Abstract

Introduction:

Written language impairment is a frequent but still incompletely understood feature of the logopenic variant of primary progressive aphasia (lvPPA), and although spoken language deficits have been extensively studied, less is known about the mechanisms underlying impaired written word production and the conditions under which orthographic representations become particularly vulnerable. In this study, we investigated written language processing in individuals with lvPPA using a battery of tasks designed to dissociate phonological and orthographic contributions.

Methods:

Twelve participants with lvPPA and thirteen healthy controls completed written picture naming, written word dictation, written non-word dictation, grapheme dictation, and a picture-based letter-in-word judgment task that probed orthographic knowledge without requiring overt written output.

Results:

Compared with controls, participants with lvPPA showed significant impairments in written picture naming, written word dictation, grapheme dictation, and letter-in-word judgment, whereas written non-word dictation was relatively preserved at the group level. Within grapheme dictation, vowel graphemes were disproportionately affected relative to consonants. In the letter-in-word judgment task, the largest group differences were observed in conditions requiring access to orthographic knowledge when phonological cues were non-informative.

Discussion:

Together, these findings indicate that written language impairment in lvPPA cannot be explained by a primary phonological deficit alone but instead reflects vulnerability of abstract orthographic representations, particularly under conditions of increased orthographic ambiguity. Identifying tasks that require access to orthographic knowledge independently of phonological support may therefore provide sensitive markers for characterizing written language impairment in lvPPA and contribute to a more precise understanding of language network dysfunction in neurodegenerative disease.

1 Introduction

Primary progressive aphasia is a neurodegenerative syndrome characterized by a gradual and selective decline of language functions, with relative preservation of other cognitive domains in the early stages of the disease (Mesulam, 2003, 2013). Current consensus criteria distinguish three main clinical variants: the non-fluent/agrammatic variant, the semantic variant, and the logopenic variant (Gorno-Tempini et al., 2011). Among these, the logopenic variant of primary progressive aphasia (lvPPA) is typically associated with impaired word retrieval, reduced word and sentence repetition, and deficits in phonological short-term memory, reflecting predominant involvement of left temporoparietal language networks (Gorno-Tempini et al., 2008; Henry and Gorno-Tempini, 2010).

While spoken language impairments in lvPPA have been extensively documented, written language production has received comparatively less systematic attention. This is notable given that writing is frequently affected early in the disease course and may provide sensitive markers of underlying language system disruption (Graham, 2014; Neophytou et al., 2019). Moreover, written language offers a unique window onto the organization of lexical, phonological, and orthographic representations, allowing dissociation of processing components that are more difficult to isolate in spoken production.

From a neuropsychological perspective, written word production is supported by multiple interacting mechanisms, including lexical–semantic access, phonological processing, phoneme–grapheme conversion, access to abstract orthographic representations, and the temporary maintenance of graphemic information during serial output (Tainturier and Rapp, 2001; Rapp and Fischer-Baum, 2014; Rapp and Purcell, 2019). Disruption at different levels of this system gives rise to distinct patterns of acquired agraphia, traditionally classified as phonological, surface, or deep agraphia (Denes et al., 2020). Although these classifications were originally derived from focal lesion studies, they continue to provide a useful conceptual framework for characterizing writing impairments in neurodegenerative conditions, including PPA (Graham, 2000, 2014).

Existing studies of spelling and writing in lvPPA have reported heterogeneous profiles that do not map neatly onto a single agraphic syndrome. Group studies and case series have documented impairments affecting both word and non-word spelling, often accompanied by a mixture of phonologically plausible errors (e.g., bato for bateau, “boat”) and non-phonologically plausible errors (e.g., bameau for bateau) (Sepelyak et al., 2011; Shim et al., 2012; Faria et al., 2013). Such patterns suggest that spelling deficits in lvPPA may reflect disruptions at multiple levels of the spelling system rather than a selective breakdown of a single route.

