Toward linguistically-fair screening and assessment of primary progressive aphasia: a transdiagnostic approach for understudied and minority languages

Introduction

This article responds directly to the special topic “Exploring the untapped in dementia research.” For decades, the linguistic and cultural realities of non-English and minority language speakers have remained a largely neglected domain in neurodegenerative research (Arslan and Peñaloza, 2025; García et al., 2023; Taiebine, 2026a; Matić Škorić et al., 2026). We hypothesize that shifting our focus from Anglo-centric norms to globally representative linguistic structures may unlock a deeper, more accurate understanding of how PPA manifests across the human spectrum, thereby addressing a critical, previously ignored linguistic gap in cognitive decline (though empirical evidence across all linguistic variations is still emerging).

As recently highlighted by (Gallée 2024), there is a pressing need for globally responsive screening materials to account for the heterogeneity inherent in dementia syndromes. This paper issues a critical call for the development and implementation of linguistically fair screening and assessment tools tailored to the specific structures of non-dominant languages. We argue that the mere literal “translation” of existing tests is insufficient and often invalid, as direct translation fails to capture the unique psycho-linguistic, socio-linguistic, and cultural properties of the target language. Instead, we propose that assessments must be grounded in cross-linguistic biomarkers and culturally transdiagnostic models that recognize the diversity of linguistic reality. Therefore, a truly holistic approach to PPA must consider how topological, biological, hodotopic, hodological, and typological factors may interact with the neuropsychological and linguistic specificities of targeted dialects and languages (see Table 1).

Table 1. Proposed integration of topological, typological, hodotopic, hodological, and RDoC taxonomy levels in PPA.

Multidimensional approaches to PPA have undergone a significant shift, moving from categorical frameworks toward a dynamic, spectrum-based model that more accurately reflects clinical heterogeneity (Ingram et al., 2020; Ramanan et al., 2022). Historically, the Gorno-Tempini et al. (2011) consensus criteria established a tripartite classification system comprising the non-fluent/agrammatic (nfvPPA), semantic (svPPA), and logopenic (lvPPA) variants. This serves as the international gold standard for clinical research by anchoring diagnoses to specific “core” and “ancillary” linguistic markers. Conversely, Mesulam's framework often advocates for a more granular, four-variant approach. This perspective frequently separates nfvPPA from primary progressive apraxia of speech (PPAOS), acknowledging that motor-speech deficits and agrammatism can be dissociated neuro-anatomically (Mesulam et al., 2014). Furthermore, Mesulam's taxonomy accounts for an “anomic” or “mixed” variant to capture patients who exhibit profound word-finding difficulties but lack the hallmark repetition deficits of lvPPA or the loss of conceptual knowledge seen in svPPA, thereby highlighting the limitations of strictly tripartite systems (Mesulam et al., 2014; Coemans et al., 2021).

PPA multidimensional continuum

Contemporary clinical and neuro-linguistic paradigms are increasingly abandoning rigid taxonomies in favor of a multidimensional continuum and spectrum model that identifies nuanced sub-types and overlapping zones (Ramanan et al., 2022; Santi et al., 2024). Within this new conception, nfvPPA is viewed as a spectrum where patients may present as “purely agrammatic” at one pole or “purely apraxic” at the other, with most individuals manifesting a hybrid profile that evolves as atrophy spreads (Illán-Gala et al., 2024). Similarly, svPPA is recognized as part of a broader frontotemporal spectrum, where linguistic semantic loss eventually converges with the behavioral features of frontotemporal dementia (bvFTD) as a pathology which extends to the right temporal lobe (Hardy et al., 2016). Regarding lvPPA, it is increasingly conceptualized as a specific linguistic phenotype of Alzheimer's pathology rather than a discrete primary aphasia (Giannini et al., 2017). Mapping the underlying cognitive processes of patients along various axes, such as syntax, semantics, phonology, and motor speech, can empower clinicians to identify mixed variants of PPA (mvPPA) or “PPA-NOS” (Not Otherwise Specified) as a dynamic state of progression (Taiebine and Faris, 2023; Taiebine and Aidi, 2025; Taiebine, 2026b). This transition toward multidimensional mapping allows for a more personalized diagnostic profile that accounts for the complex interplay between focal cortical thinning and diverse underlying proteinopathies (Mesulam et al., 2019).

The staging of PPA involves systematically tracking disease progression and symptom severity over time. Recently, approaches like the symptom-led framework have created a pivotal shift toward a more patient-centered understanding of the disease (Hardy et al., 2024a,b). This framework focuses on the step-by-step worsening of symptoms, reflecting the real-life communication challenges patients face in their daily cultural and linguistic environments. However, it is our position that current models have largely failed to address the specific needs of individuals from minority language backgrounds, leaving a critical gap in equitable care.

