Action and Event-Based Lexical-Semantic Processing in Parkinson’s Disease

Abstract

We tested the hypothesis that Parkinson’s disease (PD) impairs verbs’ event structures and/or sensory-motor semantic features, as suggested by theories of grounded cognition. Nineteen participants with PD and 16 age-matched Controls produced verbs and used a Likert scale to rate verbs’ event-based association with instruments (e.g., fork- eating versus bathing) and locations (e.g., airport- waiting versus singing). When producing instrument-related verbs, PD participants responded slower and had a lower proportion of relevant responses than Controls. Yet PD participants showed a relatively intact ability to produce location-related verbs and to rate instrument- and location-related verbs. Greater motor disease primarily impacted instrument-related verb production. 4/19 PD participants had mild cognitive impairment (per Movement Disorders Society criteria), which affected both instrument- and location-related verb production and ratings. Overall, results were consistent with ‘weak’ grounded cognition theories that propose that sensory-motor simulation may enrich action-semantic processing but is not strictly necessary for action-semantic processing.

2. Introduction

2.1. Background

2.1.1. Verb meaning in healthy adults

Verbs play a critical role in language production and comprehension because they denote events and states, and they facilitate linguistic prediction by constraining likely event participants (Altmann & Kamide, 1999; Ferretti et al., 2001; Kamide et al., 2003; Kukona et al., 2011; McRae et al., 1997; McRae et al., 1998). Widespread neural structures appear to contribute to verb processing in healthy adults and people with acquired language disorders. Many neural structures are active during both noun and verb processing, but some regions, such as the posterior right middle temporal gyrus and left dorsolateral prefrontal cortex, show preferential or selective activation in response to verb processing (Alyahya et al., 2018; Cappa & Perani, 2003; Crepaldi et al., 2013; Friederici & Frisch, 2000). According to grounded cognition theories (also known as embodied cognition) semantic processing is grounded in sensory and motor experiences in which cognition is rooted in the body’s interactions with the world (Barsalou, 2016; Barsalou et al., 2003; Glenberg & Robertson, 2000; Hesslow, 2012). If these theories are correct, then motor and somatosensory cortices should at least partially subserve verb processing.

Behavioral, neuroimaging, and transcranial magnetic stimulation (TMS) studies with healthy adults suggest that sensory-motor neural systems are involved in verb and action language processing, though the degree of reported involvement varies widely across studies. For example, Glenberg and Kaschak (2002) found that healthy adults are faster to respond to sentences when implied motions match the physical motion needed to respond, such as being faster to respond to ‘close a drawer’ when they were required to move their arm away from their body than when they were asked to move their arm towards their body. This is a phenomenon termed the “Action Compatibility Effect (ACE).” However, the ACE appears to be influenced by subtle task differences such as the timing between stimulus onset and participants’ responses (Borreggine & Kaschak, 2006; Garcia & Ibanez, 2016) and the specific items included in the stimulus set (Morey et al., 2022). While action compatibility effects are often small and may fail to reach significance in individual experiments, meta-analyses suggest that the average effect size for the ACE is significant (Winter et al., 2022; but see Morey, Kaschak et al., 2022). Along similar lines, Myung et al. (2006) found that participants in a lexical decision task were faster to press a button to confirm they recognized manipulable objects as real words (versus pseudowords) when they were primed with other objects that were similarly manipulated, such as typewriter priming piano or key priming screwdriver. Such findings suggest that action language comprehension is grounded in action production. Neuroimaging and TMS results suggest that action word processing involve corresponding regions of the primary motor cortex and surrounding motor regions (Aziz-Zadeh et al., 2006; Hauk et al., 2004; Kemmerer et al., 2008; Klepp et al., 2019; Tettamanti et al., 2005). However, some studies do not report similar motor region activation during action language processing (Kable et al., 2005; Kable et al., 2002), and new “weak” forms of grounded cognition assert that motor simulations enrich verb and action language processing but are not necessary or required for comprehension or production (Binder & Desai, 2011; Meteyard et al., 2012).

In contrast, higher-level aspects of verb processing may be subserved by brain regions that process abstract features of actions such as goals and intentions, according to Kemmerer and Gonzalez-Castillo (2010) Two-Level theory of verb meaning. In Two-level theories, verbs are represented at the lower, more sensory-motor grounded “root” level, which captures a verb’s unique idiosyncratic semantic features, and the higher, more abstracted “event structure” level, which captures shared aspects of meaning that may define verbs’ membership in a given class (Levin & Rappaport Hovav, 2005; Rappaport Hovav, 2017; Rappaport Hovav & Levin, 1998). Specifically, event structure templates are thought to represent the basic grammatically relevant elements of verb meaning, such as causal relationships within event participants and aspectual features, that is, their internal temporal make-up (Levin & Rappaport Hovav, 2005). For example, “melt,” “dry,” and “shatter” differ in semantic features but are thought to share an event structure template in which an object undergoes a transformation in state (Levin, 1993).

Comparatively little work has examined the neural bases of event structure templates. Kemmerer and Gonzalez-Castillo (2010) propose that event-level verb representations are subserved by schematic mirror neurons in the left inferior frontal gyrus/BA44. In contrast, root-level motor verb representations are thought to be subserved by somatotopically mapped mirror neurons in the left primary motor and/or premotor cortices (Kemmerer & Gonzalez-Castillo, 2010). Evidence for this hypothesis is limited, but includes findings from fMRI studies showing that BA44 mediates intention recognition (i.e., the act of interpreting why those around us act in a certain way; Brass et al., 2007; Iacoboni et al., 2005; Liepelt et al., 2008). Other evidence for this hypothesis includes studies of arm/hand observation-execution matching (i.e., the ability to simulate observed actions; Engel et al., 2008; Gazzola, Rizzolatti, et al., 2007; Gazzola, van der Worp, et al., 2007; Hamzei et al., 2003; Johnson-Frey et al., 2003; Schubotz & von Cramon, 2008), and impaired gesture recognition in people with BA44 lesions (Pazzaglia et al., 2008). The basal ganglia and their cortical connections may also play an important role in processing event semantic knowledge. In an fMRI study of healthy young adults, Tinaz et al. (2006) found that regions of the dorsolateral prefrontal cortex and the globus pallidus internal part of the basal ganglia showed greater activation in an event sequencing task involving the chronological arrangement of randomly-ordered pictures than in a control (object discrimination) task. People with Huntington’s disease, which affects the frontostriatal system (including the striatum, a central component of the basal ganglia), show impaired knowledge of certain activity-related task sequences, i.e. script events, though these impairments may correlate more with frontal and posterior cortices than with basal ganglia degradation (Allain et al., 2011). Thus, the current literature suggests that event processing may involve the basal ganglia as well as widespread regions of the neocortex.

Parkinson’s Disease (PD) presents a clear, promising avenue for furthering our understanding of verb processing because it affects multiple brain regions relevant to action and event knowledge. PD is associated with loss of dopaminergic neurons in the substantia nigra within the basal ganglia, which impairs the functioning of circuits that connect the basal ganglia to multiple areas of the cerebral cortex, including motor cortex, supplementary motor areas, and the dorsolateral prefrontal cortex (Alexander et al., 1986; J. Cummings, 1993; Middleton & Strick, 2000a, 2000b). These neurological changes lead to motor impairments such as tremor, rigidity, and bradykinesia (Jankovic, 2008; Sveinbjornsdottir, 2016). Although not required for a clinical diagnosis of PD, non-motor symptoms such as dementia, REM sleep behavior disorder, and depression are also common in this condition (Lim & Lang, 2010). The prevalence of tremor makes action language especially relevant to PD because tremor is often less well-controlled by dopaminergic medications than are other motor symptoms. Daily physical activities can be quite difficult for patients in even the early stages (Heusinkveld et al., 2018), and many PD patients’ hobbies involve minimal movement or motor control (Manson & Caird, 1985).

2.1.2. Verb-related deficits in Parkinson’s Disease

Evidence suggests deficits in verb production, particularly action verb production, in PD participants with age-typical cognitive performance as well as those with mild cognitive impairment and dementia, compared to healthy older adults. Performance on action fluency tasks is more impaired (relative to healthy adult controls) than performance on phonemic and semantic category fluency in participants with PD with (Piatt et al., 1999) and without dementia (McDowd et al., 2011; Rodrigues et al., 2015; Salmazo-Silva et al., 2017; Signorini & Volpato, 2006). Similarly, when producing nouns or verbs that are semantically related to a target word, PD participants without dementia are impaired in verb but not noun production (Crescentini et al., 2008; Péran et al., 2003; Silveri et al., 2018). Additionally, PD participants show greater impairments on verb than noun naming (Bertella et al., 2002; Bocanegra et al., 2017; Bocanegra et al., 2015; Cotelli et al., 2007; Garcia et al., 2017; Rodriguez-Ferreiro et al., 2009; Silveri et al., 2012). In narrative discourse, PD participants without cognitive impairment produce fewer action concepts (Garcia et al., 2016) and body-related verbs (Cousins et al., 2018) than do control participants. Overall, verb production impairments appear relatively early during PD across a variety of experimental and naturalistic tasks (e.g., Bocanegra et al., 2015; Cardona et al., 2013).