Importantly, these written language impairments occur within the broader language profile of lvPPA, which is characterized by impaired phonological short-term memory, reduced repetition, and word-finding difficulties associated with temporoparietal network degeneration (Gorno-Tempini et al., 2008; Henry and Gorno-Tempini, 2010). Because written word production requires coordinated access to phonological, lexical, and orthographic representations, deficits affecting phonological processing or lexical retrieval may propagate to written production, while additional impairments in orthographic representations may further compromise spelling accuracy.

Sensitivity to psycholinguistic variables such as lexical frequency, orthographic regularity, word length, and semantic properties has also been observed, highlighting the contribution of lexical–orthographic knowledge to written production deficits in lvPPA (Neophytou et al., 2019; Carroll-Duhigg et al., 2025). Together, these findings suggest that agraphia in lvPPA reflects disruption of multiple interacting components of the writing system rather than a uniform breakdown of a single spelling route.

However, much of the existing literature has relied on global accuracy measures or broad lexicality contrasts (e.g., words vs. non-words), which may obscure more selective vulnerabilities within orthographic processing. In particular, the availability and stability of abstract orthographic representations in lvPPA remain poorly understood, especially under conditions in which phonological information provides limited or unreliable support. According to influential models of spelling, accurate written production relies on access to abstract orthographic representations that cannot be derived directly from phonology, particularly in languages characterized by inconsistent phoneme–grapheme mappings (Rapp and Caramazza, 1997). Although sublexical phoneme–grapheme conversion processes are sensitive to the statistical regularities of the spelling system (Ziegler et al., 1996; Peereman and Content, 1999), they are insufficient to support accurate spelling in many cases, especially when mappings are of low consistency, thereby necessitating the contribution of lexical–orthographic knowledge (Barry and Seymour, 1988).

This issue is especially relevant in French, where phoneme to grapheme correspondences is highly inconsistent in the phonology-to-orthography direction (Ziegler et al., 1996), and where vowel phonemes are associated with substantial orthographic ambiguity (Jaffré and Fayol, 2013). For example, the French vowel phoneme /o/ can correspond to several orthographic forms (e.g., o, au, eau), and the phoneme /ε/ may be spelled in, ain, or ein. Such variability increases the demands placed on orthographic selection processes because phonological information alone does not uniquely determine the correct spelling. As a result, vowel spelling places disproportionate demands on orthographic selection processes and on the integrity of abstract orthographic representations, beyond what can be achieved through sublexical phonological conversion alone. Consistent with this view, neuropsychological evidence indicates that consonants and vowels are represented as distinct categories within the orthographic system and can be selectively impaired following brain damage (Buchwald and Rapp, 2006). Vulnerability at this level may therefore be particularly informative for identifying orthographic impairments in lvPPA.

Tasks that explicitly manipulate the relationship between phonological and orthographic information offer a powerful means of probing these mechanisms. Letter-based judgment tasks, in which participants decide whether a given letter belongs to the written form of a word, have been shown to tap orthographic knowledge independently of overt written output (Macoir and Beland, 2004; Purcell and Rapp, 2013). By crossing the presence or absence of a target letter in the orthographic form with the presence or absence of the corresponding phoneme in the phonological form, such tasks allow fine-grained examination of conditions in which orthographic decisions can or cannot be supported by phonological cues. Despite their theoretical relevance, these paradigms have rarely been applied to the study of written language impairment in lvPPA. In particular, identifying conditions under which orthographic representations must be accessed independently of phonological support may provide especially sensitive markers of written language impairment in lvPPA.

The present study aimed to provide a detailed characterization of written word production deficits in lvPPA by combining global task-level analyses with targeted manipulations designed to probe orthographic processing and phonology–orthography interactions. Using a battery of written production tasks, including written picture naming, written word dictation, written non-word dictation, grapheme dictation, and a picture-based letter-in-word judgment task, we examined both overall accuracy and qualitative error patterns in individuals with lvPPA compared to healthy control participants. Particular emphasis was placed on dissociating lexical and sublexical spelling mechanisms, assessing the influence of psycholinguistic variables known to tax orthographic representations, and identifying conditions in which access to orthographic knowledge was required independently of phonological support.