Recent trends in PPA research highlight a significant shift toward the use of Artificial Intelligence (AI) and machine learning (ML) to enhance diagnostic precision and clinical workflows (Macoir et al., 2025a,b). A primary application is the automated digital analysis of connected speech, which uses AI to detect PPA variants from natural speech patterns (Rezaii et al., 2024). Modern approaches leverage Large Language Models (LLMs) to identify “natural partitions” or clusters within patient speech, achieving high levels of agreement with independent clinical diagnoses (Matias-Guiu et al., 2019). Furthermore, AI-driven clinical workflows with NeuroScreen (Themistocleous and Stark, 2026), a machine learning model designed to automate the complex task of differentiating between various neurological conditions through language analysis, achieved up to 91% accuracy in classifying PPA and its variants. Beyond diagnosis, AI is being applied to neural assessment through EEG and cortical speech envelope tracking, offering objective measures of language network neuroplasticity to support advanced rehabilitation (Tafuri et al., 2024). These innovative trends aim to provide scalable, objective tools that reduce clinician bias and extend specialized diagnostic expertise to non-tertiary healthcare settings.

A holistic approach to PPA requires clinicians to consider how topological brain changes interact with the specific linguistic features of a patient's native language. Culturally responsive and linguistically fair screening should not be merely a supplementary goal but a clinical necessity for accurate diagnosis. Furthermore, by ensuring broader inclusion of diverse populations, researchers and practitioners help to better understand the complexity of PPA in practice. Moving forward, the development of assessment tools validated across multiple languages is essential to provide equitable treatment and improve outcomes for all individuals facing the challenges of language decline.

Discussion

From broad-linguistic to dialect-specific screening of PPA

The world currently hosts approximately 7,168 living languages, though the count of regional dialects remains significantly higher and more difficult to quantify due to dynamic sociopolitical boundaries (Rehm and Way, 2023). Despite this immense variety, English has emerged as the absolute lingua franca of the scientific landscape (O'Neil, 2018). This creates a striking paradox where a single language dominates intellectual discourse, even as thousands of others maintain the world's cultural heritage (Saputra and Lidyawati, 2025).

We suggest that this over-representation of English has led to a critical scarcity of screening tools tailored to the vast majority of the world's languages. To respond to this testing gap, the Mini Linguistic State Examination (MLSE) has emerged as a pivotal instrument for the brief yet multidimensional screening of PPA, specifically designed to bridge the gap between time-intensive neuropsychological batteries and overly simplistic cognitive screens (Garrard, 2024). Recent updates emphasize its robust cross-linguistic validity in over 20 languages while enabling a standardized screening tool for international clinical trials (Garrard, 2024). The paper-pen version of MLSE evaluates five core linguistic domains including motor speech, phonology, semantics, syntax, and verbal working memory, through 11 subtests that take approximately 15 to 20 min to administer. Furthermore, the recent development of a web-based version of MLSE (Garrard, 2024) represents a significant shift toward digital accessibility in the screening of PPA. We anticipate that the automatic scoring of patients' performance might reduce the burden on clinicians and enhance the consistency of diagnostic profiling across key domains. However, its automated digital scoring currently remains heavily restricted to the English language, which limits its utility in a globalized clinical landscape. To overcome this, we propose that future versions of MLSE should use cross-linguistic adaptations, cross-cultural validations, as well as advanced speech recognition technology, leveraging existing machine learning algorithms capable of detecting subtle acoustic markers of motor speech disorders in multilingual clinical settings.

In parallel to the MLSE, the Dépistage Cognitif de Québec (DCQ) has established itself as a promising instrument for screening atypical dementias, including PPA. Validated in French, English, and Spanish, the DCQ provides a comprehensive screening of social cognition and language that complements standard batteries (Bergeron et al., 2015; Laforce et al., 2017; Fernández-Romero et al., 2024; Meilleur-Durand et al., 2025). While tools like the DCQ are a notable exception demonstrating progress, they still only cover a fraction of global language families, highlighting the ongoing scarcity of truly culturally and linguistically responsive research (Quique et al., 2025).

To bridge this gap in current clinical practice where tailored cognitive screens are unavailable, combining specific socio-linguistic investigations using the Bilingual Aphasia Test (BAT) (Kambanaros and Grohmann, 2012) with the MLSE or DCQ could help profile a patient's history and preferences with specific markers of PPA in terms of typological linguistic features. Thus, we suggest that, as an interim measure, the BAT might serve as a highly practical tool for taking a detailed sociolinguistic history, allowing clinicians to understand an individual's unique bilingual and dialect usage before applying standardized and potentially biased tests.