PD may also impair action comprehension. Garcia et al. (2018) found that PD participants without cognitive impairment showed selective deficits in comprehending verb-related information in narrative texts with high action content. In contrast, PD participants with mild cognitive impairment showed deficits in comprehending both verb-related and circumstantial information in both high- and low-action texts. These differences in action comprehension differentially affect how PD participants (versus controls) learn new verbs: PD participants were less likely to learn grammatical features of novel verbs, whereas both control and PD participants showed difficulty learning their semantic features (Murray Grossman et al., 1994). Additionally, ACEs have been shown in PD (Ibanez et al., 2013; Melloni et al., 2015), and smaller/absent action compatibility effects may correlate with reduced basal ganglia volume in PD participants (Melloni et al., 2015). In a sentence sensibility judgment task, Fernandino et al. (2013b) found that PD participants without dementia were slower to respond to sentences containing literal and idiomatic action compared to abstract sentences, whereas control participants’ response times did not differ among conditions. Similarly, PD participants, but not healthy older adults, were less accurate at judging the semantic similarity of action versus abstract verbs (Fernandino et al., 2013a). In contrast, Kemmerer et al. (2013) found that PD participants were generally as accurate as controls at judging verbs’ semantic similarity, and that although PD participants were slower than controls, they were slower across action and non-action verbs alike. Thus, although verb processing impairments are less consistently reported in comprehension tasks than in production tasks, verb processing may be impaired overall in PD.

2.1.3. Proposed explanations of verb impairments in PD

There are multiple competing accounts of verb impairments in PD. Of greatest relevance to the present studies are the following hypotheses: 1) Verb impairments may reflect a grammatical-class-based deficit, 2) Verb impairments may reflect impaired executive functioning skills, 3) Verb impairments may reflect a deficit in sensorimotor simulation abilities, and 4) Verb impairments may stem from deficits in event knowledge.

2.1.3.1. Grammatical explanation of verb impairments in PD.

Verbs may be represented separately from nouns (Caramazza & Hillis, 1991; Shapiro et al., 2006; Zingeser & Berndt, 1990; but see Crepaldi et al., 2013, Tyler et al., 2001) in frontostriatal brain regions that are susceptible to deterioration in PD (Alexander et al., 1986; Middleton & Strick, 2000b). Additionally, PD participants with and without cognitive impairment may experience grammatical processing deficits on a range of language comprehension and production tasks (Grossman, 1999; Grossman et al., 1992; Hochstadt, 2009; Johari et al., 2019; Kemmerer, 1999; Kotz et al., 2003). Consequently, it has been argued that action language impairments in PD reflect a grammatical class-based verb deficit (Grossman et al., 1994; Piatt et al., 1999). However, a number of studies that reported action language impairments in PD contrasted performance on high-action versus low-action items of the same grammatical class, with some specifically finding impairments related to high-action nouns (Bocanegra et al., 2017; Fernandino et al., 2013a, 2013b; E. Herrera et al., 2012; Johari et al., 2019). Therefore, differences in noun versus verb grammatical processing do not appear to fully account for action processing deficits in PD. In light of these findings, it is worth examining the role of PD-related cognitive and motor impairments in action language impairments.

2.1.3.2. Executive functioning explanation of verb impairments in PD.

Executive functioning (EF) skills are commonly impaired even in early stages of PD (Kudlicka et al., 2011; Fengler et al., 2017) and may be required to process complex or ambiguous language. For example, PD participants are impaired (compared to controls) at producing verbs within sentences, with greater verb impairment correlated with poorer set switching and working memory abilities (Colman et al., 2009). Importantly for this hypothesis, Silveri et al. (2018) and Di Tella et al. (2018) devised tasks in which verb production was easier than noun production owing to a smaller set of valid response options in the verb condition. In these tasks, PD participants showed greater impairment for nouns than verbs, suggesting that PD participants are challenged by increased selection demands, not by grammatical class per se. Interestingly, stimulating the substantia nigra (a core locus of neurodegeneration in PD) increases errors for items with lower selection constraint, suggesting the substantia nigra plays an important role in lexical selection (Castner et al., 2008). However, many studies that excluded participants with dementia or even mild cognitive impairment report altered verb processing in PD (e.g., Signorini & Volpato, 2006; Bertella et al., 2002; Bocanegra et al., 2017), suggesting that there may be additional causes of verb impairment in PD.

2.1.3.3. Sensorimotor simulation explanation of verb impairments in PD.

If sensory-motor simulation deficits degrade action language representation in PD, then greater motor impairment severity should correlate with higher error rates, slower response times, and/or altered recruitment of neural resources for action language processing compared to non-action language processing. However, correlations between motor severity rating scales and behavioral/neural outcomes reveal mixed findings. While a few studies reported significant correlations in the expected direction (Melloni et al., 2015; Salmazo-Silva et al., 2017), in others, motor severity scores did not correlate with action language processing (Garcia et al., 2018; Murray Grossman et al., 1994; Johari et al., 2019; Péran et al., 2003; Signorini & Volpato, 2006), while still others show correlations meeting significance only at a relaxed statistical threshold (Bocanegra et al., 2017; Cousins et al., 2018; Peran et al., 2009) or following outlier removal (Melloni et al., 2015). Contrary to grounded cognition predictions, Cousins et al. (2018) found that PD participants with more severe motor impairments produced more body-related verbs than those with less severe motor impairments, though this finding may be due to body-related verbs being more concrete and therefore easier for late-stage PD participants to activate. However, many studies correlated their outcomes to part III of the Unified Parkinson’s Disease Rating Scale (UPDRS), and it has been argued that the UPDRS part III does not assess fine motor function and may not reflect subtle changes in sensorimotor neural representations (Bocanegra et al., 2017; Cousins et al., 2018), despite its status as one of the most widely used measures of disease progression in PD (Opara et al., 2017). Additionally, the UPDRS part III may not correlate with biomarkers reflecting underlying disease pathology, such as the protein α-synuclein (Mollenhauer et al., 2019), and it is not designed to assess changes in sensation and other modalities involved in action simulation. While significant correlations between the UPDRS part III and measures of action language processing would strengthen claims that sensory-motor simulation plays a role in verb processing in PD, the relative lack of significant correlations does not negate this possibility.

Other studies support the sensorimotor simulation hypothesis using alternative approaches. In an action decision task, Roberts et al. (2017) found that PD patients with predominantly upper limb impairments took longer to respond to upper-limb verbs than lower-limb verbs. Participants with predominantly lower limb impairments did not show a predicted deficit in processing lower limb verbs compared to upper limb verbs. To explain the lack of relationship between lower limb impairment and processing of lower-limb-related verbs, Roberts et al. (2017) suggested that neuronal representations of lower limb actions may be less extensive and therefore less susceptible to PD-related deterioration. Thus, there is some evidence for effector-specific action language impairments in PD, in line with grounded cognition. Additionally, action language impairments may be more severe when PD participants are off dopaminergic medication or deep brain stimulation than when they are treated for their PD symptoms (Boulenger et al., 2008; Elena Herrera et al., 2015; E. Herrera & Cuetos, 2012; Kemmerer et al., 2013; Phillips et al., 2012; Silveri et al., 2012). Overall, it remains plausible that action language deficits are at least partially due to impaired sensory-motor simulation in PD, with most available evidence focused on the relationship between limb motor impairment and root-level motor verb representations.

2.1.3.4. Event-based explanation of verb impairments in PD.

Action knowledge may be disrupted in PD at the event structure level in addition to (or instead of) the root level. Both root-level and event-structure level features of verbs, such as goal recognition, are potentially representable via embodied action simulation (e.g., Rizzolatti & Sinigaglia, 2010); however, the neural mechanisms hypothesized to support each of these levels of representation differ (e.g., Kemmerer and Gonzalez-Castillo 2010), making it important to examine event representations specifically in PD. Evidence for event representation deficits in PD participants includes both behavioral and neuroimaging experiments that require the use of event-level knowledge such as temporal sequencing of actions. For example, Godbout and Doyon (2000) asked PD participants without dementia to produce the steps involved in everyday events such as “going to the cinema” in forward and backward order. In both conditions, PD participants, despite their absence of reported cognitive impairment, produced a greater proportion of sequencing errors, perseverative errors, and irrelevant intrusions than did matched controls. PD participants also produced relatively few minor supporting contextual elements in scripts produced in forward order. Godbout and Doyon suggested that the basal ganglia play a crucial role in representing and describing event knowledge. In an fMRI study, Tinaz et al. (2008) tested nondemented PD and healthy adult control participants in an event processing task, in which participants had to mentally sequence scrambled image sequences (e.g., a banana in various stages of being eaten); these participants also completed a control task that required them to identify living versus nonliving objects. In this study, PD participants showed intact performance in the control task but trended towards impaired accuracy on the event sequencing task. PD participants also showed abnormal patterns of neural activity in structures that have previously been found to be relevant to event processing (Tinaz et al., 2006; Seger & Cincotta, 2005): the prefrontal cortex, caudate, and “default mode” network. Similarly, when asked to indicate which sequences of events were either mis-ordered or contained irrelevant sub-events, PD participants without cognitive impairment produced a higher rate of errors than healthy older adults (Wyrobnik et al., 2022). Also, in response to content violations, PD participants did not show a right lateralized N400 effect that was present in control participants, suggesting that PD impairs the processing not only of event sequencing but also of event content (Wyrobnik et al., 2022). Additionally, when watching videos of everyday activities in an ERP study, PD participants have shown abnormal segmentation of event boundaries, with reduced theta band activity around identified event boundaries compared to healthy adult controls (Wyrobnik et al., 2023).

