic testing ground for surface dysgraphia. The spelling of most of the words in Hebrew is unpredictable: most Hebrew words cannot be spelled unambiguously and correctly through grapheme-to-phoneme conversion. The sources of this unpredictability are polygraphy and irregularity. Firstly, there is a great deal of polygraphy in PGC: 6 consonant phonemes can be converted to one of two or three different letters, two vowels can be converted to one of two or three vowel letters, and one vowel can be converted to a letter or to a zero grapheme. This creates a situation in which 9 phonemes have polygraphic conversions to letters, and 13 letters are polygraphic conversions of phonemes. Secondly, many words are not only unpredictable but irregular, namely, violating conversion rules. Irregularity in Hebrew stems from three sources (see the Appendix for a list of PGC rules and their violations): a) in cases of very biased polygraphy, irregular words are those that are spelled with the far-less-frequent conversion of the phoneme. For example, in Hebrew the two letters that represent the phoneme /s/ differ in conversion frequency: one of them (ס) is far more frequently associated with /s/ than the other (ש). This makes words with /s/ spelled with the less frequent letter irregular. The same is true for irregular conversion of vowel phonemes to another vowel letter than dictated by the PGC rule. b) Another source of irregularity relates to silent letters – zero phonemes that are converted to a letter (except for three words, all the silent letters in Hebrew are vowel letters). c) Another source of irregularity in Hebrew refers to the orthographic representation of vowels: in the end of the word, all the Hebrew vowel phonemes are consistently represented orthographically with a corresponding vowel letter; in the middle of the word, the vowels /o/ and /u/ are consistently represented, /a/ and /e/ are consistently not represented orthographically, and /i/ is often represented but not always (Hebrew words never begin with a vowel phoneme, but rather with a glottal stop). Words whose spelling violates these rules are irregular: words ending with a vowel phoneme which is not marked with a vowel letter; vowels in the middle that should be marked with a vowel letter but are not; and vowels in the middle of the word that are usually not marked with a vowel letter, but are marked with a letter in the target word. See the Appendix for a list of irregularities in Hebrew.³

These characteristics of the Hebrew orthography cause spelling via the sublexical route to be error-prone, and make surface dysgraphia very easy to detect. In fact, in an analysis we made of the 20,000 most frequent words in Hebrew (5000 most frequent nouns, verbs, adverbs, and adjectives, from Linzen’s database, 2009) we found that 94.9% of these words are unpredictable for spelling in one of the ways we defined above.

A specific type of unpredictable words, which are especially relevant for our current study, and which are quite common in Hebrew, are potentiophones: words that, when spelled via the sublexical route can be spelled as another existing word. i.e., words for which a phonologically-plausible misspelling may result in other words, such as spelling the target word “know” sublexically as its potentiophone now.

Surface dysgraphia is the most common dysgraphia in Hebrew: 186 of the 427 participants with developmental dysgraphia tested in Yachini and Friedmann (2022) had selective surface dysgraphia (and additional 118 had surface dysgraphia on top of other types of dysgraphia, so their surface dysgraphia may have been secondary to their other dysgraphia).

2. General methods

2.1. Participants

The participants with developmental surface dysgraphia were native Hebrew-speakers, 11 adults and adolescents, 5 males and 6 females. Background information about each of the dysgraphic participants is summarized in Table 1. None of the participants sustained brain damage and none of them had neurological conditions (other than their dysgraphia / dyslexia), so they were all developmental cases. Our participants are all adolescents or adults, so their deficit cannot be ascribed to their still being at the stage of spelling acquisition.

Table 1. Background description of the participants.

Participant	Age	Gender	Education (years)
MSH	15;0	M	9
EI	15;5	M	9
LIO	17	M	11
NAM	25	F	12
LIZ	31	F	12
SHY	14;0	F	8
HEN	14;6	F	8
RAM	17	M	11
OR	17	M	11
AN	24	F	12
HIL	27;9	F	12

2.2. General procedure

Each participant was tested individually in a quiet room. During the writing testing sessions, the participants were instructed not to erase their writing if they thought they made an error. Instead, they were requested to put their first response in parentheses and try again. Only the first writing response was used for all our analyses (namely, corrections that could occur because the participant read their own writing and corrected accordingly were not included in the analyses). The experimenter wrote down every response that differed from the target. No time limit was imposed during testing, and no response-contingent feedback was given by the experimenter, only general encouragement. The participants were told that whenever they needed a break, they could stop the session or take a break.

2.3. Data analysis

The performance of each dysgraphic participant was compared to the performance of the control group using the Crawford and Howell’s (1998; Crawford & Garthwaite, 2002, 2012) t-test. An alpha level of 0.05 was used in all analyses. Within-participant comparisons between performance in two conditions were conducted using a chi-square test.

3. Writing assessment: establishing surface dysgraphia

We used dictation and written naming tasks to identify participants with surface dysgraphia, namely, individuals who made spelling errors that result from spelling via the sublexical route instead of via the lexical route, at a rate that was significantly higher than controls of their age.

3.1. Writing to dictation – screening tests

The initial writing test, used to examine whether the participants had surface dysgraphia, was the TILTAN screening writing test (Friedmann et al., 2007). This test was constructed to identify the various types of dysgraphia. The test includes 76 Hebrew single words of 2–11 letters (M = 5.2 letters, SD = 2.1). The words were selected so that they would enable the detection of various types of dysgraphia (as they include long words, words with double letters, unpredictable words, function words, abstract words, migratable words, and words that create other existing words by neglect of the left or right side). In addition, words were selected to allow for the assessment of various effects on spelling (word length, predictability, and lexical category).

All 76 words in the screening test included at least one polygraphic phoneme (a phoneme that has two or more possibilities for conversion into a letter via the sublexial route), so they were all sensitive to surface dysgraphia; 22 of the words were potentiophones (i.e., target words that, when spelled via the sublexical route can be spelled as another existing word).

The second part of the screening test was a nonword writing test, including 28 non-words. Nonwords are expected to be spelled well by individuals with surface dysgraphia. (Notice that in Hebrew every orthographically-licit letter sequence is a possible word, and therefore there is no difference between nonwords and pseudowords.) Nine of the 11 participants were tested with the nonword part of the screening test in addition to the word writing part.

The participants were further tested using an additional surface dysgraphia word list that included 328 unpredictable words, which cannot be spelled unambiguously and correctly through grapheme-to-phoneme conversion. Of this list, 326 words were polygraphic, and 45 were irregular (according to the definition of irregularity in Section 1.3 above and the Appendix). Fifty five of the target 328 words were potentiophones.

The dictated words also allowed for the detection of other types of dysgraphia. All the 404 words (in the screening and the irregular word lists) also allowed for the detection of vowel dysgraphia (manifested in vowel letter errors), letter omission, addition, migration and non-homophonic letter substitution characteristic of orthographic output buffer deficit and neglect dysgraphia, and 390 words allowed for voicing errors.

Results

The performance of each experimental participant was compared to the performance of a control group of 140 adolescents aged 13–15 years (128 adolescents in the screening test control group, Yachini & Friedmann, 2016, 2018; 12 participants in the dedicated irregular word list control group). This age group was selected for the control group because it was age-matched to the youngest dysgraphic participants. The rationale was that if the older participants make significantly more surface dysgraphia errors than even the younger children, they should definitely be considered to have surface dysgraphia.

The performance of each of the participants in writing the TILTAN word reading test and the additional unpredictable word list (76 words in the screening list and 328 words in the unpredictable list, with a total of 404 words) is summarized in Table 2. Each of the participants made significantly (p < .0001) more errors in spelling compared to the control group. Importantly, as can be seen in Table 2, the most prominent errors for each of these participants were surface dysgraphia errors – words spelled with the incorrect yet phonologically plausible letter in case of polygraphy, and regularization errors in cases of irregular target words as described in 1.3 above. Below we refer to errors of this type as “surface errors”. Some of these surface errors created other existing words (e.g., writing sublexically in a way that results in the potentiophone of the target words, such as “now” instead of “know”); other surface errors resulted in nonwords. Other, non-surface, types of errors were relatively few, even though the target words allowed for the whole range of dysgraphias. When a single response included multiple error types, each type of error was counted separately.

Table 2. Writing 404 words to dictation: % correct and number of errors of each type.

		Number of errors of the various types
Participant	% Correct	Surface dysgraphia errors	Orthographic buffer errors	Voicing errors	Vowel letter errors	Other errors
MSH	86%*	54
EI	81%*	71	4	1
LIO	84%*	63
NAM	82%*	68	2		2	1
LIZ	75%*	90	9	2	4
SHY	74%*	84	8	1	2
HEN	82%*	69	5
RAM	62%*	154		1
OR	75%*	99	1
AN	84%*	61	4
HIL	86%*	53	3		2

*Significantly below the control group, *p < .0001. Orthographic buffer errors: consonant letter omissions, additions, migrations, and non-homophonic letter substitutions. Voicing errors: substitution of a consonant with its voiced or unvoiced counterpart. Vowel letter errors: omission, addition, migration, and non-homophonic substitution of vowel letters.

