Psycholinguistic Variables and Spelling Accuracy for People with Logopenic Variant Primary Progressive Aphasia: A Cross-Sectional Study

Abstract

Background:

Logopenic variant of primary progressive aphasia (lvPPA) has been identified as more challenging to diagnose compared to other PPA-variants. A more nuanced characterization of lvPPA symptoms may be needed to facilitate accurate and timely diagnoses. Spelling engages a range of neural networks, some of which are more susceptible to lvPPA-atrophy than others, resulting in variable spelling outcomes, depending on which systems are recruited for the task. This variability may be influenced by the specific psycholinguistic characteristics of the word being spelled.

Aims:

The primary aim of this study is to examine the effects of psycholinguistic variables of words on spelling accuracy for people with lvPPA.

Methods and Procedures:

Sixteen participants completed an abbreviated spelling assessment of 40 words taken from the Arizona Battery of Reading and Spelling (ABRS). Thirteen psycholinguistic variables, potentially relevant to orthographic, phonological, and/or lexical-semantic processing, were identified in the literature. Values for psycholinguistic variables of interest for each word were pulled from the SCOPE: South Carolina Psycholinguistic Metabase. Multilevel logistic regression modeling was used to evaluate the effect of psycholinguistic variables (at word-level) on spelling accuracy, as well as participant-level clustering effects. The main outcome variable was spelling accuracy (accurate/inaccurate). The final parsimonious model included regularity (REG), age-of-acquisition (AOA), orthographic (ONLD) and phonological neighborhood (PNLD) measured by Levenshtein’s distance (LD), and emotional valence (VAL) as predictors. Predicted probabilities were calculated to estimate effect sizes for each significant variable.

Results:

REG, AOA, ONLD, and VAL were significant predictors of lvPPA spelling accuracy. The effect of PNLD was not significant. REG had the largest effect on spelling accuracy, with a 93% probability of spelling accuracy on REG words compared to a 67% probability on irregular (IRREG) words. Words acquired at an earlier age (low AOA), as well as those with denser orthographic neighborhoods (lower ONLD) and positive emotional valence (high VAL) were significantly associated with greater probability of spelling accuracy.

Conclusions:

Differences in psycholinguistic variables of words significantly impact spelling accuracy for people with lvPPA. Findings illustrate the utility and value of spelling assessments in PPA diagnosis. Further research is needed to determine if these effects are distinct to lvPPA.

Introduction

Primary Progressive Aphasia

Primary Progressive Aphasia (PPA) is a dementia syndrome, or a heterogeneous group of symptoms related to progressive loss of speech-language function. It is caused by neurodegenerative proteinopathies commonly associated with frontotemporal lobar degeneration (FTLD) (e.g. Pick’s Disease) or dementia with Lewy bodies (DLB), and/or the presence of amyloid-beta plaque and intracellular neurofibrillary tangles, as in Alzheimer’s disease (AD) (Gorno-Tempini et al., 2016; Harris & Jones, 2014; Ruksenaite et al., 2021). In PPA, these pathophysiological processes initially target the language centers of the brain, before slowly spreading to other areas. Patterns of cortical atrophy, along with associated clinical features (i.e., speech-language symptoms) and underlying histopathology, are used to differentially diagnose three main PPA variants including non-fluent/agrammatic (nfvPPA), semantic (svPPA), and logopenic (lvPPA), as well as a motor-speech programming variant, primary progressive apraxia of speech (PPAOS) (Duffy et al., 2021; Gorno-Tempini et al., 2016).

Diagnosis is not limited to these four variants (e.g., mixed, unclassifiable, etc.) and heterogeneity of symptoms both across and within variants adds additional complexity to the diagnostic process (Santi et al., 2024; Teichmann, 2021). Symptom variability is representative of differences in the location and spread of neurodegeneration but is also moderated by compensatory neural processes (i.e., synaptic reorganization) that may differ across individuals (Gorno-Tempini et al., 2016). Some individuals meet diagnostic criteria for more than one variant (“PPA-extended” or “mixed PPA”), which may be indicative of later stage disease when atrophy has spread to surrounding areas of the brain (Rogalski et al., 2011; Ulugut et al., 2022). In contrast, some individuals with PPA do not meet the required criteria for a particular variant and receive an unspecified diagnosis (e.g., “PPA-unclassified”) (Harris et al., 2013); this may be due to the presence of milder symptoms during earlier stages of disease progression (Tippett, 2020) and illustrates the challenge to early diagnosis.

Challenges to Diagnosing Logopenic Variant

LvPPA has been identified as more challenging to diagnose compared to other variants (Gorno-Tempini et al., 2016; Harris et al., 2013; Sajjadi et al., 2012). Although commonly associated with AD pathology, lvPPA cases have also been linked to FTLD TDP-43 type A pathology, DLB, and Creutzfeldt-Jakob disease (Harris et al., 2013). Atrophy generally begins in the left temporal-parietal area, often including the angular gyrus (AG), supramarginal gyrus (SMG), posterior superior temporal gyrus (STG), inferior temporal gyrus (ITG), and inferior parietal cortex (IPC) (Mesulam et al., 2014; Petroi et al., 2020). Volume loss in these areas is correlated with common clinical features of lvPPA, including difficulty with repetition, word retrieval, and confrontation naming, as well as the presence of phonemic paraphasias, but spared semantic knowledge and motor-speech ability (Gorno-Tempini et al., 2016).

Some researchers have argued that lvPPA is largely a diagnosis of exclusion. Because the core features of lvPPA (Gorno-Tempini et al., 2011) partially overlap with other variants (e.g., “impaired naming” with features of svPPA; “impaired repetition” with features of nfvPPA), diagnosticians must rely heavily on the exclusion of other features to arrive at an lvPPA diagnosis (e.g., agrammatism/motor speech deficits for nfvPPA; motor speech deficits for PPAOS; semantic knowledge for svPPA) (Harris et al., 2013; Sajjadi et al., 2012). These overlapping features may put people with lvPPA at increased risk of misdiagnosis. Additionally, word-finding pauses, common among individuals with lvPPA, may be misinterpreted as effortful and halted speech, which is a feature of nfvPPA. Similarly, impaired sentence comprehension may be present in all three variants but attributed to deficits in grammar for people with nfvPPA, loss of semantic knowledge for people with svPPA, and auditory/phonological-working memory deficits in lvPPA (Gorno-Tempini et al., 2016; Mesulam et al., 2014). LvPPA is also considered to have a faster rate of progression compared to other variants and is associated with greater cognitive impairment (Ulugut et al., 2022). For all of these reasons, it is necessary to better understand and identify lvPPA-specific features to improve differentiation and support more accurate/earlier diagnoses.