In narrative discourse production, PD participants with and without dementia produce significantly lower proportions of main event and information content than healthy older adults (Ash et al., 2017; J. L. Cummings et al., 1988; Holtgraves et al., 2013; Murray, 2000; Reddy et al., 2016; Roberts & Post, 2018). For example, given the discourse elicitation stimuli depicting an argument scene between a man and a woman (Nicholas & Brookshire, 1993), PD participants may omit the main concepts that a couple is fighting and that the woman crashes her car into a tree, or may omit some elements from these main concepts (Roberts & Post, 2018). However, in a visual-world eye-tracking study, PD participants without cognitive impairment and age-matched controls listened to sentences in which the target objects could be predicted from the subject and verb using event knowledge, such as “The fisherman rocks the boat” (Aveni et al., 2023). PD participants did not significantly differ from control participants in the proportions of looks to target objects, suggesting that people with PD can use their knowledge of real-world events during relatively constrained language comprehension tasks. Thus, the nature and extent of event knowledge impairments in PD remain unclear.

2.2. Overview of Present Studies

We investigated whether PD impairs event-based thematic verb processing. PD and healthy older adult participants generated and identified verbs denoting events that are commonly performed with a given instrument or at a given location. Given that motor networks are disrupted in PD, and given that verb instruments/tools particularly engage motor networks (Creem-Regehr & Lee, 2005; Grafton et al., 1997), we predicted that PD participants would show greater impairment for processing instruments than locations in both studies. Additionally, we predicted that PD participants with greater degrees of motor slowing (upper limb bradykinesia, assessed via part III of the UPDRS) would show greater impairment for instrument processing. We also predicted that PD participants with cognitive impairment (assessed using the Toronto Cognitive Assessment) would show greater impairment across all stimuli than PD participants without cognitive impairment. Study-specific rationale, predictions, methods, and results are presented separately for each study, followed by a combined Discussion.

All study procedures were approved by the Institutional Review Board at Northwestern University (Study Number STU00214374). Participants provided written consent for study participation and were compensated for their study participation. All data were collected and analyzed at Northwestern University in AR’s lab.

3. Study 1

3.1. Study Rationale and Predictions

Participants generated actions that people commonly perform using various instruments/at various locations, and we analysed the speed and relevance of participants’ responses. In healthy adults, both instruments and locations prime event-related verbs (McRae et al., 2005). Furthermore, instruments/tools cause greater involvement of motor cortices than do other objects (Grafton et al., 1997), and instruments may be more central to action verb processing than are locations (Rissman et al., 2015), thereby making instrument prompts high-action-simulation conditions and location prompts low-action-simulation prompts. Therefore, if PD primarily impairs sensory-motor processing of root-level features of verbs, then, in keeping with theories of grounded cognition, and particularly Kemmerer and Gonzalez-Castillo (2010), we predicted that PD participants should be selectively impaired on instrument-related but not location-related verb production. However, competing accounts of verb impairments in PD make alternative predictions, summarised in Appendix A. For example, if PD degrades verbs’ event representations more broadly, then PD participants should be impaired at producing verbs given both instruments and locations as cues. Alternatively, if PD does not affect the computation of verbs and their corresponding event structures, then PD and control participants should not differ on instrument- or location-related verb production.

The severity of PD participants’ motor disease and cognitive impairment may both affect semantic processing in PD. PD-related cognitive changes, particularly changes in executive functioning, are highly prevalent even in the prodromal stage (Fengler et al., 2017). Thus, it is both important to consider the impact of cognitive impairment and also infeasible to include only PD participants who have no degree of cognitive change. We predicted that PD participants with cognitive impairment would show greater impairments than PD participants without cognitive impairment across both instrument- and location-related verb production, based on findings that greater executive function impairments and higher selection demands lead to poorer performance on lexical processing tasks in PD (Castner et al., 2008; Colman et al., 2009; Di Tella et al., 2018; Silveri et al., 2018). We also predicted that participants with greater motor disease severity would show greater impairments for instrument items, based on existing findings of sensory-motor bases for verb processing impairments in PD (Boulenger et al., 2008; Herrera et al., 2015; Herrera & Cuetos, 2012; Melloni et al., 2015; Roberts et al., 2017; Silveri et al., 2012).

3.2. Method

3.2.1. Participants

Healthy older participants and participants with Parkinson’s disease were recruited from the research registry ResearchMatch (Harris et al., 2012) and from community sources (flyers, support group outreach programs) using a convenience sampling approach. Eligible participants between 50 and 90 years of age were required to have a minimum grade 10 reading ability based on the Quick Adult Reading Inventory (QARI; Chall et al., 2003) and speak English as their primary language. Multilingual participants were not excluded as long as they spoke English as their primary language in daily activities (e.g., work, home). Participants were also required to have adequate vision and hearing (with accommodations) to complete all neuropsychological testing and experimental surveys. Adequate hearing was evidenced by participant self-report and by their demonstrated ability to hear and follow spoken commands via video conference. Adequate vision was evidenced by self-report and at least 45/50 accurate responses on the QARI. Participants were also required to have access to the internet via a computer, tablet, or smartphone with a working microphone and camera. Participants were excluded if they self-reported a history of diagnosed major psychiatric or neurological illness (other than Parkinson’s disease). All PD participants were on stable PD medication, except for one participant who was PD medication de novo (i.e., had never taken medication for PD symptoms). One PD participant had undergone a deep brain stimulation procedure, and symptoms were being managed with stimulation and oral medications.

Healthy older adults were required to have intact cognition, and PD participants were required to have no more than mild cognitive impairment according to the brief version of the Cognitive Change Questionnaire (CCQ8; Damin et al., 2015), an initial screening tool that was followed later by a more comprehensive cognitive assessment. Specifically, healthy older adults were required to score 2 or less on the CCQ8, and participants with Parkinson’s disease were required to score 4 or less, based on the recommended cutoff scores for detecting mild cognitive impairment and dementia, respectively (Damin et al., 2015). The CCQ8 includes some items that may be affected by motor impairment, such as difficulty in using the phone to make calls and difficulty learning to use tools. Therefore, we encouraged PD participants to consider their cognitive rather than motor limitations when answering this questionnaire, and we used a higher cutoff score to minimise the risk of excluding otherwise eligible PD participants.

Forty-eight individuals indicated interest in the study and subsequently underwent screening. One person with PD was excluded due to uncertainty regarding the PD diagnosis. The remaining individuals with PD (N = 27) were enrolled. All healthy adult Control participants (N = 20) met the screening criteria and were enrolled. Of the enrolled participants, 3 (2 PD, 1 Control) did not respond to follow-up communications to schedule study procedures, 7 participants (6 PD, 1 Control) failed to complete all of the experimental surveys, and 2 Control participants were excluded when extended cognitive testing revealed them to score at a level consistent with mild cognitive impairment, making them ineligible per our study inclusion criteria. Thirty-five participants (n = 16 healthy older adults, n = 19 PD) completed all study procedures and were included in the analyses.

Summary demographic and disease data are presented in Table 1. PD and Control participants did not significantly differ in age or education. The gender distribution also did not significantly differ between groups, although there were more male than female PD participants, consistent with meta-analytic findings that PD affects approximately 1.46 males per 1 female (Taylor et al., 2007). PD participants’ motor symptoms ranged from mild to severe, with most participants exhibiting mild-moderate disease severity. Neuropsychological assessment data are presented in Table 2. There were no significant differences between PD and Control participants on any of the administered cognitive assessments. All Control participants had age-normal cognition and were independent in their activities of daily living (ADLs), as reported using the Lawton-Brody assessment (Lawton & Brody, 1969). PD participants had either age-normal cognition (15/19) or mild cognitive impairment (4/19) and reported minimal impairments in their activities of daily living, with any ADL impairments generally noted as being driven by their motor symptoms.

Table 1.

Demographic data and disease profile.

Variable	PD	Control	Sig. Testa
Gender	31.6% Female, 68.4% Male	56.3% Female, 43.8% Male	χ2 = 2.16 p = .142
Race or Ethnicity	94.7% White, 5.3% Hispanic	93.8% White, 6.3% Af. American	χ2 = 0.02, p = .900d
Handedness (self-reported)	84.2% Right, 5.3% Left, 10.5% Ambidextrous	75.0% Right, 25.0% Left	χ2 = 0.46, p = .497e
	Mean (SD) 95% Confidence Interval [Lower, Upper] Range
Age (yrs.)	69.83 (7.76) [66.09, 73.58] 51.8 – 80.7	68.31 (10.54) [62.69, 73.93] 52.4 – 83.2	t(33) = 0.49, p = 0.627
Education (yrs.)	18.47 (2.87) [17.09, 19.86] 12 – 23	18.56 (2.37) [17.30, 19.82] 16 – 24	t(33) = 0.10, p = 0.922
Self-maintaining ADL impairments (/24)	0.95 (1.18) [0.38, 1.51] 0 – 4	0.00 (0.00) [0.00, 0.00] 0 – 0	t(18) = 3.51, p = 0.003
Instrumental ADL impairments (/8)	0.42 (0.84) [0.02, 0.82] 0 – 3	0.00 (0.00) [0.00, 0.00] 0 – 0	t(18) = 2.19, p = 0.042
MDS UPDRS Part II ( /52)b	13.58 (8.63) [9.42, 17.74] 1 – 31	-	-
MDS UPDRS-III ( /48)b, c	16.53 (6.54) [13.38, 19.68]	-	-
	6 – 32
MDS UPDRS-III Upper limb bradykinesia ( /24)	9.84 (5.18) [7.35, 12.34] 2 – 24
Years since PD diagnosis	5.72 (3.93) [3.82, 7.61] 0.3 – 13.0	-	-

Note. ADL = activities of daily living

^aLower df reflects Welch’s t-test for unequal variances

^bAll PD participants were assessed using parts of the Unified Parkinson’s Disease Rating Scale (MDS-UPDRS). The UPDRS-II is a standardised self-report measure of motor impairment, and the UPDRS-III is a standardised examiner-assessed measure of motor impairment (Goetz et al., 2008). For the UPDRS-III, a subset of items were administered as described in the text

^cUpper limb bradykinesia score, per formula described in Goetz et al. (2008)

^dTested percentage of non-white participants by group

^eTested percentage of non-right-handed participants by group

Table 2.