In the nonword writing part of the test, the participants performed in a way that is typical of surface dysgraphia: 8 of the 9 participants who participated in the nonword writing test performed well, not differently from the controls. They spelled correctly 89%−100% of the nonwords, with an average of 95% (seven participants had 0–2 errors, p > .32, one had 3 errors, one of them had allograph error, p > .10). LIZ wrote only 82% of the nonwords correctly, as she made 5 buffer errors. Nonword spelling for all of them was within 1.2 SD of the control group mean (which was 95.3%, SD = 4.9%, Yachini & Friedmann, 2022).

This finding also further supports our conclusion that they did not have orthographic output buffer dysgraphia (possibly except LIZ), voicing dysgraphia, vowel dysgraphia, deep dysgraphia, or phonological dysgraphia.

3.2. Do surface dysgraphia errors occur across spelling tasks and input modalities?

Surface dysgraphia should cause errors not only in writing to dictation, in which the input to the orthographic output lexicon comes from the phonological input lexicon, but also in written naming, in which input comes from the conceptual-semantic system. To make sure that the errors indeed result from the lexical spelling route and not from a deficit in phonological input, and to examine the semantic-orthographic connection, we assessed the participants’ writing from semantics. We used two tasks with semantic input: sentence completion and written picture naming.

Written naming: sentence completion and picture naming. We selected 50 words from the list of 328 unpredictable words that were dictated (Section 3.1) and used them again in tasks of writing from semantics. Of these words, 44 words were tested in a sentence completion task (“Each hand has five … ”). The sentence was presented auditorily and the participant was asked to write down the word that completes the sentence. The spelling of 6 other potentiophonic, morphologically simple words was tested in a written picture naming test in which the participant saw a picture of an object for unlimited time and was requested to write its name. (The test initially included 25 objects for written naming but 19 were excluded from the current analysis because they were morphologically complex).

Results

The rate of surface dysgraphia errors that each participant made in spelling from semantics is summarized in Table 3, and is compared to the participant’s spelling of the same words to dictation. All the participants made surface errors in written naming, and none of the participants showed a significant difference between the surface error rates in the two input modalities. (The surface error rate here is relatively low in both tasks, and lower than the words in Table 2 and in Table 6 below, probably because of the properties of the target words, many of which included a polygraphic phoneme but the letter in the word was the more frequent choice for the phoneme).

Table 3. Writing from semantics (sentence completion and written naming) and writing to dictation of the same 50 unpredictable words: % words written with surface dysgraphia errors.

Participant	Writing from semantics (naming)	Writing from phonology (dictation)
MSH	8%	8%
EI	4%	2%
LIO	2%	10%
NAM	12%	10%
LIZ	6%	6%
SHY	10%	12%
HEN	8%	4%
RAM	44%	40%
OR	10%	16%
AN	8%	6%
HIL	8%	16%

4. Are there types of surface dysgraphia?

Now that we have established that our participants have surface dysgraphia, we can proceed to the main challenge of this study, which was to examine whether two types of surface dysgraphia can be characterized. As we explained in the Introduction, two different loci of impairments in the lexical route can force the writer to spell via the sublexical route and show surface dysgraphia. One is a deficit in the orthographic output lexicon, the other is a deficit that keeps the orthographic lexicon intact but results from a disconnection between the semantic lexicon (and possibly also the phonological input and output lexicons) and the orthographic output lexicon (disconnection surface dysgraphia). In the case of disconnection surface dysgraphia, spelling can still benefit from the intact (yet disconnected) lexicon: either by spelling through a connection from the phonological output lexicon to the orthographic lexicon, or by feedback: spelling through PGC conversion, ending up with a string of letters in the orthographic output buffer, which receives support from the intact orthographic output lexicon.

An analysis of these two impairments in the lexical route suggests three possible ways to distinguish between the two subtypes (see the predictions in detail in section 1.2.1). The first should be the difference in the lexicality of the erroneous responses in spelling. In spelling through the sublexical route when the orthographic lexicon is involved, surface dysgraphia errors will mainly create existing words. In contrast, when the orthographic output lexicon is impaired, surface errors will lead to the production of non-lexical responses (as well as lexical ones).

A second predicted difference between the surface dysgraphia subtypes relates to the target words in which a surface error creates another existing word vs. those in which surface errors only create nonwords. The prediction is that individuals with surface dysgraphia whose orthographic output lexicon is functioning normally would make fewer errors in target words for which no phonologically-plausible misspelling creates an existing word, than in spelling homophonic and potentiophonic words, for which a phonologically plausible misspelling may create other existing words. This is in contrast to individuals with surface dysgraphia whose deficit is in the orthographic output lexicon itself, who are not expected to show differences between the two types of stimuli.

A third point we set out to examine was whether there is a difference between the two loci of impairment with respect to the effect of the frequency of the target word on spelling performance. If the orthographic output lexicon participates in spelling, the effects that operate on it should affect spelling. Therefore, in the case of disconnection surface dysgraphia, we expect an effect of the frequency of the target words on surface errors.

We will now examine these three predictions to explore the possibility of two patterns of surface dysgraphia.

4.1. Do individuals with surface dysgraphia differ with respect to the lexicality of erroneous responses?

To examine whether there is a difference between different surface dysgraphia subtypes in the tendency to produce an existing word in the surface errors, we created the lex-nonlex list, a new list of 91 words in which at least one surface error creates an existing word and at least one other surface error yields a nonword. For example, the word face can be spelled with a surface error as the existing word phase or as the nonword fase (or, for our Hebrew readers, the word שבט can be spelled with a surface error as the existing word שבת or as the nonlexical שווט). We selected the target words so that the number of possible nonlexical surface error responses for each word was always the same or larger than the number of possible lexical surface error responses. We tested these words in three tasks: 22 of the words were dictated, 42 were incorporated into a written sentence completion task as the target word, and 27 were included in a written picture naming task.

In addition, we analyzed the whole list of surface dysgraphia errors made by each individual in writing the total of 404 words in the screening and the unpredictable word lists. 54 of these words were such that at least one surface error creates an existing word and at least one other surface error yields a nonword.

Of all the surface errors that each participant made in the 145 (91 from the lex-nonlex list + 54 from the screening list and the unpredictable word list) words with both lexical and nonlexical potential for a surface error, we counted how many errors created a lexical response (writing be instead of bee), and how many were non-lexical (bea).

We only included in all these tests and analyses words without inflectional morphology and without bound function words, because such words are not represented as wholes in the lexicons (we assume that the phonological and orthographic lexicons only include the stems of the word, “orange”, “walk”, and “Mary”, but not “oranges”, “walked” and “Mary’s”, Badecker et al., 1990; Badecker et al., 1996; Friedmann & Coltheart, 2017; Hamilton & Coslett, 2007; Taft, 2015), and therefore words with inflectional morphology, for example, may require a different, and more complex, spelling process.

We expected that individuals whose disconnected yet functioning orthographic lexicon influences spelling will make mainly lexical surface errors, whereas participants with surface dysgraphia whose orthographic lexicon is impaired will show no preference for their surface error response to be lexical.

Results

The rates of lexical and non-lexical surface errors of each participant, out of the words with a potential for both types of errors this participant wrote, are summarized in Table 4 in the two left columns. Four participants (MSH, EI, NAM, LIZ) made significantly more lexical errors than non-lexical errors, a pattern that indicates that they were consulting the orthographic lexicon in spelling, even though they could not use the lexical route normally.

Table 4. The lexicality of surface error (out of the total words with a potential for both lexical and nonlexical surface response) and a comparison between surface errors in spelling words with and without a lexical potential for surface error.

	Lexical surface errors out of the 145 words with lexical and nonlexical surface error potentials		Surface errors in words with a lexical potential for surface errors vs. words without such potential
Participant	lexical errors	non-lexical errors	words with lexical potential N = 145–162	words without lexical potential N = 333	Frequency effect
MSH	13%*	6%	19%**	10%	Yes***
EI	18%*	6%	24%**	15%	Yes***
LIO	16%	20%	36%**	18%	Yes***
NAM	14%*	6%	20%	16%
LIZ	13%*	5%	18%	17%
SHY	12%	10%	22%	18%
HEN	10%	6%	17%	16%
RAM	23%	18%	41%	35%
OR	13%	10%	22%	22%
AN	10%	8%	18%	11%
HIL	8%	7%	15%	11%

Shaded perfromance cells indicate performance that involved the orthographic output lexicon:

*Significantly more lexical errors than non-lexical errors, p ≤ .03.

**Significantly more errors in words with lexical potential for surface errors than in words without such potential, p ≤ .02.

***A significant frequency effect, p ≤ .008.

Participants’ names shaded dark orange are the ones who we classify as disconnection surface dysgraphia, having access to the orthographic output lexicon, based on 3 indications for the effect of the orthographic output lexicon on spelling. The participant shaded golden orange is the one with two indications for disconnection surface dysgraphia. Participants shaded purple are the ones with only one indication for disconnection surface dysgraphia. Blue-shaded participants’ names indicate a deficit in the orthographic output lexicon itself, based on their absence of lexical effects.