Based on current cognitive models of spelling (Rapp and Caramazza, 1997; Tainturier and Rapp, 2001) and previous observations in lvPPA, two main hypotheses were formulated. First, we predicted that individuals with lvPPA would show disproportionate difficulty in tasks requiring access to orthographic knowledge when phonological information provides limited support. Second, given the high degree of phoneme–grapheme ambiguity associated with French vowel phonemes, we expected vowel graphemes to be particularly vulnerable relative to consonant graphemes. These predictions guided the selection of tasks designed to dissociate phonological and orthographic contributions to written word production.

By adopting this fine-grained approach, the present study also sought to clarify the nature of agraphia in lvPPA and to determine whether specific components of the orthographic system, most notably those involved in vowel processing and phonology-independent orthographic access, constitute points of particular vulnerability in this variant. Beyond contributing to cognitive models of written language production, this work also aimed to inform clinical assessment by identifying tasks and conditions that were especially sensitive to written language impairment in lvPPA.

2 Method2.1 Participants

Thirteen individuals meeting diagnostic criteria for the logopenic variant of primary progressive aphasia (lvPPA) were initially recruited through the Clinique Interdisciplinaire de Mémoire (CIME) at the Centre Hospitalier Universitaire (CHU) de Québec. All participants were evaluated and diagnosed by a neurologist specializing in neurodegenerative cognitive disorders, in accordance with the consensus criteria proposed by Gorno-Tempini et al. (2011). Diagnosis was based on a comprehensive clinical language assessment and neurological examination and was supported by available biomarkers in 12 of the 13 patients.

Diagnostic support was obtained from structural and/or functional neuroimaging findings and, when available, biological biomarkers derived from cerebrospinal fluid or plasma. Specifically, four patients underwent FDG-PET (fluorodeoxyglucose positron emission tomography); two underwent FDG-PET combined with lumbar puncture; one underwent FDG-PET combined with PrecivityAD2 testing; three underwent computed tomography (CT); one underwent CT combined with lumbar puncture; and one underwent magnetic resonance imaging (MRI). When available, neuroimaging findings were consistent with the typical pattern associated with lvPPA, including predominant involvement of left temporoparietal language regions. In FDG-PET examinations, this corresponded to reduced metabolic activity in left temporoparietal areas, whereas structural imaging (MRI or CT) revealed atrophy patterns compatible with degeneration of the same network. These imaging findings were interpreted by the treating neurologist as supportive of the clinical diagnosis in accordance with the consensus criteria for primary progressive aphasia (Gorno-Tempini et al., 2011).

All participants exhibited a clinical profile characteristic of lvPPA, including prominent word-finding difficulties, impaired word and sentence repetition, and reduced phonological short-term memory, in the absence of motor speech disorders or predominant semantic impairment. A comprehensive medical and psychiatric history was obtained for each participant. Individuals were excluded if they had a history of neurological or cerebrovascular disease other than lvPPA, any current or past psychiatric illness as defined by the DSM-5 (American Psychiatric Association, 2013), traumatic brain injury, untreated medical or metabolic conditions (e.g., diabetes or thyroid dysfunction), prior intracranial surgery, or uncorrected auditory or visual impairments.

The comparison group consisted of 13 neurologically healthy adults recruited through the Centre de recherche CERVO. Control participants were excluded if they met any of the following criteria: (1) performance below a Z score of −1.33 on the Montreal Cognitive Assessment (MoCA; Nasreddine et al., 2005), based on normative standards established by the research team (Larouche et al., 2016); (2) performance below the fifth percentile (approximately 1.5 standard deviations below the mean) on the Detection Test for Language Impairments in Adults and the Aged (DTLA; Macoir et al., 2017, 2022); (3) presence of an active or unstable psychiatric condition; (4) history of moderate to severe traumatic brain injury; (5) prior stroke or other neurological disorder; (6) history of psychotic illness; (7) substance abuse; or (8) uncorrected auditory or visual impairment. These exclusion criteria were applied to reduce the risk of including individuals with subtle or preclinical cognitive deficits that could confound group comparisons. As reported in Table 1, the control and lvPPA groups were comparable in age, sex distribution, and years of formal education. Time since diagnosis for participants with lvPPA is also reported in Table 1. At the time of testing, patients had received their diagnosis an average of 8.75 months earlier (range = 2–23 months), indicating that most participants were assessed relatively early in the course of the disease.