TorCA Neuropsychological assessment data.

Variable	Mean (SD) 95% Confidence Interval [Lower, Upper] Range
	PD	Control	Significance Testa
Summary Scores
Orientation ( /12)	11.74 (0.45) [11.52, 11.95] 11 – 12	11.81 (0.40) [11.60, 12.03] 11 – 12	t(33) = 0.52, p = 0.61
Immediate Memory Recall ( /30)	20.95 (3.73) [19.15, 22.75] 15 – 30	21.88 (2.60) [20.49, 23.26] 17 – 27	t(33) = 0.84, p = 0.41
CERADb trial 1 ( /10)	5.63 (1.77) [4.78, 6.48] 3 – 10	5.81 (1.33) [5.11, 6.52] 4 – 9	t(33) = 0.34, p = 0.74
CERAD trial 2 ( /10)	7.42 (1.21) [6.83, 8.01] 5 – 10	7.69 (0.95) [7.18, 8.19] 6 – 9	t(33) = 0.71, p = 0.48
CERAD trial 3 ( /10)	7.89 (1.15) [7.34, 8.45] 6 – 10	8.38 (0.89) [7.90, 8.85] 7 – 10	t(33) = 1.36, p = 0.18
Delayed Memory Recall
CERAD recall ( /10)	6.05 (1.78) [5.20, 6.91] 3 – 10	6.88 (1.88) [5.89, 7.86] 3 – 10	t(33) = 1.34, p = 0.19
Benson figure recall ( /17)	12.06 (2.90) [10.61, 13.50] 6 – 16	12.40 (3.02) [10.73, 14.07] 8 – 17	t(31) = 0.33, p = 0.74
Delayed Memory Recognition
CERAD recognition ( /20)	19.58 (0.96) [19.12, 20.04] 17 – 20	19.56 (0.73) [19.17, 19.95] 18 – 20	t(33) = −0.06, p = 0.96
Benson figure recognition ( /1)	0.89 (0.32) [0.73, 1.05] 0 – 1	0.93 (0.26) [0.79, 1.08] 0 – 1	t(31) = 0.43, p = 0.67
Visuospatial
Clock drawing ( /15)	14.11 (1.32) [13.45, 14.77] 10 – 15	14.00 (0.97) [13.49, 14.51] 12 – 15	t(32) = −0.28, p = 0.78
Benson figure copy ( /17)	15.95 (1.31) [15.32, 16.58] 12 – 17	15.47 (1.19) [14.81, 16.12] 12 – 17	t(32) = −1.11, p = 0.28
Working Memory / Attention / Executive Control
Longest digit span forward ( /9)	7.37 (1.26) [6.76, 7.97] 5 – 9	7.38 (1.31) [6.68, 8.07] 5 – 9	t(33) = 0.02, p = 0.99
Longest digit span backwards ( /8)	5.42 (1.77) [4.57, 6.28] 3 – 8	5.25 (1.18) [4.62, 5.88] 3 – 7	t(33) = −0.33, p = 0.74
Serial subtractions – 7 ( /13)	12.28 (1.45) [11.56, 13.00] 8 – 13	12.56 (0.81) [12.13, 13.00] 10 – 13	t(27.33)a = 0.72, p = 0.48
Serial subtractions – 3 ( /13)	12.42 (1.43) [11.73, 13.11] 7 – 13	12.94 (0.25) [12.80, 13.07] 12 – 13	t(19.31)a = 1.55, p = 0.14
Alternating sequences ( /2)	1.58 (0.77) [1.21, 1.95] 0 – 2	1.94 (0.25) [1.80, 2.07] 1 – 2	t(22.38)a = 1.92, p = 0.07
Verbal fluency (letter) (no maximum)	15.89 (5.69) [13.15, 18.64] 6 – 27	14.75 (4.01) [12.61, 16.89] 7 – 23	t(33) = −0.67, p = 0.50
Language
Verbal fluency (animal) (no maximum)	22.74 (6.76) [19.48, 26.00] 11 – 37	23.13 (4.35) [20.81, 25.44] 14 – 27	t(33) = 0.20, p = 0.84
Naming (MINT) ( /15)	14.68 (0.48) [14.45, 14.91] 14 – 15	14.69 (0.48) [14.43, 14.94] 14 – 15	t(33) = 0.02, p = 0.98
Sentence repetition ( /10)	9.79 (0.54) [9.53, 10.05] 8 – 10	9.56 (0.81) [9.13, 10.00] 8 – 10	t(33) = −0.99, p = 0.33
Sentence comprehension ( /8)	6.95 (1.27) [6.34, 7.56] 4 – 8	7.44 (0.89) [6.96, 7.91] 5 – 8	t(33) = 1.30, p = 0.20
Single word reading comprehension ( /2)	2.00 (0.00) [2.00, 2.00] 2 – 2	2.00 (0.00) [2.00, 2.00] 2 – 2	N/A
Single word reading ( /12)	12.00 (0.00) [12.00, 12.00] 12 – 12	11.94 (0.25) [11.80, 12.07] 11 – 12	t(15)a = −1.00, p = 0.33
Single word comprehension ( /8)	8.00 (0.00) [8.00, 8.00] 8 – 8	8.00 (0.00) [8.00, 8.00] 8 – 8	N/A
Semantic knowledge ( /10)	9.79 (0.42) [9.59, 9.99] 9 – 10	9.56 (1.03) [9.01, 10.11] 6 – 10	t(19.15)a = −0.83, p = 0.42

Note. Missing data are as follows: N = 2 (1 PD, 1 Control) Benson figure recall; N = 2 (1 PD, 1 Control) Benson figure recognition; N = 1 (PD) Clock drawing; N = 1 (Control) Benson figure copy; N = 1 (PD) serial 7 subtractions.

^aLower df reflects Welch’s t-test for unequal variances

^b Chandler et al., 2005

3.2.2. Materials

We selected 30 instruments (e.g., fork and pen) and 30 locations (e.g., airport and office) that were expected to be familiar to all participants (see Appendix B). Items were drawn in large part from McRae et al. (2005) and were supplemented with items that were expected to elicit diverse action responses. In particular, the added instruments were expected to elicit actions that saliently involved upper limb manipulations, making them high-action-simulation conditions. Sample items for Study 1 were as follows:

• List the actions people commonly perform using a FORK

• List the actions people commonly perform in/at an AIRPORT

3.2.3. Procedure

3.2.3.1. General study procedures.

Because participants were recruited during the COVID-19 pandemic, all procedures were completed remotely. Participants were screened via a brief (10–15 minute) telephone call by the first author or a trained research assistant. Items that required participants to view stimuli (i.e., the QARI) were sent to their email addresses. Eligible participants then scheduled their remote assessment session and were sent an electronic consent form through REDCap to review and sign. To confirm a valid consent, at the beginning of the remote study visit, the examiner asked the participant true/false questions about key aspects of the study and reviewed the corresponding sections of the consent form if needed. Remote neuropsychological and motor assessment sessions lasted approximately one hour.

Following completion of the neuropsychological and motor assessment session, the examiner sent the participant a personalised link to take the survey via Qualtrics. Participants completed this survey independently, generally within 1–14 days of their remote assessment. Participants were asked to take one minute per item to generate actions that they felt were commonly performed using the provided instruments or in/at the provided location (the full instructions are reported in Appendices C.1 and C.2). Participants were shown one item per page, with no option to return to prior items, and were provided 10 blank lines per item, with the number of blank lines remaining constant regardless of the participants’ actual productivity. Participants were required to spend a minimum of 30 seconds on each item, but there was no maximum time limit to allow participants to take breaks, and to accommodate participants with significant motor impairments. PD participants were scheduled in their optimal ON state based on their self-reported best time of day and were asked to complete the surveys themselves. While this possibly contributed to longer completion times for PD participants, no participants reported they were unable to complete the study due to the typing demands, and all participants’ responses were legible with minimal typographic errors. The median survey completion time was 142.7 minutes (232.8 PD, 123.2 Control). PD participants spent a mean of 109.4 seconds per item, compared to a mean of 83.2 seconds per item for Control participants.

3.2.3.2. Neuropsychological and motor assessment.

Participants completed neuropsychological assessments via HIPAA-compliant Zoom video call, and PD participants completed motor assessments during that same call. Participants were assessed by the first author or by a trained research assistant.

To assess the severity of PD participants’ motor disease, we remotely administered parts of the Unified Parkinson’s Disease Rating Scale (UPDRS; Goetz et al., 2008) to all PD participants according to the recommendations published by the Movement Disorders Society (Telemedicine in Your Movement Disorders Practice: Step 4, 2021). Recent studies suggest that remote administration of the UPDRS part III (motor exam) generally is feasible, although some items, such as those assessing participants’ muscle rigidity, cannot be administered remotely (Abdolahi et al., 2013; Schneider et al., 2020; Stillerova et al., 2016; Tarolli et al., 2020). In a reliability study, Tarolli et al. (2020) reported excellent correlations between remote and in-person assessments for the patient-reported motor symptom scale (part II) and moderate correlations between remote and in-person assessments for the motor exam (part III). In the present study, we administered all items of the UPDRS part II scale as well as the following items from the UPDRS part III scale: speech, facial expression, finger tapping (R + L), hand movements (R + L), hand pronation-supination (R + L), postural tremor of the hands (R + L), and kinetic tremor of the hands (R + L). Disagreements in UPDRS part III scores were resolved through consensus between the two primary raters and, when necessary, by consulting a third expert rater. As detailed in Table 1, PD participants’ UPDRS Part III scores generally indicated mild or mild-moderate upper limb impairment.