Spelling Characteristics of Logopenic Variant

One prominent feature of lvPPA is a breakdown in phonological networks, which are linked to impaired repetition, naming, auditory comprehension (of longer sentences), phonological paraphasias, and difficulties with reading and spelling (Gorno-Tempini et al., 2016; Petroi et al., 2020). Spelling, in particular, provides a rich context for analyzing nuanced differences across PPA variants, as it involves a complex interaction of multiple systems (e.g., orthographic, phonological, lexical-semantic), engaging a wide range of distinct areas and pathways in the brain that correspond with specific spelling-related tasks (Neophytou et al., 2019; Sepelyak 2011; Shim et al., 2012). For example, spelling irregular words (i.e., words with low sound-letter correspondence, such as “light”), which rely on lexical-semantic, or “whole-word” processing, has been found to activate the left inferior frontal gyrus (IFG) pars orbitalis, left middle temporal gyrus (MTG), right STG, left SMG, and AG – many of the areas commonly affected by lvPPA (Henry et al., 2012; Mesulam et al., 2014; Norton et al., 2007; Petroi et al., 2020). Conversely, spelling “regular” words (i.e., words with strong sound-letter correspondence, as in “hard” and “spring”) may engage phonological, or “sublexical” processes needed for “sounding out words”, with activation seen primarily in the left posterior STG, right precentral gyrus (PCG), and right fusiform gyrus (FG) (Norton et al., 2007). The left posterior STG, another area commonly impacted in people with lvPPA, is also involved in spelling non-words and believed to play an especially important role in phonological processing (Mesulam et al., 2014; Norton et al., 2007; Petroi et al., 2020). However, the relationship between regular/irregular word spelling and sublexical/lexical recruitment is not necessarily mutually exclusive, as spelling regular words may leverage either lexical or sublexical routes or both (Beeson et al., 2008; Henry et al., 2012). In contrast, spelling non-words, for which whole-word representations have not been previously stored, relies primarily on sound-to-letter conversion via phonological/sublexical pathways.

Previous studies have reported that individuals with lvPPA have greater difficulty spelling non-words compared to real words (Europa et al., 2020; Henry et al., 2015) and that this effect of lexicality is more prominent among those with lvPPA compared to other PPA variants (Henry et al., 2015). However, other studies have conversely suggested that the difference in spelling accuracy between non-words and real words is bigger among individuals with nfvPPA (i.e., having significantly greater accuracy on real words), and those with lvPPA exhibit comparable difficulty on both non-words and real words (Neophytou et al., 2019; Shim et al., 2012). Although conflicting findings have been reported for lvPPA and nfvPPA, there is general agreement across studies that individuals with svPPA show a prominent lexicality effect in the opposite direction, with greater accuracy on non-words compared to real words, specifically irregular real words (Henry et al., 2015; Neophytou et al., 2019; Shim et al., 2012).

Although difficulty with spelling has been demonstrated across all three PPA variants, spelling impairment (i.e., “surface dysgraphia”) is listed as a (secondary) diagnostic feature only for svPPA; there is no other mention of spelling impairments in consensus criteria for nfvPPA or lvPPA (Gorno-Tempini et al., 2011). As a result, clinicians may overly rely on “difficulty spelling irregular real words” as evidence of surface dysgraphia (i.e., use of sound-to-letter conversion to spell irregular real words) to differentially diagnose svPPA. This is especially concerning, given evidence that shows individuals with lvPPA may also exhibit signs of surface dysgraphia (Graham, 2014; Sepelyak et al., 2011).

Given the diagnostic ambiguity surrounding surface dysgraphia in svPPA and lvPPA, as well as conflicting reports on lexicality effects in lvPPA and nfvPPA, lexicality on its own may not be a reliable measure for differentiating lvPPA from other variants. As such, the present study examines the effect of a wider range of psycholinguistic variables on spelling accuracy, with the goal of identifying key word-level differences that extend beyond lexicality and provide a more nuanced characterization of real word spelling impairment in lvPPA. While non-word spelling should be a critical component of spelling assessment, particularly for differentiating svPPA from other variants, clinical attention remains primarily focused on difficulties spelling irregular real words as being indicative of svPPA. This emphasis on real word spelling warrants further consideration to encourage a more nuanced understanding of spelling profiles in PPA.

Role of Psycholinguistic Variables in Spelling Accuracy

Two studies evaluated the impact of concreteness, frequency, imageability, neighborhood density (orthographic and phonological), word class, and word length on spelling accuracy across PPA variants, both of which reported non-significant effects for lvPPA (Neophytou et al., 2019; Sepelyak et al., 2011), indicating a need to identify which variables have the strongest effect on spelling accuracy for people with lvPPA, and evaluate the effect of these variables in the context of other PPA-variants. The influence of age-of-acquisition, arousal, valence, frequency, lexical-semantic decision-making response time, and neighborhood density (orthographic, phonological, and/or semantic) have been studied in non-spelling contexts, with orthographic neighborhood density showing a significant effect on lexical-semantic decision-making (i.e., “is this a word?” – yes/no) for both lvPPA and nfvPPA, and age-of-acquisition appearing to be a strong predictor of naming accuracy in lvPPA (Jebahi et al., 2023; Vonk et al., 2019). Jebahi and colleagues (2023) also reported that latent factors for the Boston Naming Test, “lexical-semantic usage” (indicated by frequency, familiarity, semantic neighborhood density, age-of-acquisition) and “emotional intensity” (indicated by arousal and valence) were associated with significant differences in naming accuracy among lvPPA. Additionally, Vonk et al. (2019) found that people with lvPPA made lexical decisions more quickly and more accurately for higher frequency words that also had more dense orthographic neighborhoods (as determined by Levenshtein’s distance [LD]). Given the variability of tasks, methodology, results reported across studies, further research is needed to evaluate the effect of these variables, specifically in the context of lvPPA spelling.

Study Aims

The primary aim of this analysis is to identify and describe which psycholinguistic variables significantly predict spelling accuracy of real words for people with lvPPA. This study is the first to evaluate a wide range of psycholinguistic variables in one cohesive analysis to determine their significance and effect on spelling accuracy for people with lvPPA.