VariablesHC (n = 13)lvPPA (n = 12)M (SD)min-maxM (SD)min-maxU/χ2prDemographicsAge69.85 (6.67)59–8270.08 (7.88)55–82U = 79.50.8170.050Sex (male/female)5/8–5/7–χ2(1) = 0.030.87–Education (years)14.85 (2.34)11–1814.69 (2.36)11–18U = 88.50.8540.040Time from diagnosis (months)––8.75 (6.68)2–23–––Cognitive screeningMoCA27.46 (1.76)25–3015.85 (7.57)5–30U = 151.50.0010.674DTLA95.23 (3.56)90–10069.00 (17.64)40–96U = 162.0< 0.0010.779Verbal short-term and working memoryShort-term memory (digit span forward)5.92 (0.64)5–74.31 (1.25)3–6U = 143.50.0020.593Working memory (digit span backward)4.08 (0.64)3–52.23 (1.01)0–4U = 159.0< 0.0010.749Executive functionsAlphaflex time A (sec.)8.54 (2.60)6–1521.85 (27.47)7–110U = 32.00.0070.528Alphaflex error A0.46 (0.66)0–24.23 (6.29)0–18U = 63.50.2410.211Alphaflex time B (sec.)17.77 (4.11)12–2734.62 (22.70)20–90U = 14.0< 0.0010.709Alphaflex error B0.69 (0.755)0–24.77 (4.71)0–12U = 39.00.0170.458

Sample characteristics and background cognitive measures in lvPPA and healthy control groups.

Values represent means (standard deviations) and ranges. Group comparisons were performed using Mann–Whitney U-tests due to non-normal distributions in the lvPPA group, while comparison for sex was made using the χ2 test. Effect sizes are reported as r, calculated from the standardized U statistic, with values of approximately 0.10, 0.30, and 0.50 interpreted as small, medium, and large effects, respectively. Higher scores indicate better performance except for Alphaflex completion times and error counts, for which lower values reflect better performance.

During the clinical interview, participants were also asked about their medical and educational history. None reported a history of developmental reading or writing disorders (e.g., dyslexia or dysgraphia), and the comparable levels of formal education observed in the lvPPA and control groups (Table 1) further suggest typical premorbid literacy acquisition.

One participant in the lvPPA group exhibited extremely low performance across written word production tasks. This individual was a 64-year-old man with 18 years of education who had received a diagnosis of lvPPA 15 months prior to testing. Although global cognitive screening remained preserved (MoCA = 26), language screening revealed severe impairment (DTLA = 40), accompanied by marked deficits in verbal short-term and working memory (digit span forward = 3; backward = 0) and executive functioning (Alphaflex A time = 110 s, errors = 15; Alphaflex B time = 20 s, errors = 12). Written language performance was near floor level across most tasks (written picture naming = 2/24; written word dictation = 1/47; written non-word dictation = 0/15; grapheme dictation = 8/30), although performance remained relatively preserved on the letter-in-word judgment task (19/24). Because this performance profile precluded meaningful interpretation of the targeted linguistic effects and exerted a disproportionate influence on group-level statistical analyses, this participant was excluded from inferential analyses.

Inferential statistical analyses were therefore conducted on a final sample of 12 individuals with lvPPA and 13 healthy control participants. All individuals gave written informed consent after receiving a full explanation of the study procedures. Ethical approval was obtained from the Ethics Committee for the Sector Research in Neurosciences and Mental Health (Project MP-13-2017-164).

2.2 Materials2.2.1 Cognitive screening

Overall cognitive status was examined using the MoCA (Nasreddine et al., 2005), a standardized instrument commonly employed for the detection of mild cognitive impairment. The MoCA provides a brief assessment of several cognitive domains, including attention, executive functioning, memory, language, visuospatial abilities, and temporal and spatial orientation. Administration followed standard guidelines, yielding a total score ranging from 0 to 30.