All participants provided their demographic information and completed the Lawton-Brody assessment of self-maintaining and instrumental activities of daily living (Lawton & Brody, 1969). The experimenter then administered all orientation, immediate recall, delayed recall, working memory/attention/executive control, and language subtests of the Toronto Cognitive Assessment (TorCA; Freedman et al., 2018) according to published protocols, with the exception of the trail-making tests, which were not administered due to the constraints of remote testing. Participants were asked to conduct both their Zoom call and their survey responses on a desktop or laptop computer (Mac or Windows) due to the larger screen size and access to a keyboard, and only one participant was unable to do so. To accommodate this participant’s request to use their iPad, we conducted a preliminary Zoom call in which we provided a sample word list, a sample image, and a sample survey. This call determined that the participant’s use of the iPad would not be a hindrance to their participation, and the rest of the procedure went forward in the same way as it did with the other participants. Assessment sessions were audio/video recorded for offline double-scoring and fidelity procedures. Disagreements in neuropsychological test scoring were resolved through the same consensus procedures that were used for UPDRS scoring.

To categorise participants as cognitively impaired or unimpaired, we compared raw test scores to published normative data for the TorCA (Freedman et al., 2018). Following published guidelines, participants were categorised as impaired on a given test if they scored at or below the 5^th percentile (1.65 standard deviations below the mean) for their age group. Participants were considered to have cognitive impairment if they were rated as impaired on two or more tests (either within one domain or across multiple domains), in keeping with the Movement Disorders Society’s recommendation for comprehensive (Level II) cognitive assessment (Litvan et al., 2012). In cases where participants’ performance may have been impacted by test administration errors, the affected scores were not applied towards the determination of cognitive impairment. Amongst the Control cohort, 2/18 met the criteria for cognitive impairment and were excluded from further analyses, leaving 16 participants in the Control group.

3.2.4. Analyses

We calculated three primary measures of participants’ overall generativity and response relevancy: the number of seconds per response, and the number and percentage of responses that were relevant (both a valid action and not a repetition of an earlier response). Table 3 presents the full list of both primary and intermediate outcome measures, along with their definitions. To minimise the impact of participant ‘idle’ time on seconds per response, we removed outliers, defined as response durations above and below 1.5 times the interquartile range, leaving 2007/2100 items (and 13496/13969 total responses to these items) available in the latency analysis.

Table 3.

Definitions of free-response categories and outcome measures.

Measure	Definition
Total responses	Number of separate responses provided by the participant for a given item. Responses were counted as separate if they were written in different response fields. Responses were also counted as separate if they were written in the same response field but could otherwise count as two (or more) valid responses.
Seconds per response	Seconds spent on an item ÷ Total responses for the item
Non action	Responses that neither explicitly stated a verb nor implied a valid action.
Implied action	Responses that did not explicitly state a verb but that implied the performance of an action and/or an action use for an instrument item.
Invalid verb	Responses that explicitly stated a verb but were not considered semantically valid.
Non-instrumental action	Responses that contained a verb that could not plausibly be performed using the given object as an instrument.
Literal repetition	Responses that re-used the same verb or verb phrase (regardless of whether they were semantically repetitive).
Semantic repetition	responses that were determined to refer to the same action (regardless of whether they were literally repetitive).
Total verbal relevant responses	Raw number of literally valid and novel responses. - ‘Total responses’ minus ‘Non actions’ minus ‘Implied actions’ minus ‘Literal repetitions’.
Total semantically relevant responses	Raw number of semantically valid and novel responses. - ‘Total responses’ minus ‘Non actions’ minus ‘Invalid verbs’ minus ‘Semantic repetitions’.
% Verbal relevant responses	Percentage of literally valid and novel responses. - Total verbal relevant responses ÷ Total responses * 100
% Semantically relevant responses	Percentage of semantically valid and novel responses. - Total semantically relevant responses ÷ Total responses * 100

Note: Primary outcome measures are in bold text.

Next, to capture the number and percentages of relevant responses produced by each participant, responses were coded for whether they contained an explicit (literal) verb and whether they indicated a valid action that might be performed with a given instrument or at a given location. ‘Non action’ responses neither explicitly stated a verb nor implied a valid action. These responses were generally nouns or noun phrases, as in “speedometer needle (mph in a car)” for needle, and “boundary” for rod. ‘Implied actions’ did not explicitly include a verb but implied an action and/or an action use for an instrument item, as in “flight (witches only)” for broom, “[used] for ventilation” for hose, and “laundry” for detergent. ‘Invalid verbs’ explicitly stated a verb but were not considered semantically valid, such as “play rock paper scissors” for paper, and “chant ‘up your nose with a rubber hose’” for hose. Additionally, we coded two kinds of repetitions: ‘literal repetitions’ and ‘semantic repetitions.’ Literal repetitions re-used the same verb or verb phrase, typically accompanied by a different object. Examples include “breaking up ice” and “breaking up stone” for axe, “serving as a bib” and “serving as a lap protector” for napkin, and “use bathroom” and “use reference material” for library. Conversely, semantic repetitions referred to the same action, even if the word(s) differed. Examples include “pick up” versus “lift” for fork and “stick into” versus “insert” for needle. Literal and semantic repetitions were not mutually exclusive, and one response could be counted as both. Three percent of responses were coded as literally repetitive, and 1.9% were coded as semantically repetitive. All responses were double-coded, and all discrepancies were resolved either via consensus or by the first author.

For the outcome measures reflecting relevant responses, we calculated one set based on the literal verbs that were produced and a second set based on the semantic content. For example, verbal relevant responses were required to contain a unique verb, whereas semantically relevant responses contained any unique action, whether or not a verb was provided. As a result, the percentage of semantically relevant responses was generally equal to or greater than the percentage of verbal relevant responses, with the exception of a few responses that were semantically but not verbally repetitive (e.g., “stick into” versus “insert” for needle) and responses that were semantically invalid but contained verbs (e.g., “play rock paper scissors” for paper).

Within the instrument responses, we identified a number of ‘non-instrumental actions’ that contained a verb that could not plausibly be performed using the given object as an instrument. In most of these cases, the instrument had instead been assigned to a patient role. Examples include “eat them” for crayons, “install” for deadbolt, and “wash it” for fork. In total, 1.8% of all responses were coded as non actions, 2.4% as implied actions, 0.1% as invalid verbs, and 9.2% of instrument responses were coded as ‘non-instrumental actions’. Sample participant responses and their resulting outcome values are illustrated in Appendix D.

Finally, we used the Hand_arm.mean action values from the Lancaster Sensorimotor Norms (Lynott et al., 2020), which ranged from 0 (low hand/arm involvement) to 5 (high hand/arm involvement) to analyze the degree of upper-limb action content inherent in participants’ free responses. For multi-word responses, the core action word was identified and was used to determine the Lancaster normative rating (e.g., the value for “answer” was used for the response “answer questions,” the value for “fill” was used for the response “fill potholes”). Inflection-specific values were used whenever possible (e.g., “baking” versus “bake” received different ratings based on Lancaster normative data). Minor spelling mistakes were corrected on the participant’s behalf where possible (e.g., “badmitton” was coded as “badminton” and “art work” was coded as “artwork”). However, responses were excluded if a single core word could not be unambiguously identified (e.g., “get help with downloading documents,” “Mac the Knife (song)”. In total, Lancaster normative data were available for 12,774 out of 13,279 responses (96.2%).

For the proportions of relevant responses, there were no missing free response data, resulting in 2100 observations. (35 participants [19 PD, 16 control] * 60 items [30 locations, 30 instruments]). However, these observations were not independent, as responses were nested within participants. Therefore, we analysed the free response data using mixed effects models, with fixed effects of group and item type as well as their interaction effect and random intercepts for participants and items, e.g., lmer(% Relevant Verbs ~ Group * QuestionType + (1 | Participant_ID) + (1 | Item)), with the exception of the seconds per response model, for which a random effect of item could not be calculated. Mixed effects models were conducted in R v4.2.2 using lme version 1.1–30, and p-values were estimated using the lmerTest package, version 3.1–3 (Kuznetsova et al., 2017). Random slopes were not included for any models in order to avoid model overfitting and potential failures of model convergence, and all models converged with no errors or warnings. Group, item type, and cognitive status were sum coded (Control −0.5, PD +0.5; Instruments −0.5, Locations +0.5; Age-normal cognition −0.5, Mild cognitive impairment +0.5).

3.3. Results

3.3.1. Comparisons between PD and Control participants

We assessed main effects and their interactions with group (Control versus PD) and item type (Instruments versus Locations) for the number of seconds per response, the percentage of verbally and semantically relevant responses, and mean hand/arm action values per Lancaster norms. Detailed results are presented in Tables 4 and 5.

Table 4.

Descriptive statistics for the event-related verb production survey (non-instrumental responses included).

	Mean (SD)
	Sec. per Response		% Relevant Verb Resp.		% Sem. Relevant Resp.		Hand/Arm Action Rating
	Instruments	Locations	Instruments	Locations	Instruments	Locations	Instruments	Locations
Control	22.18 (19.35)	13.22 (12.12)	88.31 (21.32)	94.06 (13.81)	92.93 (16.65)	97.39 (8.42)	3.18 (0.98)	2.50 (1.14)
PD	27.37 (21.71)	15.21 (9.29)	92.28 (15.54)	95.26 (10.20)	95.13 (13.71)	98.02 (6.25)	3.17 (1.01)	2.42 (1.18)

Table 5.