Materials and Methods

Study Design

This is a preliminary analysis and report of a subset of data pulled from a larger cross-sectional observational IRB-approved study investigating the communication abilities and life participation of people with PPA. Spelling ability was evaluated as part of a comprehensive battery of speech-language and psychosocial assessments. Data collection began in 2016 and is ongoing. Findings are reported using “Strengthening the Reporting of Observational Studies in Epidemiology” (STROBE) guidelines (Vandenbroucke et al., 2007).

Participants

Inclusion criteria included a minimum age of 18 years; adequate auditory and visual acuity; English as the primary language used at home, work/school, or community, at time of testing or prior to disease onset; a confirmed diagnosis of PPA per a neurologist, and a variant diagnosis of lvPPA made by a speech-language pathologist (SLP) and confirmed by the neurologist. If the participant did not have a variant diagnosis at the time of enrollment, a diagnosis was provided by a research SLP in the lab using guidelines provided by Gorno-Tempini and colleagues (2011) and confirmed with the neurologist. Exclusion criteria included a primary diagnosis of brain injury not related to progressive disease (e.g., TBI; stroke) or other neurodegenerative condition without speech-language involvement (e.g., non-language-dominant dementia; Parkinsons). Participants were recruited from a wide range of sources (e.g., direct referral from neurologists, word-of-mouth, social media, and IRB-approved flyers posted throughout the community). Participants made initial contact, after which researchers followed-up using IRB-approved phone and email scripts to determine eligibility and schedule the assessment.

Spelling Assessment

All participants were assessed at the University of New Mexico Neuroscience of Rehabilitation Lab (“Brain Crew”) or in the participant’s home setting, during which a spelling assessment from the Arizona Battery of Reading and Spelling (ABRS) (Beeson et al., 2010; Henry et al., 2007; Rapcsak et al., 2009) was administered. The complete ABRS includes two 40-word lists that can be interchangeably used for spelling or reading, each consisting of 20 regular (REG) words with predictable sound-letter correspondence (e.g., broom) and 20 irregular (IRREG) words that do not have reliable sound-letter correspondence (e.g., friend), and a third list consisting of non-words (e.g., flig). Within each set of regular and irregular words, there are 10 high frequency words and 10 low frequency words.

The 40-word list selected for spelling assessment for this study is the one that begins with “broom” and ends with “plastic”. Each word was presented verbally and participants were asked to repeat the word before attempting to spell it. Spelling accuracy was scored by counting the number of correctly spelled words, for a total possible score of 40; self-corrected responses were also counted as correct. Scoring was conducted by a trained rater with access to video recordings and paper copies of spelling responses. A second rater was used to independently verify scoring. Assessors were SLPs, SLP-Clinical Fellows, or trained and supervised 2^nd year graduate student clinicians.

Values for each psycholinguistic variable, for each word in the list, were pulled from SCOPE: South Carolina Psycholinguistic Metabase (Gao et al., 2022). The database includes 245 variables organized into categories such as orthographic, phonological, semantic, etc. Values are derived from numerous large-scale/combined studies investigating the psycholinguistic qualities of over 40,000 words. After entering in a list of words and selecting the variables of interest, the system generates a .csv file of your word list along with values for each selected variable.

Variables

The main outcome variable in this study is spelling accuracy at word-level across participants (i.e., binary outcome indicating whether or not a word was spelled accurately). Participant-level data, including sex, age, years of education, and time since onset (years) were evaluated as potential covariates. Predictors included “general” psycholinguistic variables (e.g., regularity [REG]; frequency [FREQ]; age of acquisition [AOA]; familiarity [FAM]); orthographic variables (e.g., orthographic neighborhood density measured with Levenshtein’s distance [ONLD]; number of letters [NLETT]), phonological variables (e.g., phonological neighborhood measured with Levenshtein’s distance [PNLD]; number of phonemes [NPHON]), semantic variables (e.g., semantic neighborhood density [SND]; imageability [IMAG]; concreteness [CONCRETE]; arousal [AROUS]; valence [VAL]), and one response variable (e.g., auditory lexical decision response time - scaled version [AUDLEX_Z]) (see Table 1 for a detailed description of each predictor variable). These variables were chosen based on their inclusion in prior research (Henry et al., 2012; Jebahi et al., 2023; Neophtou et al., 2019; Sepelyak et al., 2011; Vonk et al., 2019), as well as the underlying theory that age-of-acquisition, frequency of use, the rules/conventions used to spell a word (i.e., regularity), level of familiarity, its orthographic and phonological composition, similarity to other words, meaning/salience, and lexical-semantic access, all play an important role in a person’s ability to spell a word.

Table 1.

Description of Psycholinguistic Variables (in alphabetical order)

Psycholinguistic Variable (“SCOPE reference”)	Description	Citation
AOA (“AoA_Kuper”)	Average response from participants asked to enter the age (in years) when they thought they had learned the word.	Kuperman et al., 2012
AROUS (“Arousal_NRC”)	Low/High emotional arousal associated with a word from 0 (low) to 1 (high)	Mohammad & Turney (2010)
AUDLEX_Z (“LexicalD_RT_A_AELP_z”)	The mean auditory lexical decision latency (in msec) for a particular word from the Auditory English Lexicon Project (scaled z-score).	Goh et al., 2020
CONCRETE (“Conc_Brys”)	Degree to which word can be experienced by senses from 1 (abstract) to 5 (concrete).	Brysbaert, M., 2014
FAM (“Prevalence_Brys”)	Participants provided a list of words and nonwords and asked whether or not they knew the word. Range of prevalence scores is −2.576 to +2.576.	Brysbaert et al., 2019
FREQ (“Freq_SUBTLEXUS”)	Log10 frequency norms based on SUBTLEXus corpus.	Brysbaert & New, 2009
IMAG (“Imag_Composite”)	Degree of effort needed to generate a mental image of the word.	Graves et al., 2010
NLETT (“NLett”)	The number of letters in a word.	(Count variable)
NPHON (“NPhon”)	The number of phonemes in a word.	Balota et al., 2007
ONLD (“OLD20”)	The mean distance of a word to its 20 closest orthographic neighbors based on Levenshtein’s distance (LD). LD is based on how many letter substitutions, deletions, etc. are required for a word to become a different word. Words with lower LD (fewer letter changes needed to form a different word) are considered to have high orthographic neighborhood density with and to have low orthographic neighborhood density with a higher distance.	Yarkoni et al., 2008
PNLD (“PLD20”)	The mean distance of a word to its 20 closest phonological neighbors (based on LD). See “ONLD” for more details.	Balota et al., 2007
REG (Regularity)	Binary variable pulled from ABRS designation. Regular words have reliable phoneme-to-grapheme correspondence (i.e., words are spelled as they sound) and relies more on phonological route for spelling. Irregular words do not have reliable phoneme-to-grapheme spelling, and therefore, relies more on lexical/semantic route.	Henry et al., 2011; Norton et al., 2007
SND (“Sem_N_D”)	Average distance between words in a semantic neighborhood and the target word.	Shaoul & Westbury (2006)
VAL (“Valence_NRC”)	Negative/Positive emotional association with a word from 0 (negative) to 1 (positive).	Mohammad & Turney (2010)