4.2. Is spelling affected by whether the target word has lexical potential for surface dysgraphia error?

The effect of the existence of lexical phonologically plausible misspellings of the target word was examined by comparing surface errors in words with at least one possible lexical surface misspelling, to words with no such lexical possibility. We compared the surface error rate in all the target homophonic and potentiophonic words the participants wrote, taken from all the writing tests (a total of 145–162 words per participant⁴) to the surface error rate in unpredictable target words with no possible lexical surface error (a total of 333 words). (e.g., we compared target words like תוכן, which could be spelled both as the lexical טוחן and as the nonlexical טוכן and תוחן, to target words with no lexical possibility apart from the target word, like תוסף).

Results

The comparison of surface error rates in spelling words with and without a lexical potential for a surface error is summarized in the 3rd and 4th columns of Table 4. Two of the participants who made more lexical than non-lexical surface errors (MSH and EI), as well as LIO, made significantly more errors on words with a potential of lexical surface error than on words in which no surface error creates an existing word. This result indicates that the spelling of these participants is supported by the orthographic output lexicon.

Interestingly, in Hebrew only few of the phonemes that can be converted to two different letters have a phoneme-to-grapheme conversion probability that is clearly biased toward one of the letters. Such is the case of /s/, which is much more often spelled with the letter ס than with the letter ש. And therefore we take words with /s/->ש to be irregular (see the Appendix for a list of irregularities in Hebrew). Still, in such irregular words with /s/ that are spelled with the less-probable ש (as in the word משחק), when there was no lexical potential for surface errors, the participants with the lexical effects did not spell the word with the PGC-probable ס, which would have led to a non-lexical response. This forms yet another indication for the support from the orthographic output lexicon.

4.3. Is spelling affected by frequency?

To examine the effect of word frequency on surface errors in the two surface dysgraphia subtypes, we estimated the frequency effect, for each participant, on all the words s/he wrote that had the potential for at least one lexical and one non-lexical surface errors, in all the writing tasks. We calculated Point Biserial Coefficient between the log frequency of the target words (according to the HeTenTen21 web-crawled corpus of written Hebrew words, encompassing more than 2.7 billion words from 43,000 + web domains, Adler, 2007; Jakubíček et al., 2013; Suchomel & Pomikálek, 2012, accessible by SketchEngine, Kilgarriff et al., 2014) and the existence/non-existence of a surface error in spelling them. Three of the participants, MSH, EI, LIO, showed an effect of target word frequency on surface errors, whereby they made more surface errors in spelling lower-frequency target words than on higher frequency target words (r_pb = 0.30, p = .001; r_pb = 0.28, p < .001; r_pb = 0.39, p = .008, respectively). This could be explained in the following way: when the letter string resulting from the PGC in the orthographic output buffer is checked against the orthographic output lexicon, the more frequent word in the lexicon is supported, leading to the spelling of the more frequent potentiophone. When the target is the more frequent potentiophone, spelling through this process is correct, but when the target is the lower-frequency potentiophone, the result of the process is the other, more frequent option, and hence, incorrect.

4.4. Interim summary: two types of surface dysgraphia

To summarize, two participants (MSH and EI) consistently manifested lexical effects that suggest that their surface dysgraphia is supported by the orthographic output lexicon. Both of them made significantly more lexical- than non-lexical surface errors⁵; both of them, in spelling words with several possible sublexical conversions, made significantly fewer surface errors on target words for which there was no lexical phonologically plausible conversion other than the target, than for target words with several lexical options; and the surface error rate of both of them was affected by the frequency of the target word.

Another participant (LIO) manifested two of these effects: her spelling was significantly worse for (unpredictable) potentiophonic than (unpredictable) non-potentiophonic words and her surface errors were affected by frequency. So her deficit is probably also a disconnection rather than a deficit to the orthographic output lexicon.

Two participants, NAM and LIZ, manifested only one effect: they made significantly more lexical than nonlexical errors in spelling. The other six participants were not affected by any of the three lexical effects.

Based on these findings, it seems that MSH, EI, and possibly also LIO have disconnection surface dysgraphia whereas the others’ surface dysgraphia results from a deficit in the orthographic output lexicon itself. Whereas it is theoretically possible to find individuals with disconnection of the orthographic output lexicon only from the phonological lexicon or only from the semantic system, the three participants with disconnection surface dysgraphia in the current study showed surface dysgraphia both in writing to dictation and in writing from semantics, suggesting that their orthographic output lexicon is disconnected from both the phonological lexicon and the semantic system. Given that the participants with the disconnection type of surface dysgraphia sometimes spelled target words correctly, we assume that their connection was impaired, but not completely inaccessible. Thus, like in other neuropsychological impairments, the term “disconnection” describes what fails when spelling fails, but does not imply that the connection can never be operational.

5. Is it simply a “read after write” phenomenon? Do lexical responses result from reading, detecting the incorrect response, and correcting it?

As we described in the Methods section, when we analyzed the results, we only counted first written responses, and did not analyze the final response after correction. That way, we made sure that the lexicality of responses was not a result of the participants reading their own response, realizing it was not a word in their orthographic input lexicon (as we show in Section 7 below, seven of the participants had intact orthographic input lexicon), and correcting it into a lexical response.

To further examine whether the lexicality of response was assisted by their later reading of the written response, and to rule out an alternative explanation according to which they were using their preserved reading ability to monitor their writing and to weed out nonword responses, we asked the participants to write in a way that did not allow them to read what they had written.

5.1. The carbon paper test

We selected 50 words with a potential for a surface error out of the 328 unpredictable word list that were dictated to them (see Section 3.1): 15 low-imageability unpredictable words, 15 high imageability unpredictable words, 10 unpredictable function words, and 10 potentiophones. 25 of the words had a potential both for lexical and for nonlexical phonologically plausible misspelling. We asked the participants to write these words again, in a separate session, only this time we prevented them from using visual feedback. This was done using a carbon paper.

We placed a sheet of carbon paper between two sheets of white paper and asked the participants to write on the top sheet with a soft, inkless pen that did not leave marks (so that the participants could not see what they were writing while they were writing or afterwards, but we could look at the second page after the session and see what the participants had written).

We compared the participants’ surface errors on writing to dictation of these 50 words in standard writing, when they were able to see what they were writing, with their writing of the same words with no visual feedback.

Results

We found no significant difference in surface error rates and lexicality between the participants’ first responses to the same words with and without visual feedback, at the individual and the group levels (Table 5).

Table 5. Number of surface errors in writing the same 50 words with and without visual feedback.

Participant	With visual feedback	No visual feedback (carbon paper)
MSH	6	7
EI	7	8
LIO	7	6
NAM	9	7
LIZ	11	14
SHY	14	9
HEN	6	5
RAM	23	25
OR	7	12
AN	7	8
HIL	13	8

The tendency of the participants with disconnection surface dysgraphia to produce lexical responses also did not change when they wrote without reading feedback, and MSH, EI, and LIO (as well as RAM, OR, and HEN) still produced more lexical than nonlexical responses on the 25 words that had both lexical and nonlexical potential.

When writing with the carbon paper, and therefore without reading feedback, MSH and LIO had each only a single non-lexical error with the carbon paper, and LIO had only 2 non-lexical errors.

This suggests that it was not their reading that corrected their spelling and shaped the pattern of surface errors.

6. Where is the disconnection? And how does the orthographic lexicon influence spelling in cases of disconnection?

Now that we established that MSH and EI (and possibly also LIO) show support from the orthographic lexicon to their sublexical spelling, we can tackle the following questions: 1) where in the model is their disconnection, and 2) what is the path through which the orthographic lexicon supports spelling.

The results summarized in Table 3 above were that all the surface dysgraphia participants made surface errors both in written naming and in writing to dictation. Focusing on the three cases of disconnection surface dysgraphia, this already suggests that their disconnection is a double disconnection of the orthographic output lexicon, from the semantic system and from the phonological input lexicon (Figure 2b).

Figure 2. Two possibilities for the locus of disconnections in disconnection surface dysgraphia. (a) disconnection only from the semantic system to the orthographic output lexicon (OOL), (b) disconnections from the phonological input lexicon (PIL) and from the semantic system to the OOL. In dictation, lexical contribution in (a) is expected through PIL-OOL connection, and in (b) through a feedback between the OOL and the orthographic output buffer, supporting input from the sublexical route.

To further examine the connection between the phonological input lexicon and the orthographic output lexicon, we examined the homophonicity of the surface dysgraphia errors. If the three disconnection surface dysgraphia cases are able to use the phonological input lexicon-orthographic output lexicon connection in writing to dictation (as in Figure 2a), their lexical surface error responses should be homophonic to the target word.

If, however, their orthographic output lexicon is also disconnected from the phonological input lexicon, writing to dictation would have to proceed through the sublexical route. Lexical support in this case of double disconnection would arise from the feedback between the orthographic output lexicon and the orthographic output buffer (Figure 2b). In this case, the orthographic output lexicon is expected to support any written word that exists in it, not only ones that are homophonic with the target. It is therefore expected to support both homophones and potentiophones of the target word, namely, existing words that do not sound like the target, but are a phonologically plausible misspelling of the target.

Thus, these two possibilities yield different predictions with respect to potentiophonic errors in writing to dictation: spelling through the phonological lexicon is expected to cause surface errors that are homophones but not potentiophones of the target; spelling only with orthographic lexical support to the orthographic output buffer is expected to result in both homophonic and potentiophonic surface errors.