Analyses of PD versus Control responses in the event-related verb production survey (non-instrumental responses included).

Fixed Effects	Seconds per Response				% Relevant Verb Responses				% Sem. Relevant Responses
	Est.	S.E.	t value	p value	Est.	S.E.	t value	p value	Est.	S.E.	t value	p value
Intercept	19.59	1.19	16.53	< .001	92.48	0.90	103.28	< .001	95.87	0.66	145.34	< .001
Item type (Instruments/Locations)	−10.67	0.68	−15.80	< .001	4.37	1.19	3.68	< .001	3.68	1.26	5.75	< .001
Group (Control/PD)	3.50	2.37	1.48	0.150	2.59	1.49	1.74	.091	1.41	0.64	1.13	0.269
Item type x Group	−3.43	1.35	−2.54	.011	−2.78	1.28	−2.18	.029	−1.58	0.99	−1.60	0.111
Random Effects
τ00 Participant_ID	44.81				15.65				11.56
τ00 Item	N/A				14.97				2.46
σ2	227.79				211.94				127.86
N Participant_ID	35				35				35
N Item	N/A				60				60
Observations	2007				2100				2100

Note: Bolded values are significant at the p > .05 level. τ00 = variance associated with random intercepts, σ2 = residual variance

For the number of seconds per response, there was a significant group by item type interaction, with PD participants taking more time to respond to instruments than Control participants, whereas there was no difference for location cues. The main effect of item type was also significant, with participants taking longer per response on instruments than locations, while the main effect of group did not reach significance.

For the percentage of verbal relevant responses, item type and group again interacted. Control participants had a higher percentage of verbal relevant responses to locations than instruments, whereas PD participants showed relatively similar performance on both item types. In contrast, when we analyzed the percentage of semantically relevant responses, only the main effect of item type was significant, indicating that when responses were not required to contain a literal verb, PD and Control participants showed similar patterns of performance across both item types.

As a secondary set of outcome measures, we also analyzed the number of total responses and the number of verbal and semantically relevant responses. Results are detailed in Appendix E. For all of these measures, there were significant group by item type interactions, such that greater numbers of responses were generated for locations than instruments, particularly for PD participants.

For Lancaster Hand_arm action values, see Table 4 for descriptive statistics. There was a significant main effect of item type, with higher action values produced in response to instruments versus locations, β = −0.71, S.E. = 0.10, t = −7.41, p < .001. There was no significant main effect of group, β = −0.05, S.E. = 0.04, t = −1.20, p = 0.237. However, there was a marginally significant group by item type interaction, β = −0.08, S.E. = 0.03869, t = −1.95, p = 0.051, with PD participants showing lower action content in response to location cues than Control participants, whereas there was no difference for instrument cues. The random intercept variance associated with items was 0.13, the random intercept variance associated with participants was 0.01, and the residual variance was 1.09, with a total of 12773 observations in the model (60 items, 35 participants).

PD and Control participants did not significantly differ in the percentage of non-instrumental actions produced in response to instrument items (e.g., “eat them” for crayons), β = 3.94, S.E. = 2.66, t =1.48, p = .15. Nonetheless, to understand whether excluding non-instrumental responses influenced the findings for the seconds per response or percentages of relevant responses, we repeated these analyses with non-instrumental responses excluded. As detailed in Appendices F and G, the pattern of findings did not change, with the exception that the main effect of group on seconds per response became statistically significant, with PD participants taking longer per response.

3.3.2. Motor and cognitive correlations in PD participants.

We next assessed whether total upper limb bradykinesia scores (raw scores from the UPDRS-III) or cognitive status (sum-coded) significantly correlated with PD participants’ free response measures. We modeled fixed effects of item type, cognitive status, and bradykinesia scores as well as interactions between item type and cognitive status and between item type and bradykinesia, as well as random intercepts for participant and item, e.g., lmer(% Relevant Verbs ~ Type * Cog_status + Type * UPDRS_total_upper_bradykinesia + (1 | Participant_ID) + (1 | Item)). A three-way interaction among item type, cognitive status, and bradykinesia scores was not estimated because of a lack of a clear theoretical reason to expect a three-way interaction and because of the low number of PD participants with both cognitive decline and notable bradykinesia. Detailed inferential statistics are presented in Table 6. Table 7 presents descriptive statistics for PD participants with age-normal cognition versus those with mild cognitive impairment. Significant interactions between bradykinesia scores and primary outcome measures are depicted in Fig. 1.

Table 6.

Effect of bradykinesia and cognitive status on event-related verb production in PD participants (non-instrumental responses included).

Fixed Effects	Seconds per Response				% Relevant Verb Responses				% Sem. Relevant Responses
	Est.	S.E.	t value	p value	Est.	S.E.	t value	p value	Est.	S.E.	t value	p value
Intercept	22.06	3.98	5.55	< .001	93.77	1.68	55.79	< .001	100.54	1.21	82.94	< .001
Item type	−7.02	2.21	−3.18	< .01	5.63	1.80	3.13	< .01	−2.50	1.35	−1.85	.064
Bradykinesia score	0.60	0.36	1.65	.116	0.09	0.15	0.59	.562	−0.41	0.11	−3.66	.002
Cognitive status	−2.48	4.54	−0.55	.591	−4.22	1.88	−2.24	.039	0.43	1.39	0.31	.759
Item type x Bradykin.	−0.65	0.18	−3.59	< .001	−0.31	0.15	−2.05	.040	0.59	0.12	4.83	< .001
Item type x Cognition	5.10	2.33	2.19	.029	1.86	1.86	1.00	.318	−2.24	1.52	−1.47	.142
Random Effects
τ00 Participant_ID	52.86				7.96				4.01
τ00 Item	17.20				9.44				1.10
σ2	211.93				154.05				103.08
N Participant_ID	19				19				19
N Item	60				60				60
Observations	1069				1140				1140

Note: Bolded values are significant at the p > .05 level.

Table 7.

Descriptive statistics for the effect of cognitive impairment on event-related verb production (non-instrumental responses included).

			Mean (SD)
	Sec. per Resp.		% Relevant Verb Resp.		% Sem. Relevant Resp.
	Instruments	Locations	Instruments	Locations	Instruments	Locations
Age-Normal	27.41 (21.79)	14.62 (9.27)	93.21 (14.48)	95.99 (8.93)	95.26 (14.09)	98.23 (5.68)
Cog. Impaired	27.23 (21.49)	17.76 (8.94)	88.80 (18.65)	92.50 (13.67)	94.65 (12.19)	97.20 (8.01)

Figure 1.

Interactions between upper limb bradykinesia scores and item type in PD participants’ event-related verb productions.

Note. A) model-estimated seconds per response, B) model-estimated percentage of verbal relevant responses, and C) model-estimated percentage of semantically relevant responses. Error bars represent 95% confidence levels for estimated marginal means.

For seconds per response, item type significantly interacted with bradykinesia severity, with higher bradykinesia scores relating to longer responses for instrument items. Next, for the percentage of verbal relevant responses, there were significant main effects of item type and cognitive status, with greater percentages of relevant responses found for cognitively age-normal participants and for locations. There was also a significant interaction between bradykinesia scores and item type, such that the difference in percentage of verbal relevant responses between locations and instruments was diminished for participants with higher bradykinesia scores. Finally, for the percentage of semantically relevant responses, there was a significant interaction between bradykinesia scores and item type as well as a significant main effect of bradykinesia scores. Higher bradykinesia scores were associated with a lower percentage of semantically relevant responses, particularly for instrument items.

3.2.2.1. Summary of Study 1 findings.

On average, PD and Control participants showed similar generativity and response relevancy in the free response survey, which suggests that PD does not globally impair the activation of event knowledge during lexical-semantic processing. Yet the interactions between group and item type and between bradykinesia scores and item type, as well as reduced hand/arm action content in PD participants’ location responses, suggest that impaired sensory-motor simulation may alter PD participants’ ability to activate high-action verbs in relation to events. Importantly, the significant interaction in seconds per response cannot be readily explained by differences in typing speeds, since slower typing speeds in PD participants would be expected to impact responses to instruments and to locations equally.

The fact that both groups generated a greater number of responses (and responded faster) for locations than instruments indicates that the instruments condition was more difficult overall, so it may be argued that this finding was driven by cognitive differences between groups, particularly in executive functioning. However, on average, PD and Control participants did not significantly differ on any measures of executive functioning, suggesting that group differences in executive functioning do not fully account for the observed group differences. Additionally, while there was a significant main effect of cognitive status on the percentage of verbal relevant responses, a similar effect was not found for the percentage of semantically relevant responses or for seconds per response, suggesting that PD-related cognitive changes impair verb processing but not action concept processing more generally. Finally, if cognitive changes underlie PD participants’ slower responses on instruments versus locations, then cognitively impaired PD participants should show an exaggerated effect of item type compared to cognitively age-typical participants; but instead, cognitively impaired PD participants showed less of a difference in response speed between instruments and locations than did age-typical participants. Overall, these observations suggest that generating instrument-related actions was impaired over and above any broader effects of cognitive change on verb processing in PD.