(Abbreviations: AOA=Age of Acquisition; AROUS=Arousal; AUDLEX=Auditory Lexical Decision Response Time; CONCRETE=Concreteness; FAM=Familiarity/Prevalence; FREQ=Frequency; IMAG=Imageability; NLETT=Word Length by Letter; NPHON=Word Length by Phoneme; ONLD=Orthographic Neighborhood Levenshtein’s Distance; PNLD=Phonological Neighborhood Levenshtein’s Distance; REG=Regular; SND=Semantic Neighborhood Density; VAL=Valence;)

Statistical Analysis

Sixteen participants (level-2 between-group “clusters”) and a list of 40 words (level-1 within-group observations) were included in the analysis, resulting in a total of 640 observations. As data collection is ongoing, the current sample size is based on the number of participants with lvPPA who completed spelling assessments at the time of analysis. See Table 2 for participant demographics. A random-intercept multilevel logistic regression fit by maximum likelihood estimation (Laplace approximation) was used to evaluate the effect of covariates (participant-level characteristics) and psycholinguistic variables (word-level predictors) on spelling accuracy (binary outcome). All analyses were conducted in statistical software R, version 4.4.0.

Table 2.

Participant Demographics and Overall Spelling Score (n=16)

Participant ID	Sex	Age	Educ. (yrs)	TPO (yrs)	Score	% Correct
1	M	68	16	3	28	70
2	M	80	22	2	36	90
3	F	58	16	4	40	100
4	F	78	18	1	32	80
5	M	75	18	4	40	100
6	M	77	18	3	39	97.5
7	F	74	22	2	31	77.5
8	M	81	16	3	25	62.5
9	F	61	14	1	26	65
10	M	66	14	4	27	67.5
11	M	63	16	5	10	25
12	F	56	16	2	18	45
13	F	77	18	3	35	87.5
14	M	69	22	3	18	45
15	F	54	18	1	35	87.5
16	F	82	16	4	34	85
Mean		69.9	17.5	2.81	29.6	74.1
SD		9.07	2.5	1.19	8.67	21.7
Range		54–82	14–22	1–5	10–40	25–100
Skewness		−0.34	0.67	−0.09	−0.75	−0.68
Kurtosis		−1.27	−0.51	−0.96	−0.24	−0.57

(TPO=Time Post Onset)

Descriptive statistics and histograms were used to evaluate distribution of variables. Poorly distributed variables were excluded from inferential statistical analysis. See Table 3 for descriptive statistics for all 13 psycholinguistic variables (at word level). Correlations between variables were also assessed (see Supplementary Table A). Assumptions of linearity between the outcome (logit) and predictor variables were evaluated using scatter plots. Skewness and kurtosis of residuals were checked for normality (i.e., near zero). Variance inflation factor (VIF) values were evaluated, using a threshold of <5 (Kim, 2019), to check for multicollinearity between variables. Outliers were evaluated for possible influence using the “influence.ME” package in R, as modeled by Nieuwenhuis et al. (2012) and Huang (2023).

Table 3.

Descriptive Statistics at Word Level (n=40)

Psycholinguistic Variable	Mean	SD	Min - Max	Range	Skew	Kurtosis
Included:
REG	(binary)
AOA	6.68	2.05	3.57 – 11.78	8.21	0.60	−0.70
ONLD	1.74	0.38	1.00 – 2.70	1.70	0.40	0.37
PNLD	1.47	0.39	1.00 – 2.50	1.50	0.61	−0.19
VAL	0.52	0.23	0.10 – 0.93	0.83	−0.19	−0.87
Removed:
FREQ	3.18	0.75	1.00 – 4.77	3.77	−0.15	0.47
NLETT	5.10	0.93	4.00 – 7.00	3.00	0.56	−0.55
NPHON	4.03	0.97	3.00 – 7.00	4.00	0.93	0.60
FAM	2.42	0.12	2.13 – 2.58	0.44	−0.51	−0.28
SND	0.60	0.11	0.19 – 0.70	1.50	−2.24	4.75
IMAG	4.59	1.24	2.50 – 6.50	4.00	−0.03	−1.35
AROUS	0.46	0.19	0.15 – 0.85	0.70	0.29	−1.12
AUDLEX (sec)	937.47	66.99	825.51 – 1081.29	255.78	0.40	−0.54
AUDLEX_Z (scaled)	−0.43	0.26	−0.87 – 0.12	0.99	0.33	−0.40

(Abbreviations: REG=Regular; AOA=Age of Acquisition; AUDLEX=Auditory Lexical Decision Response Time; AUDLEX_Z=Auditory Lexical Decision Response Time Z-Scale; AROUS=Arousal; FAM=Familiarity; FREQ=Frequency; IMAG=Imageability; NLETT=Number of Letters; NPHON=Number of Phonemes; ONLD=Orthographic Neighborhood Levenschtein’s Distance; PNLD=Phonological Neighborhood Levenschtein’s Distance; VAL=Valence; SND=Semantic Neighborhood Density)

Using the “lme4” package, a baseline unconditional model [logit(probability of accurate spelling) = g₀₀ + u_0j + r_ij] was fitted to determine baseline variance and to evaluate the extent to which unspecified differences between participants accounted for the variability in spelling accuracy. The effect of covariates was evaluated to examine the influence of specific participant-level differences (e.g., sex, age, years of education, and time since onset in years) on the model. To find the most parsimonious model while also ensuring inclusion of at least one variable from each psycholinguistic category (i.e., general, orthographic, phonological, semantic, and response), word-level predictors were fitted in a stepwise nested manner beginning with broad “general” psycholinguistic variables, followed by specific psycholinguistic categories, in order of hypothesized relevance to spelling (i.e., most-to-least) – “orthographic” variables, then “phonological”, then “semantic”, and lastly, “auditory-lexical response time”. Variables with significant effects in each nested model were retained. The final simplified model included REG, AOA, ONLD, PNLD, and VAL as predictors [logit(probability of accurate spelling) = a_j + β^reg*reg + β^aoa*aoa + β^onld*onld + β^pnld*pnld + β^valence*val + e_i]; model fit was compared to a “full” model containing all 13 predictors. Non-convergence was only observed in the “full” model, which was addressed using the “bobyqa” optimizer in R. Model fit at each stage was also evaluated using likelihood ratio testing. A summary of model building steps and goodness-of-fit comparisons are reported in Table 4.