The lack of culturally and linguistically responsive research has direct, detrimental effects on patient outcomes, as general screening tools rarely account for the nuances of non-dominant languages in bilingual and multilingual populations (Grasso et al., 2023; García et al., 2025). Therefore, we maintain that the transition toward dialect-specific screening is fundamentally a matter of health equity and clinical excellence (Taiebine et al., 2026). Practitioners and researchers should work together to create inclusive taxonomies that capture the complexities of PPA in every linguistic context.

Navigating a new PPA transdiagnostic framework

Moving away from a rigid, hierarchical framework toward a multidimensional, network-based approach is fundamentally important, as the categorical boundaries of traditional taxonomies often fail to capture the dynamic, overlapping nature of neurodegenerative decline across different languages. When clinicians are limited to assessing neurolinguistic impairments among non-English speakers using categorical, Anglo-centric diagnostic criteria, we hypothesize that the risk of misclassification and delayed treatment significantly increases. It is important to emphasize that while the interaction between specific linguistic structures and PPA pathology is strongly motivated by theoretical linguistic principles, it is highly possible that there is not yet empirical evidence demonstrating this across all minority language populations.

We propose rethinking PPA as a dynamic framework which emphasizes the importance of a multi-staged understanding of the interconnections among six core pillars: (1) Typological (linguistic structures), (2) Topological (regional brain atrophy), (3) Hodotopic level (the “locus” of a function within a brain network), (4) Hodological (network connectivity), (5) Biological (genetic level), and (6) Psychosocial (environmental/cultural context).

At the typological level, PPA assessment should account for the structural diversity of the world's languages. Most diagnostic tools are designed for English and may fail to detect deficits in morphologically complex or “agglutinative” minority languages such as Arabic or Turkish (Taiebine, 2025a).

From a topological perspective, we should map how focal atrophy correlates with these linguistic structures. In PPA, “topology” refers to the specific brain regions such as the left perisylvian cortex or the temporal poles that are undergoing degeneration. In minority language speakers, the neural representation of certain language-specific features may vary based on the age of acquisition or the frequency of use (Cheng et al., 2025). The current reliance on topological models derived primarily from Western, Educated, Industrialized, Rich, and Democratic (WEIRD) and English-dominant cohorts (Henrich et al., 2010) introduces significant sampling bias. This oversight compromises the diagnostic sensitivity required to detect subtle, early-stage cortical thinning associated with PPA (Katsumi et al., 2023) in linguistically and culturally diverse populations.

Furthermore, the hodotopic level bridges the gap between local regions and the networks they form. It considers not just where the damage is located, but how that damage disrupts the “locus” of language within the wider brain architecture (Catani and Mesulam, 2008). We posit that for a patient using an understudied language, the hodotopic organization of their mental lexicon might involve different co-activated regions compared to a monolingual English speaker. We further argue that failure to account for this feature could lead to clinical oversights where a patient's difficulty in accessing dialect-specific vocabulary is misattributed to general memory loss rather than a specific focal language breakdown.

Expanding this further, the hodologic approach focuses on the “tracts” or white matter pathways that connect these language hubs (Dronkers et al., 2017). Specific conditions, such as classic conduction aphasia (Bernal and Ardila, 2009) and variants of primary progressive aphasia (PPA-particularly the logopenic variant-often involve the breakdown of the dorsal pathway (superior longitudinal fasciculus and the arcuate fasciculus) (Montembeault et al., 2018). A hodologic assessment ensures that we are looking at the “wiring” of the language system, ensuring that the diagnosis accounts for the integrity of the communication network as a whole.

At the biological level, we should integrate genetic and physiological processes. This involves linking clinical phenotypes not just to atrophy, but to the underlying proteinopathies (tau vs. TDP-43) (Olfati et al., 2022), which may manifest differently across diverse genetic backgrounds.

Finally, the psychosocial component acknowledges that environmental and socio-cultural factors influence test performance. This includes the state of diglossia, which is common in many minority language speakers (such as using an Arabic dialect at home and Modern Standard Arabic for official use). PPA can affect these registers disproportionately in diglossic and bilingual contexts; a patient might retain their mother tongue (L1) while losing their acquired language (L2), or vice versa (Costa et al., 2019; Taiebine, 2025a,b).

Translating the framework into clinical assessment

To move this framework from concept to practice, we suggest a multi-tiered assessment strategy that clearly distinguishes between immediate clinical interventions and long-term research imperatives.

We argue that clinical practice can be immediately refined by modifying intake protocols to prioritize sociolinguistic data. While fully validated, cross-linguistically calibrated instruments remain in development, we recommend that practitioners leverage existing resources such as the MLSE or the visual components of the DCQ paired with the BAT to establish a robust baseline of a patient's linguistic history. We believe this approach enables the systematic socio-psycho-linguistic documentation of the age of acquisition, dialectal usage patterns, and formal educational attainment, which serves to attenuate potential biases prior to the administration of standardized neuropsychological assessments.