Take, for example, the word “face”. It sounds differently from “phase”, but, when spelled through the sublexical route, it may be spelled as “phase”. Because face and phase have different phonological representations, they would have two separate entries in the phonological input lexicon, each connected to a different entry in the orthographic output lexicon. Now comes the differential prediction of the two impairments: spelling through the route between the phonological input lexicon and the orthographic lexicon would yield correct spelling for potentiophones like “face”, because they will activate the separate entry in the phonological input lexicon, and from there the appropriate entry in the orthographic lexicon. The other possibility, under which spelling proceeds via phoneme-to-grapheme conversion and only later gets support from existing entries in the orthographic lexicon, has a different prediction. The orthographic lexicon will support any output of the phoneme-to-grapheme conversion that is lexical, so face and phase will both get support. Namely, in sublexical spelling, potentiophones like “face” and “phase” and homophones like “no” and “know” will receive the same treatment and the same support.

Under this logic, we have a way to examine the connection from the phonological input lexicon to the orthographic output lexicon: if spelling proceeds through the phonological input lexicon, potentiophones will be spelled correctly, and surface errors will occur only in homophones. If spelling proceeds sublexically, and the lexical support from the orthographic output lexicon comes only as feedback between the orthographic output lexicon and the orthographic output buffer, surface errors will occur on both potentiophones and homophones. The orthographic output lexicon only supports existing written words but should not be sensitive to their phonology.

To pinpoint the exact locus of impairment and the source of the lexical support in our participants’ spelling we therefore examined their writing to dictation of target words that have potentiophones.

A similar consideration concerns the path used for writing from semantics once semantics is disconnected from the orthographic output lexicon. If they are using a bypass from semantics, through the phonological output lexicon to the orthographic output lexicon, we would also expect no surface errors in target potentiophones, or at least significantly fewer surface errors in target potentiophones than in target homophones. If, in contrast, spelling goes only from semantics through the phonological output lexicon to the phonological output buffer and from there in a sublexical route to the orthographic output buffer, and the lexical contribution is only from the feedback from the orthographic output lexicon to the orthographic output buffer, we expect surface errors in both target potentiophones and target homophones. We therefore also compared their writing from semantics of homophones and potentiophones.

Method

The participants wrote to dictation 21 potentiophones, defined as target words for which a lexical phonologically plausible misspelling yields an existing word that sounds differently from the target, e.g., face->phase (EI did the task twice, so he wrote 42 potentiophones). In written naming (sentence completion and picture naming) they wrote 44 potentiophones (EI wrote 88) and 25 homophones, defined as a target word for which the lexical phonologically plausible misspelling is a homophone of the target e.g., plain->plane (EI wrote 50⁶). The frequency of the target homophones and potentiophones was similar, with no significant difference between the two types of target words. Seven words were excluded from the analyses because more than 30% of the participants in the control group wrote them with surface errors.

We compared the rates of surface errors they made in potentiophone dictation to controls, and compared their writing from semantics of homophones and potentiophones.

Results

Writing potentiophones to dictation: do they make potentiophonic, non-homophonic surface errors?

The findings were clear-cut: in writing to dictation, all three disconnection surface dysgraphia participants made many (29%−44%) surface errors in potentiophones, resulting in non-homophonic surface errors, and significantly more than the controls, p < .003. (Table 6 presents their performance in comparison to 17 age-matched controls aged 13;5–16;0).

Table 6. Percentage of words spelled with surface dysgraphia errors in writing potentiophones to dictation and in writing homophones and potentiophones from semantics in the disconnection surface dysgraphia participants, in comparison to age-matched controls.

	Dictation Potentiophones	Written naming
		Homophones	Potentiophones
MSH	44	9	17
EI	44	20	6
LIO	29	13	13
Control	6.5 (7)	4.0 (5.2)	2.7 (3)

This shows that they did not use the phonological input lexicon to access the orthographic output lexicon directly. Rather, it seems that, at least for our three participants who had disconnection surface dysgraphia, there was a double disconnection: they could neither access their intact orthographic lexicon from the semantic system, nor could they access it from the phonological input lexicon. This suggests that the source of their lexical support in writing to dictation was the feedback between the orthographic lexicon and the orthographic output buffer (Figure 3). This feedback process would not be sensitive to whether the written word sound the same as the target word or not. Therefore it should not distinguish between homophones and potentiophones, only whether the potential response is an existing word in the orthographic output lexicon.

Writing potentiophones and homophones from semantics: Do they make surface errors in potentiophones? Do they make a similar rate of surface errors in target homophones and potentiophones?

Moving to writing from semantics, we hypothesized that if the source of the lexical contribution from the orthographic output lexicon is only a result of feedback between the orthographic output buffer and the orthographic output lexicon, target potentiophones in written naming should show similar surface error rates as target homophones (which had a similar frequency).

Figure 3. Our conclusion regarding the locus of deficit of the participants with disconnection surface dysgraphia: a double disconnection of the orthographic output lexicon from the phonological input lexicon and from the semantic system (marked by stars), spelling sublexically with lexical support through feedback from the orthographic output lexicon.

The findings, summarized in Table 6, were that the participants with disconnection surface dysgraphia made surface errors also when spelling target potentiophones from semantics. Their surface error rate in spelling potentiophones was similar to their rate of surface errors for target homophones. These results suggest that also in writing from semantics, these participants cannot use a route from the phonological output lexicon to the orthographic output lexicon (see Ellis, 1993; Goodman & Caramazza, 1986; Hatfield & Patterson, 1983, for discussion of the connection between the phonological output lexicon and the orthographic output lexicon). Had they been using this bypass, we would not expect errors in potentiophones, and we would expect more errors in homophones than in potentiophones, contrary to our findings.⁷

Interim summary

The participants in this study showed two different patterns of surface dysgraphia. Two participants (MSH and EI, and possibly also LIO) showed disconnection surface dysgraphia in which the orthographic output lexicon was preserved, but disconnected from the semantic lexicon and from the phonological input lexicon. These participants’ surface dysgraphia errors were mainly existing words, and they were affected by word frequency. They made more errors on words on which at least one of the incorrect options for spelling sublexically is an existing word (homophones and potentiophones) than on unpredictable words for which all incorrect sublexical spelling options were nonlexical. We showed that the lexical support from the orthographic output lexicon to their spelling is mainly achieved through lexical feedback from the orthographic output lexicon to the orthographic output buffer.

The grapheme sequence can arrive in the orthographic output buffer either from the sublexical route connecting the phonological input buffer and the orthographic output buffer, in case of writing from auditory input, such as in dictation tasks, or from the sublexical route connecting the phonological output buffer and the orthographic output buffer, in cases of naming from semantics, such as in free spelling and in written naming.⁸

Six other participants (RAM, OR, AN, HIL, HEN, SHY) showed orthographic output lexicon surface dysgraphia. Their orthographic output lexicon was impaired, leading to an impairment in spelling through it, and to lack of support from the lexicon to the lexical options in spelling. They did not show a tendency to produce mainly lexical surface errors, they were unaffected by word frequency, and they had the same rate of surface errors in target words that had a lexical potential for a surface error as in target words without a lexical potential.

The additional participants showed some, but not all, the properties of disconnection surface dysgraphia.

7. Do reading and writing share an orthographic lexicon? Evidence from dissociations with reading

A question that has triggered much discussion and research is whether the orthographic lexicon is shared for reading and writing (Figure 4a) or whether there are separate orthographic input and output lexicons (as we portrayed above in Figure 1, repeated here as Figure 4b).

Figure 4. (a) A shared orthographic lexicon (b) Separate orthographic input lexicon and orthographic output lexicon.

One source of evidence relevant to answering this question has been reports of dissociations between surface dyslexia and surface dysgraphia. A deficit in the orthographic input lexicon causes surface dyslexia, and a deficit in the orthographic output lexicon causes surface dysgraphia. A double dissociation between surface dyslexia and surface dysgraphia, thought some researchers, might provide an answer to the riddle. And indeed, the literature contains dissociations between surface dyslexia and surface dysgraphia (Beauvois & Dérouesné, 1981) that are especially convincing when they pertain even to the same words (Coltheart & Funnell, 1987 for acquired dysgraphia; Hanley & Sotiropoulos, 2018 for developmental dysgraphia), alongside some associations between surface dyslexia and surface dysgraphia (Baxter & Warrington, 1987; Behrmann & Bub, 1992; Burt & Tate, 2002; de Partz et al., 1992; Hatfield & Patterson, 1983; Kremin, 1985; Newcombe & Marshall, 1985; Parkin, 1993; Saffran, 1985). Starting with the reported associations, like any association in cognitive neuropsychology, these cannot be taken as evidence for a shared module (Coltheart, 2002; Ellis & Young, 1996; Nickels et al., 2011; Shallice, 1988), as, for example, they could be a result of proximity of the neural substrates for reading and writing (Bub & Kertesz, 1982; Tainturier & Rapp, 2001; See Rapp et al., 2016 for a summary of findings from functional imaging supporting similar areas activated for the orthographic lexicon in reading and spelling).