Table 4.

Summary of Stepwise Model Building and Goodness-of-Fit Comparisons (n participants=16; n words=40; n observations=640)

Model	Included Variables	Likelihood Ratio	Df	p-value
Null	Random effects only (participants)	Null vs Covariates: χ2 = 1.79	6	p=0.775
Covariates	Sex; age; education in years; time post onset in years
Model 1. +General	REG; AOA; FREQ	Model 1a vs 1: χ2 = 2.87	4	p=0.090
Model 1a. +General (significant variables)	REG; AOA
Model 2. +Orthographic	REG; AOA; ONLD; NLETT	Model 2a vs 2: χ2 = 0.00	5	p=0.956
Model 2a. +Orthographic (significant variables)	REG; AOA; ONLD
Model 3. +Phonological	REG; AOA; ONLD; PNLD; NPHON	Model 3a vs 3: χ2 = 3.81	6	p=0.051
Model 3a. +Phonological (significant variables)	REG; AOA; ONLD; PNLD
Model 4. +Semantic	REG, AOA, ONLD, PNLD, IMAG, CONCRETE, AROUS, VAL	Model 4a vs 4: χ2 = 0.25	7	p=0.970
Model 4a. +Semantic (significant variables)	REG, AOA, ONLD, PNLD, VAL
Model 5. +Response	REG, AOA, ONLD, PNLD, VAL, AUDLEX_Z	Model 5a vs FINAL: χ2 = 0.84	7	p=0.358
FINAL Simplified Model	REG, AOA, ONLD, PNLD, VAL
Full Model with all psycholinguistic variables of interest (non-convergence addressed with “bobyqa” optimizer in R)	REG, FREQ, AOA, NLETT, NPHON, ONLD, PNLD, IMAG, CONCRETE, AROUS, VAL, AUDLEX_Z	Full vs Final Model: χ2 = 6.68	7	p=0.462

(Abbreviations: AOA=Age of Acquisition; AROUS=Arousal; AUDLEX_Z=Auditory Lexical Decision Response Time_Scaled; CONCRETE=Concreteness; FREQ=Frequency; IMAG=Imageability; NLETT=Word Length by Letter; NPHON=Word Length by Phoneme; ONLD=Orthographic Neighborhood Levenshtein’s Distance; PNLD=Phonological Neighborhood Levenshtein’s Distance; REG=Regular; VAL=Valence)

Odds ratios with 95% confidence intervals (CIs) and predicted probability of spelling accuracy (i.e., probability at 1 standard deviation [SD] above or below mean values; Austin & Merlo, 2017) were obtained for each variable in the final model and reported in Table 5. The difference in low/high probability is interpreted as the size of the effect of a particular variable on spelling accuracy.

Table 5.

Multilevel Logistic Regression Results Predicting Spelling Accuracy Using Odds Ratios and Predicted Probabilities (n participants=16; n words=40; n observations=640)

Variable	Null Model	+ General (95% CI)	+ Orthographic (95% CI)	+ Phonological (95% CI)	+ Semantic (95% CI)	VIF	Predicted Probability
Int	4.23*** (2.02, 8.83)	14.55*** (4.45, 47.60)	52.02*** (11.01, 245.83)	37.33*** (7.57, 184.16)	16.88*** (3.24, 87.95)
REG		6.09*** (3.75, 9.90)	5.57*** (3.41, 9.09)	4.47*** (2.64, 7.55)	5.43*** (3.12, 9.47)	1.25	REG: 93% IRREG: 67%
AOA		0.76*** (0.68, 0.84)	0.80*** (0.71, 0.89)	0.83** (0.74, 0.93)	0.88* (0.78, 0.997)	1.29	Late: 72% Early: 90%
ONLD			0.418** (0.22, 0.80)	0.21*** (0.09, 0.51)	0.18*** (0.07, 0.45)	2.17	High LD: 72% Low LD: 89%
PNLD				2.61* (1.17, 5.83)	2.09† (0.91, 4.80)	2.07	High LD: 83% Low LD: 78%
VAL					6.07** (2.03, 18.18)	1.25	Pos: 85% Neg: 77%
R2 (marg)	0.000	0.172	0.180	0.187	0.196
R2 (cond)	0.373	0.565	0.578	0.589	0.601
AIC	632.2	538.5	533.8	528.7	526.4
BIC	641.1	560.8	560.6	559.9	571.0
ICC	0.4	0.5	0.5	0.5	0.5
RMSE	0.39	0.34	0.34	0.34	0.34

^†p<0.1

^*p<0.05

^**p<0.01

^***p<0.001

(REG=Regular; AOA=Age of Acquisition; LD=Levenshtein Distance; ONLD=Orthographic Neighborhood Levenshtein’s Distance; PNLD=Phonological Neighborhood Levenshtein’s Distance; VAL=Valence; VIF=Variance Inflation Factor; AIC=Akaike Information Criterion; BIC=Bayesian Information Criterion; ICC=Intraclass Correlation Coefficient; RMSE=Root Mean Square Deviation)

Note: High LD = Low neighborhood density (e.g., yacht); Low LD = High neighborhood density (e.g., cat)

Missing Data

Missing values for psycholinguistic variables were imputed based on information found in the literature. There were three missing data items, including AOA for the word “twenty”, as well as IMAG values for the words “twenty” and “plastic”. Reasons for missingness are unknown. It is possible that the word “twenty”, which was missing from both variables, may be more likely to be excluded from study word lists due to its more common use as a numerical value (e.g., 20) rather than an orthographically-presented word. Cheung and colleagues (2017) found that children can count to “20” by approximately 5 years old and New Mexico common core standards indicate that children typically know how to count to “20” by kindergarten. For these reasons, AOA of the word “twenty” was imputed as “5”. With regard to IMAG, this is a variable that is often coupled with “concreteness”, as they are believed to both engage dual-coding neurological processes for word processing and have been reported as highly correlated (r ≈ .83) (Connell & Lynott, 2012; Paivio et al., 1968). Therefore, concreteness values (Brysbaert et al., 2014), which were available for “twenty” and “plastic” through the SCOPE database, were used to impute the missing IMAG values.