Language functioning was evaluated with the DTLA (Macoir et al., 2017, 2022). This short screening battery was specifically designed to identify language deficits associated with neurodegenerative conditions. It comprises nine tasks probing key components of language processing, such as confrontation naming, repetition of words and non-words, verbal fluency, reading and spelling, sentence comprehension, semantic processing, and an alphabetization span task indexing verbal working memory. Together, these DTLA measures provide a brief yet sensitive overview of both receptive and expressive language abilities across multiple linguistic levels.

In addition to global cognitive screening tools, tests of verbal short-term and working memory as well as executive functions were administered. Verbal short-term and working memory was assessed using the Digit Span subtest from the Wechsler Adult Intelligence Scale–Fourth Edition (WAIS-IV; Wechsler, 2008). Participants completed both the forward and backward conditions, administered according to standardized procedures. The forward condition required immediate repetition of auditorily presented digit sequences in their original order and served as an index of simple verbal span. The backward condition involved reproducing the digits in reverse order, thereby placing additional demands on mental manipulation and working memory. Performance was quantified as the maximum sequence length correctly recalled for each condition.

Executive functioning was examined with the Alphaflex (Grotz et al., 2018), a brief orally administered task designed to assess reactive mental flexibility with minimal visuomotor involvement and comprising two conditions. In Part A, participants recited the alphabet aloud as quickly and accurately as possible, providing a baseline measure of an overlearned verbal sequence. In Part B, they were instructed to produce only every second letter of the alphabet (e.g., A, C, E), a manipulation that increases demands on inhibitory control, sequencing, and internal monitoring. For each condition, completion time, and number of errors were recorded.

Together, these measures provided a comprehensive characterization of the participants' global cognitive, language, and executive profiles.

2.2.2 Battery of written word production tasks

Written language production was assessed using a targeted battery designed specifically to examine key components of written word production in lvPPA. The selected tasks probed complementary stages of the spelling architecture, including lexical–semantic access, phonological-to-orthographic conversion, orthographic output processes, and serial grapheme selection. Together, these tasks allowed for a fine-grained characterization of both accuracy and error patterns in written production.

The battery comprised five tasks: written picture naming, written word dictation, written non-word dictation, grapheme dictation, and a letter-in-word judgment task based on picture stimuli. Although no written output is required, successful performance on the letter-in-word judgment task presupposes the internal generation and evaluation of the target word's orthographic representation, thereby engaging central spelling processes involved in written word production rather than mere visual or recognition-based mechanisms. A detailed overview of each task, including the number of items, controlled linguistic parameters, scoring procedures, and targeted cognitive processes, is provided in Table 2.

TaskDescriptionNumber of itemsControlled parametersPrimary cognitive processes targetedScoringWritten picture namingWriting the name of visually presented objects24Lexical frequency; orthographic regularity; syllabic lengthLexical–semantic access; orthographic output lexicon; graphemic buffer; serial grapheme production1 point per correctly written word (max = 24); error analysisWritten word dictationWriting auditorily presented real words47Lexical frequency; imageability; orthographic regularity; syllabic lengthLexical and sublexical spelling routes; phonological-to-orthographic conversion; orthographic working memory1 point per correctly spelled word (max = 47); error analysisWritten non-word dictationWriting auditorily presented non-words15Syllabic lengthSublexical phonological-to-orthographic conversion; graphemic buffer1 point per acceptable spelling (max = 15); error analysisGrapheme dictationWriting graphemes corresponding to spoken phonemes30Grapheme typePhoneme-to-grapheme conversion; sublexical spelling mechanisms1 point per correct response (max = 30); error analysisLetter-in-word judgment (picture-based)Judging whether a given letter belongs to the written form of a pictured word24 (4 × 6)Lexical frequency; presence/absence of target letter and corresponding phonemeOrthographic output lexicon access; orthographic knowledge without written output1 point per correct judgment (max = 24)

Overview of the written word production tasks, controlled linguistic parameters, and targeted cognitive processes.