4. Study 2

4.1. Study Rationale and Predictions

Findings from Study 1 suggest that PD participants are impaired at using event knowledge to produce verbs in high-action-simulation conditions (i.e., for instrument prompts) but not low-action-simulation conditions (i.e., for location prompts). However, as summarised in the Introduction, verb processing impairments are reported less consistently in comprehension-based tasks than in production-based tasks, making the full extent of action and event impairment in PD unclear. Therefore, in Study 2, we assessed PD participants’ ability to identify, rather than produce, event-based associates of instrument and location items. In the verb-specific grammatical account, the PD group would be expected to be impaired on rating both instrument-verb pairs and location-verb pairs. The executive function account would predict impairment across stimuli only for PD participants with cognitive decline. Based on the strong version of sensorimotor account of verb processing, PD participants would be expected to be more impaired on rating instrument-verb pairs than location-verb pairs, although the weak form of this account may permit PD and Control participants to show equivalent performance in this task. Finally, based on the event-structure-based account, PD participants should be impaired on identifying event-related verbs in addition to producing them. Alternatively, if PD only partially disrupts the use of event knowledge during language processing, then PD participants may show an intact ability to identify event-related verbs despite relatively impaired event-related verb production abilities. As in Study 1, we additionally predicted that PD participants with greater motor disease severity would show greater impairments for instrument ratings, whereas PD participants with cognitive impairment were predicted to be impaired on both instrument and location ratings.

4.2. Method

4.2.1. Participants

Study 2 participants were the same as those described in Study 1.

4.2.2. Materials

Items for Study 2 were based on the same 30 instruments and 30 locations used in Study 1. For each item, we provided participants with 4–6 verbs (or verb phrases, when necessitated by the verb’s argument structure, e.g., ‘opening envelopes’ for razor; and/or to provide semantic specificity, e.g., ‘making bubbles’ for detergent). Verbs (and verb phrases) were selected based on responses provided by healthy young adults in a similar previous study (McRae et al., 2005). We asked participants to use a Likert scale to rate how commonly each proposed action featured in events with the given instrument or location. We intentionally selected actions that were expected to range from low to high ratings. Likert survey items and mean participant ratings for each item are detailed in Appendix H and were presented to participants as shown in Fig. 2. There was one block of instruments and one of locations. Participants were randomly assigned to receive either the verb-instrument or verb-location block first (counterbalanced across participants). In addition, the ordering of the verbs within each instrument or location item was randomised for each participant.(Figure 2 about here)

Figure 2.

Sample Study 2 survey item (instruments survey).

4.2.3. Procedure

Participants completed remote cognitive and motor assessments as described in Study 1 before receiving a Qualtrics survey link. To minimise interference across studies, participants received the Likert survey approximately 7 days after completing the Study 1 free-response survey. Participants completed this survey independently, and all participants completed Study 2 within 45 days of their assessment. The full instructions provided to participants are reported in Appendices C.3 and C.4. The median survey completion time was 26.6 minutes (28.2 PD, 23.7 Control).

4.2.4. Analyses

To analyze participants’ ratings, we first determined the ‘typical’ response for each item within a given survey by calculating the per-item means of all Control participants’ responses. We then used the cor.test function in R v4.2.2 to conduct Pearson’s correlations comparing each individual PD participant’s item responses to the Controls’ item mean responses. Similarly, for Control participants, Pearson’s correlations were conducted between individual Controls’ item responses and the mean of all other Controls’ responses to the same item. This leave-one-out approach to calculating Controls’ correlations ensured that Controls participants’ own scores were not included in their comparator group and would therefore not be statistically inflated compared to the PD participants’ correlations. This procedure therefore yielded a parallel set of correlation measures in each participant group that captured the relative similarity of each participant’s item responses to the mean of all typical item responses. Sample ratings and resulting Pearson correlations are depicted in Fig. 3. One response was missing and was removed pairwise (i.e., the remainder of that participant’s data were retained and used) to allow for maximal data retention. The resulting participant-wise Pearson’s r correlation coefficients therefore reflected each individual participant’s degree of difference from the ‘typical’ response pattern. Because only one such value could be generated for each participant, these data were not nested within participants and so were most appropriately modeled with ANOVAs rather than mixed effects models. The data were entered into two-way ANOVAs to assess main effects and interactions of group and item type. ANOVA models and η_p² effect sizes were calculated using version 3.1–1 of the car package. Group, item type, and cognitive status were sum-coded as described for Study 1, and estimates were calculated using type III sum of squares.(Figure 3 about here)

Figure 3.

Sample ratings from individual participants and resulting correlations with Control means, from Study 2.

Note. Panel A depicts ratings from a participant with a high Pearson correlation value (.94). Panel B depicts ratings from a participant with a relatively low Pearson correlation value (.85). Each point represents the rating given to an individual item within the Likert event-based survey. The blue regression line shows the correlation for each participant.

4.3. Results

Fig. 4 illustrates participants’ correlation coefficients for Study 2, stratified by group and item type.

Figure 4.

Event-related verb identification (Likert) ratings by group and item type.

Note. Individuals’ Pearson correlation coefficients from the event-related Likert survey, by item type (combined versus instruments versus locations) and by group (Control versus PD).

Means and standard deviations of the Pearson correlations are reported in Table 8. A 2 group (Control versus PD) × 2 item type (instruments versus locations) ANOVA revealed a significant group by item type interaction, F(1, 66) = 4.53, p = .037, η_p² = .064, indicating that correlation coefficients were higher for location ratings mainly for control participants, whereas PD participants showed little effect of item type. However, there was no significant main effect of group, F(1, 66) = 1.71, p = .196, η_p² = .025. There was a significant main effect of item type F(1, 66) = 5.25, p = .025, η_p² = .074, with locations exhibiting higher mean coefficients.

Table 8.

Pearson correlation coefficients for event-related verb identification by group and item type.

	Mean (SD)
	Combined	Instruments	Locations
Control	0.90 (.02)	0.89 (.03)	0.92 (.02)
PD	0.91 (.03)	0.91 (.03)	0.91 (.03)

Within the PD participants, we assessed whether total upper limb bradykinesia scores (based on the UPDRS-III) or cognitive status significantly correlated with participants’ Pearson coefficients for event-related verbs. There was a significant main effect of cognitive status (F(1, 32) = 6.59, p = .015, η_p² = .17), with lower correlation coefficients (i.e., less similar response patterns) found within PD participants who had MCI. There was no significant main effect of bradykinesia (F(1, 32) = 1.41, p = .244, η_p² = .042) or item type (F(1, 32) = 0.73, p = .788, η_p² = .002). There were also no significant interactions between item type and bradykinesia (F(1, 32) = 0.04, p = .842, η_p² = .001) or between item type and cognitive status (F(1, 32) = 0.02, p = .903, η_p² = .000).

In summary, the significant group by item type interaction suggests that PD may alter sensory-motor aspects of event-related verb processing. However, PD participants’ correlation coefficients were not significantly different than Controls’ overall (with PD participants showing higher correlation coefficients, if anything). This finding suggests that PD participants were generally as capable of recognizing event-related verbs as healthy older adults. Additionally, the significant main effect of cognitive status suggests that PD-related cognitive changes broadly impair the ability to identify event-related verbs.

5. Discussion

5.1. Overall interpretation of results

Previous research suggests that verb processing is impaired in PD, but little work has examined a potential relationship between verb and event knowledge impairments. To better understand whether event-based verb processing is disrupted in PD, we asked participants to produce (Study 1) and identify (Study 2) event-related verbs. We also sought to examine whether any event-related verb processing impairments might be explained by sensory-motor simulation deficits. To do so, we manipulated the implied action content of the experimental stimuli throughout these studies. In both studies, we explored the contributions of motor disease severity and cognitive impairment to any observed deficits.

Based on previous investigations of action and event processing deficits in PD, we predicted 1) that PD would impair event-related verb processing, 2) that greater impairments would be evident for PD participants for tasks and stimuli that require greater sensory-motor simulation, (i.e., processing instrument-verb association) and that 3) greater impairments would be evident in PD participants with cognitive impairment and in those with more severe motor impairments. However, while the results generally supported the third prediction, they only partially supported the first and second predictions.

5.1.1. Impact of PD on general activation of event knowledge

Deficits in event knowledge were expected to lead to impaired verb processing even in tasks that minimised retrieval demands. Yet there were no significant main effects of group on event-based verb production or identification, suggesting that PD does not globally degrade representations of the event-based relationships between verbs and various instruments and locations. Interestingly, in Study 2, participants’ correlation coefficients were quite high across both PD and Control participants, generally ranging from .85 to .95. This finding suggests that event knowledge—at least, knowledge of which actions commonly occur in events with various instruments and locations—is similar across individuals and is stable even in the presence of PD. Thus, previous findings of PD-related impairments in event content production and identification (e.g., Godbout & Doyon, 2000; Roberts & Post, 2018; Wyrobnik et al., 2022) do not appear to be driven by deficits in activating event semantic knowledge at the lexical-semantic level. Importantly, however, the present studies found group by item type interactions, suggesting that event-related verb processing was affected in PD for some item types but not for others.