Results

Participants

Participants included in this preliminary report were assessed for eligibility between August 2016 - December 2023 (n=64), of which 37 did not meet inclusion criteria and 11 did not complete the spelling task. Recruitment is ongoing. See Figure 1 for flow chart.

Figure 1.

Participant Enrollment Flow Chart

Descriptive Statistics

Sixteen participants (level-2), with 40 words (level-1) each, were included in the analysis. Participants were all cisgender and half female/half male (n=8 each), between the ages of 54–82 years (mean=69.9; SD=9.07), with 14–22 years of education (mean=17.5; SD=2.5), 1–5 years post onset of symptoms (mean=2.8, SD=1.19), and spelling scores between 10–40 (mean= 29.6; SD=8.67) (See Table 2). No participants had missing data. Word-level variables FAM and SND were not included in the analysis due to poor distribution. Although skewness and kurtosis were within acceptable limits for FAM, it was found to have a very small range of 0.45 out of a possible range of 5.152 (Brysbaert et al., 2019).

Main Results

The null model had an intraclass correlation coefficient (ICC) of 0.37, indicating that 37% of the variability in spelling accuracy is attributable to differences across participants (i.e., random effects), and the remaining 63% is due to differences at word-level (i.e., fixed effects) (Huang, Chapter 2, 2023). According to the null model, individuals in the 20^th percentile for spelling accuracy have a ~50% predicted probability of spelling a word correctly, while individuals in the 80^th percentile have a ~95% probability, indicating a median/mean probability of accuracy of 72.5%. Covariates (sex, age, years of education, and time since onset of symptoms) did not significantly account for differences in spelling accuracy and likelihood ratio test found no significant difference between the null model and model with covariates [𝛘2 (6, n=16) = 1.79, p>0.05]. Results of likelihood ratio tests comparing each nested model are shown in Table 4.

In the final model, predictor variables REG [β=1.692; SE=0.283; p<0.001], AOA [β=−0.129; SE=0.064; p<0.05], ONLD [β=−1.725; SE=0.469; p<0.05], and VAL [β=1.804; SE=0.559; p<0.01] were significantly associated with spelling accuracy. Odds ratios with 95% confidence intervals for each variable in the final model are listed in Table 5 (regression coefficients and odds ratios for all 13 variables of interest are available in Supplementary Tables B and C). Given the predicted probability of spelling accuracy for each significant variable, REG was found to have the largest effect on spelling accuracy, with a 93% probability of accuracy on REG words and 67% probability on IRREG words. AOA and ONLD had similar differences, with a 90% chance of accuracy on earlier acquired words vs. a 72% chance on later acquired words, and an 89% chance of accuracy on words with low orthographic LD (i.e., higher orthographic neighborhood density; e.g., “mile”) vs. 72% for words with high orthographic LD (i.e., lower orthographic neighborhood density; e.g., “circuit”). The effect of VAL on spelling accuracy was smaller, with an 85% probability on words with high VAL and 77% on low VAL words. Model assumptions of linearity, normality, and non-multicollinearity were met. No influential data points were found.

Other Analysis

A post-hoc power analysis using the PowerUp! package (Dong & Maynard, 2013) in statistical software R, version 4.4.0. was conducted to estimate the sample size needed to detect an effect size of .25, with 80% power. Using ICC recommendations for multilevel cross-sectional observational studies in psychological research, as reported in Williams et al. (2024) (i.e., ICC=0.06), the power analysis found that a total of 44 participants (clusters) would be needed to reliably detect an effect size of .25, with 80% power.

Discussion

Key Results

The main purpose of this study was to examine the influence of specific psycholinguistic variables of words on spelling accuracy of real words for people with lvPPA. The analysis found that word-level differences accounted for most of the variability in spelling accuracy, with about a third of the variance explained by differences between participants (not related to participants’ sex, age, years of education, and years since onset). Among the 13 different variables identified in the literature as potentially relevant to spelling accuracy, REG, AOA, ONLD, and VAL were found to be significant predictors of spelling accuracy for people with lvPPA. (Word-level variables FAM and SND were not included in the analysis due to poor distribution.) Although the effect of PNLD on spelling accuracy was not significant after controlling for the effect of VAL, it was retained in the final model, given the theoretical rationale that phonology is an important aspect of spelling (whether or not its effect on spelling accuracy is significant in this specific population). Although ONLD and PNLD are highly correlated (r=.62) (see Supplementary Table D), ONLD was found to be a notably stronger predictor of spelling accuracy than PNLD. REG had the largest effect on spelling accuracy, followed by AOA and ONLD, and lastly, VAL.

Interpretation

These preliminary findings offer promising evidence for the use of spelling assessments to provide a more nuanced characterization of written language deficits experienced by people with lvPPA. In particular, REG was confirmed as a strong predictor of spelling accuracy in lvPPA. As phonological processing deficits are a prominent feature of lvPPA, one might assume spelling regular words, which is believed to engage sublexical phonological processes, would be more challenging for people with lvPPA. However, our results showed that regular words were associated with a higher likelihood of spelling accuracy, while controlling for the effects of AOA, ONLD, PNLD, and VAL. Although people can rely on sublexical processes to spell regular words using sound-to-letter conversion, it is known that factors associated with semantic association/memory - typically better preserved in people with lvPPA - are also involved. For example, Henry et al. (2012) reported that semantic processing appears to be essential for spelling low-frequency irregular words but is also recruited for spelling regular words. Therefore, even in the presence of damage to sublexical-phonological routes, people with lvPPA may still access whole-word lexical representations through intact lexical-semantic routes. As such, the positive effect of REG (and conversely, the negative effect of IRREG) on spelling accuracy may not necessarily reflect an over-reliance on phoneme-to-grapheme conversion (i.e., “surface dysgraphia”) in this population. As the focus of our study was to evaluate psycholinguistic variables, which are only available for real words, we did not include a non-word analysis at this time. Therefore, it is unknown if a similar negative effect would have been present on non-words, as some studies suggest would be expected in lvPPA (Neophytou et al., 2019; Shim et al., 2012). However, as individuals with svPPA are expected to perform strongly on non-words, our findings illustrate the importance of evaluating non-words spelling to differentiate lvPPA from svPPA.