All tasks were administered following standardized procedures. Accuracy scores reflect the number of correct responses. In addition to quantitative scores, written responses were transcribed verbatim to allow qualitative classification of error types. The letter-in-word judgment task was subdivided into four conditions based on the presence or absence of the target letter in the orthographic form and the corresponding phoneme in the phonological form (6 items per condition): (a) phoneme present/letter present, (b) phoneme absent/letter present, (c) phoneme present/letter absent, and (d) phoneme absent/letter absent.

Briefly, written picture naming assessed access to orthographic representations from visual–semantic input, while written word dictation examined spelling mechanisms under auditory input, with systematic manipulation of lexical frequency, orthographic regularity, syllabic length, and imageability. Written non-word dictation and grapheme dictation were included to specifically evaluate sublexical phonological-to-orthographic conversion processes independent of lexical knowledge. Finally, the letter-in-word judgment task provided a measure of orthographic knowledge without requiring full written output, thereby isolating access to the orthographic output lexicon.

All tasks were administered following standardized procedures. Accuracy scores were computed for each task, and all responses were transcribed verbatim to enable qualitative classification of error types, including phonologically plausible and non-phonologically plausible errors, substitutions, perseverations, and omissions.

2.3 Procedure

All participants were evaluated individually either in a quiet room in their home or in a dedicated testing space at the research center. Assessments were carried out by trained research assistants under the supervision of the primary investigator, and each evaluation was completed in a single session lasting approximately 90 min.

Testing followed a standardized sequence to minimize potential task interference. Participants first completed the MoCA and the DTLA, after which they were administered the digit span tasks followed by the Alphaflex. The session concluded with the administration of the battery of written word production tasks. All responses were scored immediately according to the standardized scoring criteria, and data were subsequently entered into a secure database for analysis. Recruitment and testing occurred between February 2023 and October 2025.

2.4 Statistical analyses

All statistical analyses were conducted using Jamovi (version 2.6.45; The Jamovi Project, 2025). Given the relatively small sample size and the presence of non-normal and bounded score distributions across several measures, non-parametric statistical methods were used throughout.

Between-group comparisons were first conducted on cognitive screening and diagnostic measures, including global cognitive status (MoCA), language screening (DTLA), verbal short-term and working memory (digit span forward and backward), and executive function measures (Alphaflex completion times and error counts for Parts A and B). Group differences were examined using Mann–Whitney U-tests due to non-normal score distributions. Effect sizes were reported as r. These analyses were intended to characterize group-level cognitive differences and were not adjusted for multiple comparisons.

2.4.1 Quantitative analyses of overall task performance

Between-group differences in written word production were first examined using total accuracy scores for each task, including written picture naming, written word dictation, written non-word dictation, grapheme dictation, and the letter-in-word judgment task. Group comparisons were performed using Mann–Whitney U-tests. Effect sizes were reported as r, calculated from the standardized U statistic, with values of approximately 0.10, 0.30, and 0.50 interpreted as small, medium, and large effects, respectively. To control for multiple comparisons across tasks, p-values were adjusted using the Holm–Bonferroni procedure.

In a second step, the letter-in-word judgment task was further decomposed into four subcategories defined by the presence or absence of the target letter in the orthographic form and the corresponding phoneme in the phonological form. Between-group comparisons were conducted separately for each subcategory using Mann–Whitney U-tests. To control for multiple comparisons within this set of related analyses, p-values were adjusted using the Holm–Bonferroni procedure applied across the four subcategories.

Finally, to further characterize performance on grapheme dictation, accuracy scores were examined separately for consonant and vowel graphemes. Between-group comparisons were conducted using Mann–Whitney U-tests and were interpreted as secondary, targeted analyses.

2.4.2 Influence of psycholinguistic parameters

To investigate the influence of psycholinguistic variables on written production performance, additional analyses were conducted for tasks in which these parameters were systematically manipulated. Within each group, the effects of lexical frequency, orthographic regularity, and syllabic length were examined using Wilcoxon signed-rank tests. To assess whether the magnitude of these effects differed between groups, participant-level difference scores (e.g., low-frequency minus high-frequency accuracy) were computed and compared across groups using Mann–Whitney U-tests. This approach allowed evaluation of group differences in sensitivity to psycholinguistic parameters while avoiding parametric interaction models that would be inappropriate given sample size and distributional constraints.

