Posterior cortical atrophy (PCA) is a clinico-radiological syndrome characterized by a progressive decline in higher-order visuospatial processing with relative preservation in other cognitive domains at initial presentation (Benson et al., 1988; Renner et al., 2004; Crutch et al., 2017). Common visuospatial symptoms of the syndrome include impaired object and space perception, simultanagnosia, environmental agnosia, and visual field defects (Crutch et al., 2017). From the neuroimaging perspective, the syndrome is associated with atrophy, hypometabolism, and usually tau deposition in posterior parietal, occipital, and temporo-occipital cortices (Whitwell et al., 2007; Lehmann et al., 2011). As the majority of PCA cases are due to underlying Alzheimer’s pathology, PCA is also referred to as the visual variant of Alzheimer’s disease (AD) (Levine et al., 1993; Alladi et al., 2007). While the diagnostic criteria for PCA indicate preserved functions in cognitive domains outside of visuospatial processing at symptom onset, a growing literature has documented language abnormalities in PCA emerging early in the course of the illness. Specifically, impaired category fluency and confrontation naming have been documented on formal neuropsychological assessments (Tang-Wai et al., 2004; McMonagle et al., 2006; Putcha et al., 2018, 2020). Language abnormalities are also evident during spontaneous speech, such as using higher frequency words and slowed speech rate (number of words per minute) (Crutch et al., 2013). These emerging observations suggest that there is still much to be understood about the nature and origin of language impairments in PCA.

The specific types of language abnormalities observed in PCA may be related to the network dysfunction that supports lexicosemantic retrieval, as has been previously postulated across the phenotypic spectrum of AD (Putcha et al., 2020). Another possible explanation is that the language abnormalities observed in PCA may stem from the visuospatial impairments central to the syndrome, rather than representing a primary language deficit. A large body of research supports close relationships between the visual processing of objects and the amodal semantic processing required for retrieving the names of those objects (Binder and Desai, 2011; Huth et al., 2016; Aliko et al., 2023). Recently, it has been shown that anterior to each region that is selective for the visual processing of a given category in the visual cortex, there is a corresponding area selective to its linguistic processing (Popham et al., 2021). This anatomical and functional configuration suggests that the anterior border of the visual cortex acts as a convergence zone where information from the unimodal visual system enters the amodal linguistic systems involved in linguistic retrieval. Therefore, the successful production of a word that has visual attributes requires intact visual processing, essential for providing the information needed to retrieve its corresponding linguistic representation (i.e., its name). Therefore, deficits in visual processing would theoretically impede the production of words with visual attributes. If the pathophysiology of language abnormalities in PCA involves disrupted visual processing, then it stands to reason that tasks heavily dependent on visual processing will exhibit significant language impairments. Conversely, tasks with minimal reliance on visual input should result in relatively intact language performance.

In the current study, we sought to test this hypothesis by contrasting the language used in two different speech samples as PCA participants described the Picnic scene from the Western Aphasia Battery (Kertesz et al., 2007) (visually dependent) and their prior jobs (visually independent). In each speech sample, we measured word frequency, word utterance latency, and the use of spatial relational words. For the picture description task, we hypothesized that PCA patients would use higher frequency words (e.g., replacing specific names of pictured items with superordinate words potentially including “thing”), have increased word utterance latency due to object recognition difficulty, and use fewer spatial relational words such as “into” or “underneath” compared to healthy individuals. For the non-visually dependent job description task, we expected these linguistic features to be comparable between PCA patients and healthy individuals.

Building on the hypothesis that speech patterns in visually dependent tasks reflect visuospatial processing deficits, we next sought to identify linguistic markers of these challenges. Specifically, we investigated which elements in the picnic scene presented particular retrieval difficulties and whether PCA patients could intuitively grasp and articulate the overall theme of the scene, such as using the word “picnic.” Due to the difficulty with visually integrating a scene (simultanagnosia) (Tang-Wai et al., 2004; Singh et al., 2015; Cui et al., 2022), we hypothesized that PCA patients were less likely to verbalize the term “picnic” compared to healthy individuals. Lastly, to address the clinical significance of this work, we used the language features derived from the picture description task to develop a classifier aimed at distinguishing PCA patients from healthy individuals and hypothesized a high degree of classification accuracy.

Using computational linguistic analysis in PCA, our study illuminated a distinction in language performance between visually-dependent and visually-independent contexts. At the theoretical level, we used this method to show that at least some of the language abnormalities increasingly being identified in PCA are byproducts of visuospatial deficits characteristic of this atypical AD syndrome. On the visually-independent job description task, the characteristics of language production we measured here were not impaired in patients with PCA. Translating our observations to clinical practice, we showed that computational linguistic analysis of a simple picture description task robustly classified nearly all PCA patients as distinct from healthy controls, supporting its value in clinical diagnostic evaluation.