As for the dissociations, firstly, whereas a deficit in the orthographic input lexicon causes surface dyslexia, not all cases of surface dyslexia result from a deficit in the orthographic input lexicon, both in acquired cases and in developmental cases (Coltheart & Funnell, 1987; Friedmann & Lukov, 2008; Gvion & Friedmann, 2016); similarly, as we have shown in the current study, surface dysgraphia does not necessarily imply a deficit in the orthographic output lexicon itself. Therefore, a dissociation between surface dyslexia and surface dysgraphia does not directly mean a dissociation between deficits in the orthographic input and output lexicons. As Tainturier and Rapp (2001) pointed out, dissociations may, instead, reflect a shared lexicon with distinct connections- for reading, from the orthographic visual analyzer/input buffer to the orthographic lexicon; and for writing, from the orthographic lexicon to the orthographic output buffer. Namely, a patient who demonstrates a dissociation between surface dyslexia and dysgraphia may in fact have a selective deficit in the access to or from a shared orthographic lexicon.

Therefore, for a dissociation between surface dyslexia and surface dysgraphia to genuinely shed light on the existence of two orthographic lexicons for input and output, it is essential to establish whether the deficit is in the lexicon itself or rather in the access to or from it. The current group of participants with surface dysgraphia provides an interesting opportunity to examine this, because we identified, for each of them, not only whether they have surface dysgraphia but also whether their surface dysgraphia resulted from a deficit in the orthographic output lexicon itself or in the access to it. So what we need to do next is to examine whether in reading they have surface dyslexia, and if so, whether their deficit is in the orthographic lexicon itself or in the access to/from it.

Our first step in doing this was using the methodology and tests for detecting the locus of impairment in the lexical reading route described in Friedmann and Lukov (2008). We examined the participants’ reading aloud of irregular, potentiophonic, and polyphonic words (to establish whether or not they have surface dyslexia) and then we tested their lexical decision of pseudo-homophones – to examine their orthographic input lexicon and the access to it from the orthographic input buffer. We also examined their comprehension of potentiophones and homophones as an indication of the status of the orthographic input lexicon and the access to it from the orthographic-visual analyzer and from it to the semantic system. Finally, we examined their oral naming, to test their phonological output lexicon, which may also be a source of surface dyslexia in reading aloud (Gvion & Friedmann, 2016).

If writing and reading share an orthographic lexicon (as in Figure 4a) then the prediction is that individuals with orthographic lexicon surface dysgraphia will also show orthographic lexicon surface dyslexia – namely if their lexicon in impaired in writing, it should also be impaired in reading aloud.

If, however, individuals with an impairment in the orthographic lexicon itself in spelling show an intact orthographic lexicon in reading, this will indicate that there are in fact two orthographic lexicons, as in Figure 4b.

Therefore, we tested the reading of the eight participants with surface dysgraphia who showed indications of impairment in the orthographic lexicon itself (the 6 with no lexical effects and 2 with only one effect, see Table 4).

Hebrew is a good testing ground for this question. Similarly to what we explained above about spelling, Hebrew orthography is also highly unpredictable for reading. There are very high degrees of freedom in reading every word. This results from several sources (for a more elaborate discussion see Friedmann & Lukov, 2008; and Gvion & Friedmann, 2016): some vowels (/a/, /e/, and sometimes /i/) in the middle of the word are not represented. As a result, one has to use the lexicon (and the morphological structure of the word) to know which vowels should be inserted, if at all. For example, geshem (rain) is spelled GSM (גשם), and in order to know that it should be read “geshem”, and not as the nonwords “gasham”, “gsham”, “gashem” etc., the lexical entry has to be consulted. In addition, each of the four vowel letters can also be read as a consonant, and four of the consonant letters may be converted to either of two consonants. Stress position is not marked orthographically. Again, the lexicon is the one to provide the information about the way these nine letters should be read in a specific word, and where the stress is in the target word. These properties make the reading of each word in Hebrew unpredictable, i.e., there is no word that can be unambiguously converted to a phonological string, and lexical information is required for the choice of the correct rendering of the target sequence. (Hebrew has an option of adding diacritics to the script, which reduces polygraphy for vowels and for 4 of the consonants. This option is used only in very specific contexts: poetry, prayer books, and first grade and children books. It is very rarely used otherwise).

7.1. Reading tasks and results

7.1.1. Reading aloud: do the participants with impaired orthographic output lexicon also have surface dyslexia?

Reading aloud was assessed using the TILTAN screening reading test (Friedmann & Gvion, 2003). The word reading test includes 196 2–11 letter Hebrew words: 136 single words presented in a list one above the other, and 60 words presented in pairs, with no semantic relation between the words within a pair (or in the list). The test was constructed so that it contains words sensitive to various types of dyslexia (migratable words, abstract words, function words and morphologically complex words, words with many orthographic neighbors, words for which vowel letter errors create other existing words, words in which neglect errors on their left or right side yield other words). Specifically for surface dyslexia, the properties of the Hebrew orthography yield a situation in which, similarly to spelling, and even more extremely than in spelling, no word in Hebrew can be read unambiguously and correctly through grapheme-to-phoneme conversion, and correct reading requires the lexical route. (The test includes a few words for which two lexical readings are possible, similarly to the word tear in English. We accepted both readings as correct).

Therefore, essentially all words in the screening test are sensitive to surface dyslexia. The test also includes 34 potentiophones, which are most sensitive to surface dyslexia. We also administered the nonword screening test, which includes 30 nonwords.

The performance of the adult (LIZ, HIL, NAM, AN) and 17-year-old participants (OR, RAM) was compared to the performance of 1073 adults aged 20–50 years (M = 28), with 12 years of education and above; the performance of the participants aged 14–15 (SHY, HEN) was compared to 26 7th graders with typical reading.

Results

The performance of the participants with orthographic lexicon surface dysgraphia on the TILTAN oral reading test are summarized in Table 7.‏ Four participants (NAM, OR, AN, HIL) showed normal reading aloud, which did not differ from that of their matched controls in the number of surface errors.

Table 7. Tests that assess the orthographic lexicon in reading: Number of surface errors in reading aloud (of 196 words) and % correct (and number of errors) on the lexical decision, homophone word-association and homophone-potentiophone picture-word matching tasks.

Participant with surface dysgraphia ^a= adult control ^c = adolescent control	Reading aloud (number of surface errors)	Lexical decision pseudohomophones	Homophone/potentiophone association	Homophone/potentiophone picture matching
LIZ^a	6*	89 (7)*	85 (6)*	90 (4)*
RAM^a	14*	89 (7)*	83 (7)*	71 (12)*
SHY^c	7^	91 (6)*	85 (6)*	90 (4)*
HEN^c	7^	89 (7)*	88 (5)*^	90 (4)*
HIL^a	1	100 (0)	90 (4)	90 (4)*
NAM^a	2	97 (2)	97 (1)	100 (0)
OR^a	5	97 (2)	90 (4)	93 (3)
AN^a	1	98 (1)	92 (3)	97 (1)
^aControl: Adults M (SD)	2.7 (1. 5) N = 1073	98 (1.5) N = 148	98.3 (6.0) N = 141	98 (3) N = 13
^cControl: Adolescents 7–10th grade M (SD)	7th grade 4.1 (2.6) N = 26	9th grade 99.3 (1.1) N = 59 ^c7–8th grade 98.9 (2.7) N = 24	7–9th grade 97.4 (6.8) N = 74	7–10th grade 99.1 (4.9) N = 13

*performance that is significantly below the matched controls, p ≤ .04 (these cells are also shaded), ^not significantly below the control performance but in the bottom of the control range (p = .14).

Participants whose names are shaded in blue show reading without surface dyslexia, indicating intact orthographic input lexicon.

Participants whose names are shaded in yellow showed some indication of surface dyslexia with a deficit in the orthographic input lexicon, in addition to their surface dysgraphia with a deficit in the orthographic output lexicon.

Two other participants (RAM, LIZ) made significantly more errors than the controls and produced significantly (p ≤ .05) more surface dyslexia errors, suggesting that they read more words than the controls via the sublexical route (the surface dyslexia of all three was relatively mild). Two participants, SHY and HEN, showed marginal impairment in oral reading.

The participants’ reading of non-words was preserved. All of them were within the control range, with an average of 1.5 errors (SD = 1.4).

7.2. Further testing in order to determine the locus of impairment in the lexical route for reading

The next step was to determine, for each participant, whether the orthographic lexicon was functioning normally and could be used in (silent) reading or not, and whether access from it to the semantic lexicon is preserved. We used a lexical decision task, which indicates the status of the orthographic lexicon and the access to it, and two homophone comprehension tasks, which indicate the status of the access from the orthographic input lexicon to the semantic lexicon (as well as the status of the orthographic and semantic lexicons themselves). The silent reading tasks are taken from Friedmann and Lukov (2008). The participants were requested to perform all three silent reading tasks without sounding out the words they read. We also tested oral picture naming to assess the phonological output lexicon.

Written lexical decision

The participants saw 66 pairs, each pair includes a correctly spelled word (castles, gem) and its pseudo-homophone (casles, jem). The participants were requested to circle the correctly spelled word. The control groups for this test included 148 adults, 59 9th graders, and 24 7th–8th graders.

Written homophone-potentiophone comprehension

We tested the comprehension of written homophones and potentiophones using a word-picture-matching task and a word-association task.