For written non-word dictation and grapheme dictation, which primarily probe sublexical phonological-to-orthographic conversion processes, between-group comparisons were performed on total accuracy scores using Mann–Whitney U-tests. Where applicable, exploratory within-group analyses examined the effect of syllabic length using Wilcoxon signed-rank tests.

2.4.3 Qualitative error analyses

Written responses were further analyzed qualitatively using predefined error categories. Operational definitions and examples for each error type are provided in Table 3. For each participant, error proportions were calculated relative to the total number of errors produced. Given the sparse and zero-inflated distributions typical of error data, qualitative analyses were primarily descriptive. Targeted between-group comparisons were conducted using Mann–Whitney U-tests for theoretically motivated error categories. Effect sizes (r) were reported to facilitate interpretation.

Error categoryOperational definitionExample (French)Corresponding example in EnglishPhonologically plausible errorsIncorrect spellings that preserve the phonological form of the target and conform to phoneme-to-grapheme conversion rules.Bateau (“boat”) → batoknife → nifeNon-phonologically plausible errorsIncorrect spellings that cannot be derived from the target's phonology and violate standard phoneme-to-grapheme correspondences.Bateau (“boat”) → bameauboat → boafSemantic errorsProduction of a real word that is semantically related to the target but phonologically and orthographically unrelated.chien (“dog”) → chat (‘cat)dog → catFormal visual paragraphiasProduction of a real word sharing visual–orthographic similarity with the target without semantic or phonological relatedness.table (“table”) → cable (cable)house → horseLexicalizationsProduction of a real word in response to a non-word target.plade → plage (beach)snorp → snoreOther pooled error typesResponses not fitting the above categories, including no-response errors or visually driven errors in picture naming.pomme (“apple”) → ∅
banane (“banana”) → lune (“moon”)apple → ∅ banana → moon

Qualitative error categories used for the analysis of written word production.

Error categories were defined a priori and were mutually exclusive. Each written response was assigned to a single error category.

3 Results

Results are presented in four parts: between-group differences in cognitive screening and diagnostic measures, between-group differences in overall written word production performance, effects of psycholinguistic parameters, and qualitative analyses of error patterns.

3.1 Between-group differences in cognitive screening and diagnostic measures

As shown in Table 1, participants with lvPPA differed significantly from healthy controls on global cognitive screening and language measures. The lvPPA group showed significantly lower performance on the MoCA and the DTLA, reflecting impairments in overall cognitive functioning and language abilities.

Group differences were also observed on measures of verbal short-term and working memory, with participants with lvPPA performing significantly worse than controls on both digit span forward and digit span backward.

Finally, executive function measures revealed significantly longer completion times on the Alphaflex, particularly in the more demanding Part B, in the lvPPA group relative to controls. No reliable group differences were observed for error counts.

3.2 Between-group differences in written word production

Group comparisons on total accuracy scores revealed widespread impairments in written word production in the lvPPA group relative to healthy controls (Table 4). Participants with lvPPA performed significantly worse than controls on written picture naming, written word dictation, grapheme dictation, and the letter-in-word judgment task, with medium to large effect sizes. These group differences remained significant after correction for multiple comparisons using the Holm–Bonferroni procedure.

Task (max)HC mean (SD)HC min–maxlvPPA mean (SD)lvPPA min–maxUprp (Holm)Written picture naming (24)22.85 (1.07)21–2418.92 (4.17)10–23133.00.003**0.5980.008**Written word dictation (47)44.77 (1.96)42–4738.17 (5.24)29–46142.0< 0.001***0.6960.003**Written non-word dictation (15)10.85 (1.95)6–149.08 (2.39)5–13111.50.0690.3640.069Grapheme dictation (30)27.62 (1.89)23–3024.75 (3.25)19–30122.00.017*0.4790.034*- Consonant graphemes (17)14.85 (1.99)10–1713.58 (2.02)10–17109.00.0930.34–- Vowel graphemes (13)