Our work is consistent with studies showing rich connections between networks representing information directly received from senses and information conveyed through spoken language (Damasio, 1989; Ralph et al., 2017; Popham et al., 2021). Unimodal sensory information and abstract language information are combined at multiple points across the cortex, such as inferior parietal lobule (comprising the angular and supramarginal gyri) and large swaths of posterolateral temporal cortex (Mesulam, 2000; Devereux et al., 2013), many of which can be affected in PCA. Therefore, language abnormalities in PCA may arise for at least two potential reasons. First, the neurodegeneration of PCA may extend beyond visuospatial areas to encompass regions involved in abstract language processing. In support of this hypothesis, evidence suggests that the brain regions affected in PCA overlap with those critical for word retrieval (Vonk et al., 1991; Migliaccio et al., 2009; Warren et al., 2012; Ossenkoppele et al., 2015; Leyton et al., 2017). Alternatively, language anomalies may arise as a consequence of visuospatial deficits hindering the transfer of necessary sensory information for amodal language processing.

While these two possibilities are not mutually exclusive, our results suggest that language impairments might be largely secondary to visuospatial dysfunction. In our analysis comparing a variety of speech and language properties of the narratives produced when PCA patients describe a complex visual scene versus a recounting of their primary occupation from memory, we observed speech and language impairments in only the visually dependent picture description task. We believe the increased word frequency and word utterance latency in the picture description task reflect visual deficits in object recognition. Similarly, the reduced use of spatial relational words may reflect the patients’ difficulty processing spatial relations between elements of the picnic scene. The absence of abnormalities in word frequency, word utterance latency, and relational words in the job description task provides evidence that these language impairments do not stem from an intrinsic deficit in the language system in PCA.

Since most daily communication is a blend of visuospatial cognition, episodic memory, and other cognitive domains, we anticipate that an analysis of everyday speech would reveal linguistic deficiencies proportionate to the visuospatial load of its content. This expectation aligns with prior research reporting linguistic anomalies in participants recounting their recent holiday, an account that naturally encompasses the visuospatial processing of a recent event, such as where they went and what they saw (Crutch et al., 2013). Relatedly, it has been shown that the autobiographical narratives of patients with PCA have diminished spatiotemporal and perceptual details compared to healthy individuals. This finding likely stems from the impaired access of PCA patients to visual information, which plays a crucial role in constructing autobiographical memories (Ahmed et al., 2018). Furthermore, the narratives of patients with PCA have been shown to be spatially fragmented when they were asked to imagine and describe commonplace scenes (Ramanan et al., 2018). These findings provide converging evidence for the hypothesis that language reveals the visuospatial impairments of PCA proportionate to the visuospatial load of the content.

Another consideration in interpreting these task differences is that the picture description task required the use of specific linguistic elements representing the specific visual stimulus. It is also possible that when given fewer constraints in the job description task, individuals had the freedom to choose a potentially more familiar and more easily accessible language. Our explanation of the underlying language abnormalities in PCA is consistent with findings that showed a striking discrepancy between visual and verbal comprehension tasks in this population (Rogers et al., 2006). Our conclusion is also synergistic with results reporting very mild impairment in semantic memory in PCA, indicating that the apparent semantic impairment in these conditions may be secondary to visual impairment (Rogers et al., 2006).

Based on the observation that language is a sensitive indicator of visuospatial impairments of PCA, we performed an in-depth content unit analysis of language elicited through the picture description task. First, we measured the probability of verbalizing each content unit of the picture. The most distinguishing feature between PCA patients and healthy individuals was the probability of mentioning the “fisherman.” While 93% of healthy participants mentioned this content unit, only 8% of PCA patients did so. This discrepancy could be attributed to the smaller size of this element in the picture. In addition, multiple elements are superimposed in the location of this content unit. Numerically, though not significantly, certain items, such as “clouds,” had a higher likelihood of being mentioned by PCA patients than healthy individuals. This type of analysis provides a naturalistic way of identifying the visuospatial elements that are particularly challenging for PCA patients and could help clinicians devise rehabilitative strategies to alleviate these challenges. Moreover, we observed that PCA patients often missed describing the overarching theme of the image (“picnic”), even when they identified certain components related to it (“basket”). We believe this finding represents the effects of simultagnosia, which prevents many PCA patients from grasping the integrated theme of a composite visual entity.

Finally, when specific quantitative language metrics were employed to differentiate PCA patients from healthy individuals, our predictive model achieved a high level of performance, as evidenced by an AUC of 1 and an accuracy rate of 98.15%. Automating this linguistic evaluation from an easily acquired speech sample would facilitate the integration of measures like this into digital healthcare infrastructure, which a wide array of healthcare providers could potentially use once trained. This work extends our prior research, which used narrative data from a simple picture description task for a high accuracy classification of the three primary progressive aphasia variants (Rezaii et al., 2022, 2023a,b). Future studies are needed to directly analyze the sensitivity and specificity of language predictors across a wide range of patient populations, examine the neurobiological underpinning of linguistic indicators of PCA, and address some of the limitations of this study. These limitations include the absence of neuropsychological scores for participants recruited via the MTurk online platform and the lack of normative data for tests used to determine the severity of PCA symptoms. Ultimately, these advancements will facilitate early diagnosis of PCA as well as monitoring response to disease-modifying, rehabilitative, or other therapies in this underserved atypical variant of AD.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving humans were approved by Mass General Brigham Human Research Committee Institutional Review Board in Boston, Massachusetts. The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.

NR: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing. DH: Data curation, Writing – review & editing. MQ: Data curation, Writing – review & editing. BW: Data curation, Writing – review & editing. SM: Data curation, Resources, Writing – review & editing. BD: Conceptualization, Data curation, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Writing – original draft, Writing – review & editing. DP: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Writing – original draft, Writing – review & editing.