The word-picture matching test consisted of 41 triads that included a picture (e.g., a knight), and two written words- the matching word (knight) and its homophone (night) or potentiophone. The participant was requested to point to the word that matched the picture.

The written homophone/potentiophone word association task included 40 written word triads. Each triad included a target word (money), a word that was semantically related to the target word (buy), and a homophone or a potentiophone of the related word (bye). The participant was requested to point to the word that was associated with the target word.

The control groups for the word association task included 141 adults and 74 7th–9th graders. The control groups for the picture task consisted of 13 adults and 13 7th–10th graders.

Oral picture naming

The participants’ phonological output lexicon (and the access to it from the conceptual-semantic system) was assessed using a task of oral naming of 100 color pictures of objects (SHEMESH, Biran & Friedmann, 2004, 2005; Friedmann & Biran, 2003). The participants saw a picture of an object and were asked to name it aloud. We analyzed the number of pictures named correctly immediately. Hesitations longer than 5 s and incorrect first responses were counted as incorrect responses.

Results

The results are summarized in Table 7. The important finding is that four of the participants who had a deficit in the orthographic output lexicon (HIL, NAM, OR, AN) had no deficit in the orthographic input lexicon. They showed no surface dyslexia: they read aloud well and performed well and similarly to controls also on the lexical decision and on the two homophone comprehension tests, which, together with their oral reading, indicated an intact orthographic lexicon in reading. As a result, these participants form examples of a dissociation between an impaired orthographic output lexicon and an intact orthographic input lexicon.⁹

Four other participants (SHY, LIZ, RAM, HEN) showed some indication of a deficit in the lexical route for reading. LIZ and RAM performed significantly below the matched control group on all 4 surface dyslexia tasks: they made significantly more surface errors in reading aloud (see Section 7.1 above); they performed significantly below (p ≤ .04) their matched controls in the orthographic lexical decision test and in the two homophone comprehension tests, indicating a deficit in the orthographic input lexicon (or in the access to it from the orthographic input buffer, see Figure 1). The two other participants, SHY and HEN, showed some indications of surface dyslexia although they were less clear cases: both of them were only marginally below the controls in reading aloud, SHY was significantly impaired in lexical decision and in the two comprehension tasks, and HEN was below the controls in the lexical decision task and in one of the comprehension tasks.

The oral naming of all the participants was good and within the normal range, with performance of 96%−100% correct (M = 98%). This indicates that the phonological output lexicon of each of the participants and the access from the phonological output lexicon to the phonological output buffer were also intact.

7.3. Teasing apart a deficit in the orthographic lexical representations themselves from a deficit in access from the orthographic lexicon to the orthographic output buffer: reading and writing the same words

Thus, four participants showed a dissociation, with a deficit in the orthographic lexicon for spelling and a spared orthographic lexicon (and spared access to it from the orthographic-visual analyzer and the orthographic input buffer) for reading. This dissociation can be interpreted as indicating that there are two orthographic lexicons, one for reading, the other for writing, as depicted in Figure 4b. A further way to examine whether there is a shared orthographic lexicon for reading and writing is to examine reading and writing of the same words. We tested this in three of the participants, who agreed to further testing: two who showed a dissociation between impaired orthographic lexicon in writing and spared orthographic lexicon in reading (NAM and OR) and one who showed an association (RAM). The participants read and wrote the same unpredictable words and we examined whether they made errors on the same words in reading and in writing. This task was especially important for understanding RAM’s deficit, who showed impairment in the orthographic lexicon in both reading and writing. If RAM’s deficit is in the entries stored in an impaired shared orthographic lexicon, he should make errors on the same entries in reading and writing. For NAM and OR this test was mainly used to re-establish that they did not have surface dyslexia.

Methods

The participants read and wrote the same unpredictable words (NAM: 129 words; OR: 119 words; RAM: 133 words). All these words were unpredictable both for reading and for writing. I.e., they were such that reading them via the sublexical route can yield surface dyslexia errors, and also that spelling them via the sublexical route can yield surface dysgraphia errors. The selected words were comparable with respect to conversion probability in both reading and writing (i.e., both irregular, both unbiased polygraphic, or both polygraphic spelled with the less probable letter).

Results

RAM was the participant who showed impaired orthographic lexicons in both reading and writing. However, he showed two important differences between reading and writing of the same words, indicating that his deficit is not a deficit in the entries of a shared orthographic lexicon. A) his reading was far better than his writing of the same words, even though all words had a potential for a surface dyslexia error (phonologically plausible incorrect reading). He made 76 surface errors in writing and only 15 surface errors in reading of the same 133 words. B) even when he did make surface errors in reading, he did not necessarily make them on the same words that he misspelled: he only made errors on both reading and writing in 7 of the 133 words, and none of these errors was the same in reading and writing.

For NAM and OR we also saw a very clear difference between reading and writing, which, for them, was not surprising given that we have already established that they did not have surface dyslexia (their rate of surface errors in reading was within the norm) even though they showed impaired orthographic output lexicon for spelling. NAM made 44 surface errors in writing and only 2 surface errors in reading of the same 129 words. OR made 48 surface errors in writing and only 6 surface errors in reading the same 119 words.

McNemar test analyses indicate that reading and writing of the same words was significantly different for each of the three tested participants, with p < .0001 for each of them.

These results suggest that even though RAM did have a mild surface dyslexia in reading, with a deficit in the orthographic lexicon, it was not the same lexicon that was impaired in his spelling, as he did not make errors on the same entries in reading and writing. For NAM and OR the results further supported our previous conclusion regarding their intact orthographic input lexicon alongside an impaired orthographic output lexicon. The final experiment examined whether indeed their deficit was in the orthographic output lexicon itself or in its access to the orthographic output buffer.

7.4. Re-writing the same words

Next, we examined an alternative interpretation according to which it is not the orthographic output lexicon itself that is impaired for these participants, but rather the access from the orthographic lexicon to the orthographic output buffer (Hanley & Sotiropoulos, 2018). Such alternative explanation can maintain the assumption of a single lexicon that is intact, and only the access from it for spelling is impaired (Figure 4a). A source of evidence as to whether the deficit is in the orthographic lexicon itself or in the access from/to it is consistent failure on specific lexical items. Howard (1995), for example, deduced from an aphasic patient’s consistent inability to retrieve specific lexical entries from the phonological output lexicon that the deficit lay in the lexicon itself.

Therefore, to examine whether the spelling deficit of NAM and OR, the two participants who showed a dissociation, was indeed in the orthographic output lexicon or whether it was in its output to the orthographic output buffer, we tested their spelling of the same words time and time again in several sessions. We hypothesized that consistent success in spelling specific words, and consistent failure in specific words (spelling them with a surface error though not necessarily with the exact same surface error) is better explained by a deficit in the entries stored in the orthographic lexicon itself rather than in a deficit in the access from an intact orthographic lexicon to the orthographic output buffer.

Method

We dictated to NAM and OR the same words several times, in separate sessions: NAM: 184 words (90 words 3 times, 94 words 5 times); OR: 173 words (79 words 3 times, 48 words 5 times, 46 words 6 times), and examined whether they made surface errors on the same target words. Consistent responses were defined as correct on all trials or on all but one of items that were written 5 or 6 times, and as 3 correct out of items written 3 times (at least 5 correct out of 6, at least 4 correct of 5, and all correct out of 3), and as spelling with a surface error on more than half of the trials (at least 4 errors of 6 trials, at least 3 errors on 5 trials, at least 2 errors on 3 trials).¹⁰

Results

NAM and OR showed a very consistent pattern of performance: NAM was consistent on 93% of her trials (142 consistently correct, and 29 consistently written with a surface error out of 184 items). OR was consistent on 90% of the items he wrote (109 consistently correct, and 46 consistently written with a surface error out of the 173 words he wrote repeatedly).

This finding is indicative of a deficit these two participants had in the stored items in the orthographic output lexicon (Howard, 1995) rather than in the access from the orthographic lexicon to the orthographic output buffer.

7.5. Interim summary – separate orthographic lexicons for reading and writing

We found that 4 participants showed impairment in the orthographic lexicon for writing but not in the orthographic lexicon for reading, as demonstrated by their unimpaired reading and absence of surface errors in reading aloud, in lexical decision and in comprehension tasks. This is consistent with a model with two separate orthographic lexicons, one for input the other for output, where these participants are only impaired in the orthographic output lexicon (Figure 4b). We ruled out an alternative explanation according to which the orthographic lexicon is shared, and the deficit of the participants in spelling lies in the connection between the shared orthographic lexicon and the orthographic output buffer. This disconnection explanation was ruled out for two of the four participants by the pattern of spelling of the same words in different sessions. Words were either consistently written correctly or consistently written with a surface error. Such a pattern is more easily accountable under a deficit in the stored lexical representations in the orthographic output lexicon itself rather than an access deficit. It would be difficult for a deficit in the access to the orthographic output buffer to induce errors on consistent items.