In terms of long-term research imperatives, we view the transition away from the literal translation of English-centric assessments toward the formal socio-linguistic adaptation and local development of diagnostic tools as a critical necessity. Future research should prioritize the assessment of language-specific structural markers such as tonal variance in tonal languages or agglutinative morphologies in Arabic rather than relying on literal translations of English syntactic structures, which often fail to account for the intrinsic linguistic properties and cognitive demands of the patient's primary language. Establishing these typologically informed frameworks is essential for ensuring diagnostic accuracy and equity across linguistically diverse clinical settings, though this will undoubtedly require rigorous, ongoing cross-linguistic normative research.

As a result, we propose that clinical interpretation should be hodologically informed, recognizing that a “repetition deficit” in a bilingual speaker might theoretically reflect network interference rather than the classic arcuate fasciculus breakdown seen in monolinguals. We acknowledge that such neuro-anatomical correlations are currently speculative and require much more neuro-functional investigation in patients speaking understudied languages. Therefore, we hypothesize that by ensuring an efficient mapping of these six components, clinicians may construct a personalized diagnostic profile that is accurate, equitable, and actionable.

To transcend the constraints of existing taxonomies and better translate the above-mentioned components into clinical assessment, we suggest adapting the transdiagnostic Research Domain Criteria (RDoC) model (Geraldo et al., 2019) to the PPA landscape (as shown in Table 1). Although originally conceptualized to investigate psychopathology and brain-behavior relationships, the multi-level heuristic of the RDoC has recently been extended to neuropsychological contexts (Langenecker et al., 2023). Consequently, this framework may be adapted to encompass the multidimensional nature of PPA by integrating a patient's genotype, phenotype, and neural underpinnings (Elvevåg et al., 2016) with a refined clinical phenotype.

We posit that, unlike existing traditional models, the RDoC applied to PPA could prioritize ancillary neurolinguistic features by incorporating language-specific markers, such as tone-deafness in tonal languages or clitics in Arabic. We further suggest this model could leverage psycholinguistic sensitivity by measuring processing speeds and retrieval patterns calibrated to the specific word frequency and neighborhood density of understudied languages. Therefore, we argue that the RDoC has the potential to facilitate a holistic taxonomy that preserves the core consensus categories of non-fluent, semantic, and logopenic variants while actively accommodating “dialect-sensitive” profiles in canonical PPA sub-types.

Ultimately, we conceptualize this intra-categorical classification of RDoC as a bridge between the biological reality of the disease and the specific linguistic features of the patient. We caution that assessments failing to account for these sociolinguistic realities risk under-diagnosing patients in the early stages of the disease, where the first signs of differential decline appear in the most “natural” or deeply embedded dialectical structures. We expect this new framework will ensure that the assessment of PPA is not only neurologically and neuropsychologically accurate but also linguistically fair, ultimately leading to more equitable healthcare outcomes and person-centered care for minority populations.

Conclusions

To sum up, we maintain that the pursuit of linguistic equity in the diagnosis of PPA requires a fundamental shift from “standardized” assessments to “dialect” precision. While achieving manual dialect precision globally may seem costly, overly fine-grained, or logistically daunting, we stipulate that the rapidly advancing integration of AI and LLMs may offer a realistic, low-cost, and scalable pathway to map dialectal nuances without requiring thousands of independent test translations. Furthermore, applying the RDoC model within the PPA framework acknowledges that the clinical phenotype is not a static set of symptoms, but a dynamic interaction between multi-layered variables.

Moving forward, we assert that the inclusion of ancillary neurolinguistic markers and dialect-specific psycholinguistic data into shared, open-access data repositories will be essential for clinicians and researchers. We anticipate that establishing these open repositories of diverse speech samples and normative cross-linguistic data will ensure that patients from minority and understudied linguistic backgrounds are no longer invisible to current clinical classifications. Ultimately, we view a linguistically-fair screening as more than a clinical improvement; it is a profound commitment to the cognitive rights of diverse populations. Therefore, we argue that the hodologic and topological variations inherent in multilingualism and regional dialects should be treated as an asset to move beyond current Western-centric taxonomies toward a truly global understanding of PPA. We present this call to action as a blueprint for a more inclusive future in aphasiology, where the depth of a patient's linguistic identity is used to enhance, rather than hinder, the accuracy of their diagnosis and their overall quality of life.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Tomasina M. Oh, University of Plymouth, United Kingdom

Reviewed by: Florentina Morello-García, Fleni-CONICET, Argentina