Although there is a clear effect of REG on spelling accuracy, there were a few notable exceptions that illustrate the potential influence of other psycholinguistic factors (e.g., AOA, ONLD, and VAL) on spelling accuracy. For example, in our sample, regular words like “grumble” and “plastic” were spelled incorrectly by ≥25% of participants, whereas irregular words “give” and “learn” were spelled accurately by nearly all participants. The irregular words “give” and “learn” are both early-acquired (<6 years old) words, which according to our findings, increases the chance of spelling accuracy, whereas the regular word “grumble” is a later-acquired (>8 years old) word. Although “plastic” is a regular, early-acquired word, and thus, would presumably carry a greater chance of being spelled accurately, it has a notably lower orthographic neighborhood density (i.e., higher orthographic LD), which decreases the likelihood of spelling accuracy. Additionally, “grumble” and “plastic” have neutral or negative valence values (<0.5), whereas “give” and “learn” have higher, more positive valence values (>0.5), which our findings show increases the probability of accurate spelling. However, all nine of the most commonly misspelled words (i.e., spelled incorrectly by ≥50% of the sample, or ≥8 participants) were irregular words, of which five were also later-acquired words (≥9 years old). With that said, it is possible that irregular words acquired later in life (which also tend to be used less frequently), may be inherently more challenging to spell, regardless of PPA-variant. Further research with a larger sample comparing variants is needed to explore these errors and patterns in the context of differential diagnosis of lvPPA.

The effect size of ONLD was found to be similar to that of AOA, in that the likelihood of spelling accuracy was 17% greater for words with lower LD (i.e., more dense orthographic neighborhoods) and 18% greater for early-acquired words. Although AOA has been reported to be the strongest predictor of naming accuracy for people with lvPPA (Jebahi et al., 2023), variable results have been reported for ONLD. Unlike our findings, Neophytou et al. (2019) reported that ONLD was not a significant predictor of spelling accuracy for lvPPA. It is possible that word length, which was included in Neophytou et al. (2019) but not in our final model due to non-significant results, may share some of the variance in spelling accuracy with ONLD. Meanwhile, and more consistent with our findings, Vonk et al. (2019) found that words with higher orthographic neighborhood density (i.e., low orthographic LD) were associated with greater accuracy and quicker reaction times on lexical-semantic decision making (i.e., “is this a word?” - yes/no) in people with lvPPA. Words with lower orthographic LD (e.g., mile) require fewer letter substitutions/deletions needed to form a new word, and thus, have more orthographic neighbors (i.e., words that differ by 1 letter - e.g., bile, tile, pile, mine, milk, etc.) and greater orthographic neighborhood density compared to words higher orthographic LD (e.g., circuit), which have fewer orthographic neighbors and less dense neighborhoods. Words with higher neighborhood density are surrounded by orthographically similar counterparts, providing additional examples of similar sound-letter correspondences. As a result, these densely connected words may lead to more robust neural connections that provide some protection against degeneration. Furthermore, research has shown that words with denser orthographic neighborhoods are associated with greater semantic activation. For instance, EEG studies of the N400 event-related potential (ERP) in controls have revealed that words with many orthographic neighbors elicit a similar increase in N400 responses as those elicited by words with greater semantic associations (Müller, Duñabeitia, & Carreiras, 2010). This suggests that the presence of more orthographic neighbors activates their corresponding semantic representations and increases global lexical-semantic activity (see also Grainger & Jacobs, 1996). The robust network of words that share similar sound-letter correspondence and increased lexical-semantic activity associated with denser orthographic neighborhoods helps to explain the role of ONLD in spelling accuracy in lvPPA.

The effect of PNLD on spelling outcome was not significant after controlling for the effect of semantic variables, which indicates that the variance in spelling accuracy is better explained by semantic differences across words than differences in phonological neighborhood density. As phonological impairment is more common in lvPPA, phonological variables may have less of an effect on spelling accuracy (i.e., the chance of accurately spelling a word with low vs. high phonological LD may not differ in a measurably meaningful way) in our small and relatively homogeneous sample. Although several semantic variables (e.g., IMAG, CONCRETE, AROUS, VAL) were evaluated, only VAL was significant in the model. This is the first study, known to the authors, that evaluates the effect of VAL on spelling accuracy for people with lvPPA (see Macoir et al., 2020 for an investigation of VAL in svPPA). Jebahi et al. (2023) evaluated the relationship between VAL and naming accuracy, which was not found to be significant in people with lvPPA. Studies involving manipulations of lexicality and VAL in ERP research in healthy controls have demonstrated that neural processing related to VAL is separate from, but interacts with, lexical-semantic processing (Bayer, Grass, & Schacht, 2019). In addition, studies have reported that words with positive or negative VAL, as compared to neutral words, are prioritized when processing language (Bayer et al., 2019; Citron, Weekes, & Ferstl, 2013), and learned associations between the orthography itself (i.e., the word form) and emotional valence influence lexical access and language performance (Bayer et al., 2019; de Zubicaray et al., 2023; Slavova, 2021). It is therefore not surprising that VAL was a predictor of spelling accuracy in lvPPA, especially as social cognition systems are thought to be relatively spared in lvPPA compared to other PPA variants (Fittipaldi et al., 2019).

Limitations

Small sample size (n=16 participants; 40 words) is a primary limitation of this study. Findings are preliminary and should be interpreted with caution until future analyses can be conducted with a larger sample (estimated n=44 participants, per post-hoc power analysis). However, simulation studies have found that multilevel logistic regression estimates (used to calculate odds ratios/predicted probability) remain unbiased even with very small samples (e.g., 10 clusters), although standard errors may be underestimated (Pacagnella, 2011; McNeish & Stapleton, 2014). Therefore, the reported relationship between variables is likely accurate but with potentially less precision/confidence than model results may suggest.

Additionally, poor distribution and/or limited range of psycholinguistic variables represented in the ABRS word list is a limitation. While only two variables (i.e., FAM, SND) were excluded from the analysis due to word list limitations, it prevented us from investigating the contribution of two semantic variables, which may be particularly relevant in the context of relatively preserved lexical-semantic processing.

Generalizability

Although the study’s small sample size likely underestimates confidence intervals, the reported regression estimates, odds ratios, and predicted probabilities are likely generalizable to individuals with lvPPA (McNeish & Stapleton, 2014). Demographics and time since onset did not significantly account for variance in spelling scores, and therefore, do not appear to influence the generalizability of results. However, approximately 37% of the variance in spelling score was explained by participant-level differences, which may be associated with disease-related covariates that were not specified in the model (i.e., severity of language impairment; working memory and/or other co-occurring cognitive impairments), as well as pre-morbid spelling ability. Controlling for these variables in future analyses may provide additional context to improve generalizability of results.