Finally, a participant who showed a seeming association between an impairment in the orthographic lexicon itself in reading and in writing in fact showed far better reading than writing for the same words and he did not make errors on the same items in reading and writing. There were lexical entries that he was able to activate in reading but not in writing, and some words he was able to write but not to read. A McNemar analysis yielded a completely dissociable performance in reading and writing: he made five times more surface errors in writing than in reading words that were all irregular both for reading and for writing, and made errors on different words when he read and wrote the same target words. This may be interpreted as separate deficits in the lexical orthographic entries for reading and for writing.

Thus, our results support two separate orthographic lexicons- one for reading and one for writing, as in Figure 4b.

8. Discussion

8.1. Two types of surface dysgraphia

Surface dysgraphia is a deficit that causes impaired spelling of irregular words, which manifests itself in phonologically plausible errors. This pattern is usually attributed to an impairment to the orthographic output lexicon, forcing the writer to spell via the sublexical route (McCloskey & Rapp, 2017). However, a deficit in the lexical route, resulting in sublexical spelling, can stem not only from a deficit to the orthographic output lexicon itself. Such a pattern can also stem from a disconnection of an intact orthographic output lexicon from the semantic system and the phonological input lexicon.

In this study we reported on these two types of surface dysgraphia. Two (possibly three) individuals showed disconnection surface dysgraphia: their orthographic output lexicon was not impaired and contributed to their spelling, but it was disconnected from the semantic system and from the phonological input lexicon.

Six other participants showed orthographic lexicon surface dysgraphia, which resulted from a deficit in the orthographic output lexicon itself, and their error pattern exhibited no lexical contribution.

Three main characteristics distinguished the pattern of performance of the individuals with disconnection surface dysgraphia from that of the individuals with orthographic lexicon surface dysgraphia:

1. The participants with disconnection surface dysgraphia made significantly more lexical than nonlexical surface errors: even though they misspelled the target words and instead wrote a different phonologically plausible string, in most of the cases, this error was an existing word, either a homophone or a potentiophone of the target word.

2. Relatedly, the participants with disconnection surface dysgraphia made more surface dysgraphia errors in words that had a lexical potential for surface errors, namely, in words in which there was another option to spell the string via the sublexical route which would result in an existing word. When the target word was the only lexical phonologically plausible misspelling, it was the option selected, and hence it was spelled correctly

3. Finally, the participants with the disconnection surface dysgraphia showed a word frequency effect: they made fewer surface errors in high-frequency target words.

In contrast, the six participants whose surface dysgraphia resulted from a deficit in the orthographic output lexicon itself did not show any involvement of the orthographic lexicon in spelling: they did not have a tendency for lexical misspellings, no tendency to make more errors on words with a lexical potential for a surface error, and did not show a word frequency effect.

The lack of consistent frequency effect in cases of orthographic output lexicon deficit is in line with previously reported cases with impaired orthographic lexicon which found no effect of word frequency on spelling of irregular words, in an individual with developmental surface dysgraphia (Hanley et al., 1992). Other studies have reported no consistent word frequency effect: Luzzatti et al. (1998) tested 18 cases of acquired surface dysgraphia and found that only 44% of them showed a word frequency effect. Other cases of impaired orthographic lexicon did show frequency effect (the acquired single case in de Partz et al., 1992; the acquired single case and the developmental single case in Temple, 1985; and the developmental case in Romani et al., 1999). These findings suggest that impairments in the orthographic output lexicon are not necessarily characterized by a frequency effect. This is probably an indication that there are different natures for different types of orthographic output lexicon deficits.¹¹

These three lexical effects point to the involvement of the orthographic output lexicon in spelling in cases of disconnection surface dysgraphia. We suggested that the intact-yet-disconnected orthographic output lexicon influenced spelling in these cases in the following way: the word was spelled sublexically, in a phoneme-to-grapheme conversion route (in dictation – through the route between the phonological input buffer and the orthographic output buffer; in written naming from semantics, to the phonological output lexicon and through the route between the phonological output buffer and the orthographic output buffer, see Figure 3 above). The string that arrived in the orthographic output buffer following the sublexical phoneme-to-grapheme conversion was then supported by a feedback mechanism from the orthographic output lexicon (similar to proposals by Barr et al., 2020; Kohnen, Nickels, Brunsdon, et al., 2008; McCloskey et al., 2006; and Sage & Ellis, 2006, for spelling; and Dell et al., 1997; Hillis & Caramazza, 1995; Rapp & Goldrick, 2000, for lexical feedback in phonological output). This way, the orthographic output lexicon supports any lexical phonologically plausible misspelled letter string that arrives in the orthographic output buffer, not only strings that sound the same as the target word.

This spelling path, and specifically, the lexical contribution from the orthographic output lexicon which is provided by feedback to the orthographic output buffer, is supported by the findings that the participants with disconnection surface dysgraphia made surface errors in writing to dictation not only on homophones, but also on potentiophones (such as angle-angel; no-now; comb-come). This led us to conclude that they did not write to dictation through a connection between their phonological input lexicon and the orthographic output lexicon (differently, for example, from Brunsdon et al.’s 2005 proposal for the spelling errors of the individual they studied). Had they been using such route, we would not expect to see surface errors in potentiophones (which sound differently, and hence are expected to activate different entries in the orthographic output lexicon). These considerations – the surface errors in writing potentiophones to dictation–- also led us to conclude that the disconnection of the three participants in our study was both from the semantic system and from the phonological input lexicon to the orthographic output lexicon.

Finally, a question that would be interesting to examine in future work is whether the two varieties of surface dysgraphia would benefit from different treatment approaches. It may be that the lexical deficit would benefit more from repetition of spelling the same words, whereas the disconnection variety would benefit more from strengthening the connection between the semantic lexicon and the orthographic output lexicon, for example by spelling within a sentence context, or by written naming of pictures.

8.2. Separate orthographic input lexicon and orthographic output lexicon

A long-standing question relates to whether there are separate input and output orthographic lexicons, or whether there is one shared orthographic lexicon (Baxter & Warrington, 1987; Behrmann & Bub, 1992; Coltheart & Funnell, 1987; de Partz et al., 1992; Hanley & Sotiropoulos, 2018; Hatfield & Patterson, 1983; Parkin, 1993; Saffran, 1985; Sotiropoulos & Hanley, 2017).

We used the fine-grained characterisation of the locus of impairment of our participants in the spelling model to re-examine this question. We focused on the participants for whom we had established in the spelling testing a deficit in the orthographic output lexicon itself, and asked, first, whether a dissociation can be found: whether their orthographic lexicon is intact in reading tasks. We thoroughly assessed their orthographic input lexicon using tasks of reading aloud of irregular and potentiophonic words, lexical decision of pseudohomophones, and two tasks of written homophone comprehension. These tasks pointed to four participants who showed a clear dissociation between their impaired orthographic output lexicon and intact orthographic input lexicon.

Before this dissociation can serve as evidence for the existence of two separate orthographic lexicons, we still needed to exclude an alternative explanation. Whereas in the spelling assessments we established that the deficit of these four participants is not in the access to the orthographic output lexicon from semantics or from phonological input but rather in the lexicon itself, we had not ruled out that the deficit of these four participants was in the output of the orthographic output lexicon to the orthographic output buffer.¹² So it is still possible that the orthographic output lexicon itself was intact but its connection to the orthographic output buffer was the one that is impaired, causing sublexical spelling. For this reason, we looked at the properties of the surface errors with respect to consistency. The idea was the following: deficits in specific entries are far better explained under a deficit in a lexicon (Howard, 1995) than under a general difficulty in connection between components. We therefore examined the consistency of errors – whether the participants made surface errors in the same words when spelling the same words again and again. Two of the participants who showed a dissociation were tested in re-writing the same words, and were found consistent in their patterns: the two participants with orthographic output lexicon surface dysgraphia (whose orthographic input lexicon was intact) made errors in the same target words in repeated spelling (not necessarily the same errors), or succeeded in spelling the same words again and again. This points to a difficulty within the lexicon in specific entries, and therefore supports the conclusion derived from the dissociation regarding two separate orthographic lexicons, one impaired, the other one spared.

We also saw support for separate orthographic lexicons from a participant who showed a deficit in the orthographic lexicon in both reading and writing: he showed far fewer surface errors in reading than in writing of the exact same entries, and did not show consistent errors in the same items in reading and in writing.

Thus, we believe that pinpointing of the different loci of impairment giving rise to surface dysgraphia were also helpful in investigating the question of separate input and output orthographic lexicons.

8.3. Surface dyslexia in developmental cases

Our participants had a developmental, rather than acquired deficit. Firstly, as has been reported in other cases of developmental surface dysgraphia (Angelelli et al., 2004; Brunsdon et al., 2005; Hanley et al., 1992; Hanley & Sotiropoulos, 2018; McCloskey & Rapp, 2017; Romani et al., 1999; Temple, 1985), the patterns of impairment were very similar to those reported for acquired surface dysgraphia. Furthermore, the reason some children spell sublexically when others already spell correctly is that they need more exposures (including more instances of spelling the same word) than their peers. The finding that the participants with the orthographic-lexicon surface dysgraphia do not show word frequency effect suggests that insufficient exposure was not their problem, but rather that they had a developmental deficit that resembles the one observed in acquired cases. It can be conceptualized that they were born with the same deficit that individuals with surface dysgraphia acquire following stroke, for example. Similarly, the participants who did show lexical effects, the disconnection type, showed that the lexical items did exist in their lexicon, again rejecting the idea of insufficient exposure leading to lexical items not registered in the lexicon, and supporting the idea that developmental dysgraphias may be a result of a deficit very similar to those observed in acquired deficits.