5.1.2. Impact of PD on sensory-motor simulation during language processing

Group and item type interacted in both studies. In Study 1, PD participants were slower, compared to Control participants, at generating verbs related to instrument items. Crucially, the significant interaction suggests that the differences between PD and Control participants in Study 1 were not merely due to poorer motoric (e.g., typing) abilities, as PD participants showed relatively intact performance on location items. Interestingly, PD participants also produced marginally greater total numbers of responses than Controls in response to locations but not to instruments. One possible interpretation of this result is that it reflects an advantage for processing event-based associates of locations in PD. However, PD participants also spent more time answering the free response surveys than did Controls. Thus, the greater number of responses for location items appears not to reflect a selective processing advantage for locations, but rather a selective processing disadvantage for instruments that was partially masked by PD participants’ greater overall engagement in Study 1. Therefore, these findings may support grounded cognition theories that predict that motor diseases such as PD should selectively impair action-semantic language processing (e.g., Barsalou, 2008; Binder & Desai, 2011). Also supporting this theory is the fact that PD participants’ free responses to location cues contained marginally lower action content than Controls’ responses. Considering that there were no overall differences between PD and Control groups, it is possible that the instrument cues inherently constrained participants to produce higher-action responses. In contrast, the locations appear to have supported more variability in the responses’ action content, as evidenced by higher standard deviations for locations than instruments, leading to this interaction pattern. We interpret this pattern of results with caution given the marginal interaction effect, though it is worth noting that these p values are estimates rather than precise values due to ambiguity in determining the appropriate degrees of freedom and test statistics (Kuznetsova et al., 2017). These findings must also be interpreted cautiously because the Lancaster ratings were collected from published normative data and so were not conducted on age-matched control participants and were not specific to the verb senses used by participants.

Some findings did not show the predicted interactions between group and item type, which suggests a complex relationship between sensory-motor simulation and lexical-semantic processing in PD. Compared to Controls, PD participants showed smaller effects of item type in both the percentage of verbal relevant free responses (Study 1) and in their Likert ratings (Study 2). Additionally, for the percentage of semantically relevant responses, there was neither a main effect of group nor a group by item type interaction, suggesting that PD participants’ semantic response relevancy was similar to that of Controls across both item types. Thus, the collective results did not show that PD participants were consistently impaired at processing high-action event components, but rather that the effect of implied action content on event processing generally differed in people with PD compared to healthy older adults. For example, the degree of action content had less of an impact upon PD participants than on Control participants in Study 2. Thus, these findings may support only ‘weak’ grounded cognition theories that propose that sensory-motor simulation may enrich action-semantic processing in some task contexts, rather than supporting ‘strong’ versions that propose that sensory-motor simulation is critical for action-semantic processing.

The reason for PD participants’ higher mean correlation coefficients for instrument items in Study 2 (compared to Controls) is open to interpretation. Control participants appear to have had a greater tendency to endorse non-standard uses of instruments. In contrast, PD participants appeared relatively insensitive to action content both in Study 2 and in some outcome measures in Study 1, perhaps because of impaired action simulation abilities (e.g., Cardona et al., 2013). Supporting this interpretation are findings that PD participants show reduced functional connectivity in cortical motor networks during action verb processing and increased functional connectivity in posterior brain areas (median cingulate and paracingulate gyri) implicated in amodal lexical-semantic processing (Abrevaya et al., 2017). Thus, our findings may be attributable to the fact that PD participants relied less on sensory-motor simulation than did Controls, particularly for identification-based tasks. Additionally, or alternatively, reduced cognitive flexibility (e.g., Koerts et al., 2009) may have caused PD participants to respond to event-related verbs in relatively stereotyped ways. Along these lines, some PD participants show enhanced lexical-semantic priming effects that appear to be driven by set-shifting deficits, i.e., cognitive inflexibility (McDonald et al., 1996). Both of these possibilities were explored further in the analyses of the role of motor disease severity and cognitive impairment on lexical-semantic processing in PD.

5.1.3. Impact of cognitive impairment

Event processing in PD may change as a function of cognitive impairment. In Study 1, cognitively-impaired PD participants exhibited a lower percentage of verbal relevant responses but similar rates of semantically relevant responses and similar response speeds, compared to age-normal PD participants. In Study 2, cognitively impaired PD participants showed poorer performance in identifying verbs associated with both instruments and locations, with no significant interaction between cognitive status and item type. Overall, these findings support the hypothesis that PD-related cognitive impairment exerts a relatively broad negative effect on event-based lexical-semantic processing but does not seem to fully account for changes in processing high-action stimuli in PD, as discussed in detail in the summary of Study 1 findings.

5.1.4. Impact of motor disease severity

Event processing in PD may also change as a function of motor disease severity. Higher upper limb bradykinesia scores correlated with greater seconds per response and lower percentages of semantically relevant responses, particularly for instrument items. These findings support the idea that motor disease may affect event-based language processing in PD by disrupting the ability to reactivate action simulations during verb processing. However, from an action simulation perspective, it is unclear why participants with higher bradykinesia scores showed less of an effect of item type in the percentage of verbal relevant responses. It may be that sensory-motor simulation is recruited flexibly, rather than being essential for all lexical-semantic action word processing. Considering that people with higher bradykinesia scores showed greater impairment on the percentage of semantically relevant responses for instrument items, it appears that motor disease and action simulation deficits primarily impaired the semantic processing of instrument items, not the ability to generate grammatical verbs. Of note, participants with low bradykinesia scores showed a near-ceiling-level ability to generate semantically valid responses for both instruments and locations. For the percentage of verbal relevant responses, participants with low bradykinesia scores showed a slight advantage for locations, which was consistent with PD participants’ generally higher performance on locations versus instruments.

Overall, bradykinesia severity and cognitive status significantly correlated with multiple measures of event-based lexical-semantic processing. Therefore, given the enormous heterogeneity of clinical manifestations of PD (Foltynie et al., 2002; Greenland et al., 2019), PD participant cohorts’ varying disease characteristics likely account for some of the previously described discrepancies in reports of the nature and extent of verb/action language deficits in PD. It is critical for studies involving PD participants to report measures of participants’ disease severity and cognitive abilities so that we can model these factors and chart the progression of action- and event-based language deficits in PD. Clinically speaking, action- and event-based language processing impairments appear likely to worsen over the course of the disease in tandem with impacts on other cognitive abilities and may hold potential as valuable markers of disease progression.

5.2. Strengths and limitations

The present study features several strengths. First, we examined all participants’ knowledge of event-related verbs in both production and identification conditions, using the same experimental items across conditions. This allowed us to investigate the role of contextual support in the differences between PD and Control participants in event knowledge activation patterns. Second, we compared the production and identification of event-based verbs to the production and identification of other verbs, not to nouns, thereby avoiding a potentially confounding influence of grammatical class. Third, PD participants in the present study were relatively well characterised in terms of their motor and cognitive impairments. The TorCA features multiple sub-tests of executive function and visuospatial deficits (Freedman et al., 2018), making it better suited to the assessment of cognitive function in PD than simpler screening tests such as the MMSE (Bak & Mioshi, 2007).

One limitation is that PD participants’ diagnoses were not able to be confirmed with a physician, so some PD participants may instead have had an atypical parkinsonian disorder (Poewe & Wenning, 2002) or may have been otherwise misdiagnosed. However, at the time of the study, all PD participants had been under the care of a neurologist for over a year (with the exception of one participant who had been diagnosed less than one year prior to participation but had been seen by a neurologist since the time of diagnosis). Another limitation is that the present study featured a relatively small sample size. Although the sample size of the present study was comparable to that of several similar studies that have reported action language deficits in PD (Bertella et al., 2002; Crescentini et al., 2008; Peran et al., 2009; Salmazo-Silva et al., 2017; Signorini & Volpato, 2006), this sample size limited our ability to compare the performances of PD participants with and without cognitive impairment, and with and without more severe motor disease. In particular, the low number of participants with cognitive impairment, coupled with the use of a relatively brief cognitive assessment battery, precluded an investigation of which cognitive skills or domains were most associated with impaired event-based language processing.

The remote nature of this study limited our ability to verify that participants completed all tasks fully independently and in their medication ON state (if applicable). Additionally, the assessment of participants’ motor symptom severity was limited by the fact that this study was conducted remotely, so muscle rigidity was not able to be assessed. Rigidity may sometimes interfere with range of motion (Goetz et al., 2008) and thereby affect sensory-motor simulation abilities, so the lack of rigidity scores may have lessened our ability to detect the influence of motor disease upon sensory-motor simulation during event-based verb processing. Also, because visualizing the lower limbs is challenging in the remote testing mode, we focused on upper limb bradykinesia scores. However, experimental items were selected to prominently feature upper limb movements, making upper limb bradykinesia scores particularly relevant for understanding the impact of motor disease on language processing abilities. The analysis of hand and arm action content in participants’ free responses was limited by the nature of the available normative data and the fact that this was not a matched control sample. Finally, the effect of items’ action content was assessed indirectly, via comparisons between instruments and locations. It may be that differences between instrument and location processing were partially driven by stimulus features other than action content. However, it is important to understand how PD differentially affects the processing of varying kinds of thematic role relations, given that healthy adults immediately use event-based information for thematic role assignment during online, naturalistic language processing (e.g., Altmann, 1999; Bicknell et al., 2010; McRae et al., 1998).

6. Concluding remarks

Overall, the existing literature suggested that PD participants would be impaired relative to healthy older adults on action and event-based language processing. Indeed, in Study 1, PD participants showed slower rates of responses and lower proportions of relevant responses than Control participants for instrument-related verb production. However, PD participants did not consistently perform ‘worse’ overall than Control participants across studies, but merely differently, in that they showed greater verbal productivity in response to locations and a relative insensitivity to action content in other measures of event-based verb processing. Within PD participants, the presence of cognitive impairment correlated with poorer performance on some measures of event-based language processing, and motor disease severity differentially impacted high- and low-action items. Thus, PD may not globally impair event language representations but instead may alter the typical pattern of activation of event language representations, based in part on the degree of implied action content and on the severity of PD participants’ cognitive and motor impairments.

Data Availability Statement

Deidentified ratings data for all experimental stimuli and analysis scripts are publicly available at Open Science Framework: https://osf.io/dbj34/.