Future Directions

Unique challenges to diagnosing lvPPA have been described throughout the literature (e.g., Harris et al., 2013; Gorno-Tempini et al., 2016; Sajjadi et al., 2012; Watanabe et al., 2024). Spelling is a relatively quick and easy assessment to administer, and with a thoughtfully curated word list (i.e., words well-distributed across psycholinguistic variables that are known to significantly impact spelling accuracy across variants), it may be a useful tool for differentially diagnosing lvPPA. Future studies may benefit from using the SCOPE database to generate a list of words featuring a wide range of psycholinguistic variables of interest (including semantic variables with adequate distribution for analysis). Such information is especially needed to differentiate lvPPA from nfvPPA, as nearly all of the studies evaluating spelling in PPA (mostly focusing on the effect of regularity, lexicality, and types of errors - not psycholinguistic variables) reported non-significant differences between lvPPA and nfvPPA. However, our findings also revealed potential challenges to differentiating lvPPA and svPPA; specifically, if the effect of regularity on spelling accuracy is over-interpreted as evidence of “surface dysgraphia”, which is one of the secondary diagnostic criteria for svPPA (Gorno-Tempini et al., 2011). As spelling assessments typically only include real words (i.e., regular and irregular words), this finding supports the inclusion of both non-words and real words in spelling assessments to better differentiate lvPPA and svPPA. Additionally, evaluating spelling errors at letter-level (e.g., phonological plausibility of errors; number of letters in the correct position) may provide valuable insight into the distinct compensatory neural processes underlying similar phenotypical/behavioral characteristics seen across variants (Sepelyak et al., 2011). For example, if difficulty spelling irregular words is characteristic of both svPPA and lvPPA, the presence of phonologically plausible errors may indicate recruitment of sublexical processes to compensate for lexical-semantic breakdowns (i.e., “surface dysgraphia”), whereas phonologically implausible errors may reflect impairment in both sublexical and lexical pathways. Error-analyses may also provide insights into the role of working memory deficits, which is known to be prominent among those with lvPPA (e.g., the effect of impaired graphemic buffer on spelling accuracy; Sepelyak et al., 2011).

Findings from existing and future studies, should serve as a strong foundation to assertively advocate for the inclusion of a brief spelling assessment in the National Alzheimer’s Coordinating Center Uniform Data Set (currently UDS-4), which is the standardized longitudinal assessment package utilized by all 35 ADRCs across the U.S. to evaluate and stage cognitive decline. Since reading and writing deficits become more prominent over time in lvPPA (Lavoie et al., 2023; Ulugut et al., 2022), information from reading and writing assessments, including spelling, can potentially be incorporated to track disease progression and support overall management. Incorporating an efficient measure like spelling could improve earlier detection of language-led dementias (i.e., PPA), leading to more appropriate care, and improved outcomes.

Improving assessment methods may also involve examination of self-corrections. In our sample, several participants successfully self-corrected their spelling errors, and these corrections were counted as accurate. As it relates to our strongest predictor in this study, self-corrections were made on both regular (n=8 words) and irregular (n=11 words). Poor spellers exhibited self-corrections, but top spellers did too; for example, a participant who scored 36 out of 40 made 6 self-corrections. The implications of self-corrections, and the utility of related data, remain unclear. It is not known whether they indicate atypical cognitive and/or linguistic processes, whether adjusting scores to account for these corrections impacts the sensitivity of spelling assessments, and/or if the nature of the errors (e.g., phonologically plausible, semantic substitution, pseudoword, etc.) provides valuable insights. Additionally, it remains to be seen whether self-corrections signal an unstable neural system on the verge of decline and if tracking them over time can aid in monitoring and management. This area merits further investigation, particularly for variables that have been previously reported as having a milder effect on people with lvPPA or similar effects for lvPPA and nfvPPA.

Research on spelling abilities in lvPPA may have important implications for treatment research and clinical practice. Treatments for PPA should focus on individualized, personally relevant content, leveraging relatively spared skills to compensate for those affected. Additionally, treatment should focus on reinforcing skills identified through research and clinical assessment as being potentially vulnerable to future decline (i.e., a prophylactic approach aimed at maintaining communication abilities for as long as possible) (e.g., Croot, 2018; Flurie, Ungrady, & Reilly, 2020; Henry, et al., 2018, Meyer, et al., 2024). Existing spelling treatments (e.g., Anagram and Copy Treatment [ACT], Copy and Recall Treatment [CART], group interventions, etc.) (e.g., Beeson, 1999; Beeson, Higginson, & Rising, 2013; de Aguiar et al., 2020; Jolliff & Trope, 2023) could be further improved by incorporating information related to additional psycholinguistic variables. Procedures from existing naming treatments that already include tandem targeting of lexical-semantic and phonological abilities, as well as personalization (e.g., Dial et al., 2019; Rogalski et al., 2016; Robinaugh & Henry, 2022), could be applied to development of innovative spelling treatments. Incorporating psycholinguistic variables and PPA variant profiles can also potentially enhance the effectiveness of script training. While personalized scripts are already recommended, carefully selecting vocabulary words that may be particularly vulnerable to progression, such as irregular, later-acquired, negatively-perceived words, with less dense orthographic neighborhoods, can further optimize outcomes. Tools like the SCOPE database provide researchers and clinicians with easy access to psycholinguistic metrics for target words, facilitating more tailored and effective treatment strategies.

Conclusion

This study examined the influence of specific psycholinguistic variables on spelling accuracy in people with lvPPA, finding that word-level differences largely accounted for variability in spelling outcomes. Key predictors included REG, AOA, ONLD, and VAL, with REG showing the strongest impact. The findings highlight the potential for spelling assessments to provide nuanced insights into language deficits, which could aid in earlier detection of lvPPA and improved differentiation of lvPPA from other variants. Despite the limitations of a small sample size, our findings support the utility of future research to explore the nuanced relationship between psycholinguistic variables and spelling accuracy to enhance differential diagnosis and treatment strategies for lvPPA, as well as advocacy for the inclusion of spelling assessments in standardized evaluation frameworks. Given that lvPPA is typically caused by the most common type of dementia (i.e., AD), progresses more rapidly than other variants, and is associated with greater cognitive impairment (Ulugut et al., 2022), we encourage our academic community to explore all possible opportunities, such as spelling assessment; improving detection and management is essential for ensuring the prompt translation of findings and refinements into clinical practice.