Our findings show that an orthographic output lexicon can develop, and lexical entries can be established in it, even when its connections from semantics and from the phonological input lexicon are impaired. It is intriguing to think of the developmental path here: how could these individuals develop an orthographic output lexicon even though it is disconnected from input from semantics and from the phonological input lexicon? One option is that the disconnection is partial, so the lexicon is built in cases the connection does function. In addition, other components that are not disconnected may contribute to the development of the lexical items: One option is the phonological output lexicon, which may support the creation of corresponding lexical items in the orthographic output lexicon, possibly by way of creating “orthographic skeletons”, as suggested by Wegener et al. (2017, 2020). Another path to fill the lexicon may be through the orthographic input lexicon, which may contribute to the creation of lexical entries in the orthographic output lexicon, by way of monitoring written words.

Acknowledgments

We thank Iris Rubin-Zaxenberg and Revital Mor for their important role in participants’ recruitment and testing. We are deeply grateful to the members of the Language and Brain Lab and especially Hadar Green and Maya Yachini for helpful discussions. We also thank Anne Castles, Max Coltheart, Saskia Kohnen, and Lyndsey Nickels, of the beloved yet no longer existing CCD for exciting discussions of these data and models.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

Naama Friedmann was supported by the Lieselotte Lab for Child Development and the Branco-Weiss Chair for Child Development and Education. The research was supported by grants from Human Frontiers Science Program (RGP0057/2016, Friedmann) and from the Cukier-Goldstein-Goren Center for Mind and Language (MILA).

Notes

1 The term “potentiophones” is usually used in reading (Friedmann & Lukov, 2008). We use it here also for spelling. Examples for potentiophones in spelling in Hebrew include קוטב-כותב, צבת-צוות, שפה-ספה, פרא-פרה.

2 If the orthographic output lexicon impairment is in some way related to frequency (e.g., if the lexicon requires far more exposures than normal in order to encode an entry), then frequency may be involved in determining whether a word will be spelled through the lexicon, and hence whether it will be spelled correctly or not. However, it might be that the lexicon impairment is of a different kind, which has nothing to do with word frequency. In this case, because writing does not pass through the lexicon, no word frequency effect is expected.

3 Irregularity may be considered a special case of polygraphy, where the much-less probable option is selected; or it may be considered a violation of a PGC rule, where the rule takes only the very probable option and does not specify the very improbable rare conversion rule. Take phoneme P, which is converted in a certain orthography in 99.99% of the cases into grapheme G1, and in 0.01% of the cases into G2. The PGC rules may be P->G1 (possibly weighted 99.99) and P->G2 (0.01), in which case an irregular word with P->G2 is a case of polygraphy. Alternatively, the PGC rule may be just P->G1, in which case a word with P->G2 is a case of rule violation, an irregular word.

Silent letters can be seen in a similar way – either as a case of irregularity and rule violation: a rule dictates that a zero phoneme is converted to a zero grapheme. In irregular words like sword and doubt, a zero phoneme is represented with a grapheme, violating the rule. Silent letters may also be cast in terms of biased polygraphy: the rule of conversion of zero phonemes is a conversion to a zero grapheme in most cases, and to w, b, t, l in a tiny fraction of the cases, as in sword, doubt, listen, and talk.

We take here the definition of irregularity as cases of violation of PGC rules. Namely, an extremely biased conversion of a phoneme is a PGC rule, and the very few exceptions are violations of this rule, creating irregularity. These are cases of conversion that no Hebrew writer with normal spelling would use in writing nonwords with the target phonemes.

4 Some words were excluded for some participants due to technical reasons.

5 Could it be that the lexical effects were a mere result of phoneme-to-grapheme conversion probability that happened to be biased toward existing words in the target words we had selected? We think not. Had this been the case, we would expect the PGC probability to affect all participants with surface dysgraphia in the same way. The fact that some of the participants showed a very clear preference for lexical responses for the same words on which others showed no lexical response preference indicates that there is something beyond the PGC probabilities that modulated their spelling. It might indeed be that some of the lexical results of all the participants’ surface errors were due to phoneme-to-grapheme conversion probability, and this rate, for the impaired-lexicon participants may be the base-rate of lexical responses by chance and by conversion probability for all participants. Still, the rate of lexical responses of the participants with disconnection surface dysgraphia was far beyond this base rate.

6 There was no difference between the two runs in EI’s performance, so we preferred to use as much data as we could, and therefore we analyzed both runs.

7 Table 6 also indicates better writing of potentiophones from semantics than from dictation, for two of the participants. At this point we are not sure what we can conclude from this. Unlike in Table 3, the potentiophones in dictation and written naming were different words, so it may be that the ones we used in the dictation task were more susceptible to surface errors, and this is in fact also what we see in the control group. If this is a real difference, then it may be that the connection of the semantic system to the orthographic output lexicon is less impaired for these two participants, or it may indicate that sublexical writing is more readily available in dictation than when writing from semantics.

8 We assume a direct POB-OOB route (in line with models by, e.g., Beeson, 1999; Kay et al., 1996; Krajenbrink et al., 2015; Laiacona et al., 2009; Ogden, 1996; Whitworth et al., 2005). Such route would, for example, allow a child in the first stages of building the orthographic lexicon to spell a word that exists in their POL but not yet in the OOL, by means of accessing the entry in the POL, from there to the phonemes in the POB, and from there, by way of PGC, to a sequence of graphemes in the OOB. Whereas in research we rely on dictation tasks, and hence the sublexical PIB-OOB route, the ecological everyday tasks of writing words not via the orthographic lexicon probably originate in the semantic system and proceed via this POB-OOB route. Notice that this POB-OOB route is not necessary for writing nonwords to dictation, and the direct PIB-OOB seems to be used for this, as demonstrated and discussed in Friedmann et al. (2020).

9 Three of the participants, OR, AN, and HIL are clear instances of a dissociation. NAM did show one lexical effect in writing so she might be a less strong case.

10 We assume that a missing representation and writing via the sublexical route can randomly result in a correct sublexical writing.

11 Concerning individuals with orthographic output lexicon deficits, we assume that different impairments within this lexicon may contribute to the occurrence of surface dysgraphia errors. In some individuals, frequency determines whether or not a lexical entry is stored and accessible. For others, impairment to certain entries is not determined by word frequency.

12 A deficit in the orthographic output buffer itself is ruled out, because, as summarized in Table 2, the participants had very few errors of substitution, omission, and addition of non-homophonic letters. A selective deficit of phonologically-plausible misspellings could not arise from the orthographic output buffer.

Appendix: Unpredictability and irregularity in spelling Hebrew

1. Irregularity – violating regular rules: (phonemes for PGC appear here in English letters without slashes)

Regular rules:

1. regular and non-polygraphic PG conversion of consonants: b-ב,g-ג,d-ד,z-ז,y-י,l-ל,m-מ,n-נ, p-פ,f-פ,ts-צ,r-ר,sh-ש – these rules are not violated (except for cases of assimilation).

2. very biased polygraphy /s/ -> ס

3. mark every vowel in the end of the word with a vowel letter (a,e-> ע/א/ה, o,u->ו, i->י)

4. mark /o/ and /u/ in the middle of the word with the vowel letter ו

5. do not mark orthographically /a/ and /e/ in the middle of the word (except:

   • e1. mark /a/: by u, o, i and followed by any consonant except x-> ע/א/ה

   • e2. Do not mark /a/ preceded by u, o, i and followed by /x/ (uax, oax -> וח, iax->יח, éax->ח)

   • e3. within the diphthong aw and ai – polygraphic with or without א

6. do not mark zero phonemes orthographically

As a result, irregular words are word in which PGC is the following:

1. s->ש (safa->שפה)

   (/s/ is essentially polygraphic but it is much more biased towards PG conversion to ס, so we listed it in the regular rules, so that /s/-> ש is an irregular violation of the rule).

2. A word ending with a vowel in which the vowel is not marked orthographically (paru’a-> פרוע; katavta->כתבת)

3. o-> ה (eifo->איפה)

4. o-> א (zot->זאת)

5. i-> א (rishon->ראשון)

6. a-> י (stav->סתיו)

7. e -> י (en->אין)

8. ey in the end of the word -> ה/א

9. a and e in the middle of the word that are marked with א (kan->כאן)

10. vowel /o/ and /u/ in the middle of the word that is not marked orthographically (kol->כל)

11. silent letters: zero phonemes that are marked orthographically (kishu->קישוא; bari->בריא; shu'it->שעועית).

2. Polygraphy:

1. /t/ -> ת/ט

2. /k/ -> כ/ק

3. /x/ -> כ/ח

4. /v/ -> ב/ו/וו

5. /glottal stop/ -> א/ע/ה

6. /i/ in the middle may or may not be marked with a letter. When it is marked, it is with the letter י

7. /a/ in the end of the word -> ע/א/ה

8. /e/ in the end of the word -> א/ה

9. /ai/ in the end of the word -> אי/י