c o r t e x 1 4 1 ( 2 0 2 1 ) 1 1 2 e1 2 7 Available online at www.sciencedirect.com ScienceDirect Journal homepage: www.elsevier.com/locate/cortex Research Report Neural signatures of predictive language processing in Parkinson's disease with and without mild cognitive impairment n-Cabrera a, Javier Pagonabarraga b,c,d,e,**, Joaquı́n Morı́s f, Patricia Leo Saül Martı́nez-Horta b,c,d,e, Juan Marı́n-Lahoz b,c,d,e, Andrea Horta-Barba b,c,d,e, Helena Bejr-Kasem b,c,d,e, Jaime Kulisevsky b,c,d,e and Antoni Rodrı́guez-Fornells a,g,* a Cognition and Brain Plasticity Unit (CBPU), Department of Cognition, Development and Educational Psychology, Institute of Neuroscience, University of Barcelona and Bellvitge Biomedical Research Institute (IDIBELL), Barcelona, Spain b Movement Disorders Unit, Neurology Department, Hospital de la Santa Creu i Sant Pau, Barcelona, Spain c Biomedical Research Institute (IIB-Sant Pau), Barcelona, Spain d Universitat Autonoma de Barcelona (UAB), Spain e Centro de Investigacion en Red-Enfermedades Neurodegenerativas (CIBERNED), Spain f Department of Educational and Evolutive Psychology, University of Granada, Granada, Spain g Catalan Institution for Research and Advanced Studies (ICREA), Barcelona, Spain article info abstract Article history: Cognitive deficits are common in Parkinson's disease (PD), with some PD patients meeting Received 8 December 2020 criteria for mild cognitive impairment (MCI). An unaddressed question is whether linguistic Reviewed 07 February 2021 prediction is preserved in PD. This ability is nowadays deemed crucial for achieving fast Revised 4 March 2021 and efficient comprehension, and it may be negatively impacted by cognitive deterioration Accepted 31 March 2021 in PD. To fill this gap of knowledge, we used event-related potentials (ERPs) to evaluate Action editor Michael Schwarze mechanisms of linguistic prediction in a sample of PD patients (on dopamine compensa- Published online 30 April 2021 tion) with and without MCI. To this end, participants read sentence contexts that were predictive or not about a sentence-final word. The final word appeared after one sec, Keywords: matching or mismatching the prediction. The introduction of the interval allowed to Semantic prediction capture neural responses both before and after sentence-final words, reflecting semantic Semantic processing anticipation and semantic processing. PD patients with normal cognition (N ¼ 58) showed Parkinson's disease ERP responses comparable to those of matched controls. Specifically, in predictive con- Mild cognitive impairment texts, a slow negative potential developed prior to sentence-final words, reflecting se- mantic anticipation. Later, expected words elicited reduced N400 responses (compared to unexpected words), indicating facilitated semantic processing. PD patients with MCI (N ¼ 20) showed, in addition, a prolongation of the N400 congruency effect (compared to matched PD patients without MCI), indicating that further cognitive decline impacts * Corresponding author. Cognition and Brain Plasticity Unit Department of Cognition, Development and Educational Psychology Faculty of Psychology, University of Barcelona, L'Hospitalet de Llobregat, Barcelona 08097, Spain. Fax: þ34 934024268. ** Corresponding author. Movement Disorders Unit, Neurology Department, Hospital de la Santa Creu i Sant Pau, Barcelona, 08041, Spain. E-mail addresses: jpagonabarraga@santpau.cat (J. Pagonabarraga), Antoni.rodriguez@icrea.cat (A. Rodrı́guez-Fornells). https://doi.org/10.1016/j.cortex.2021.03.032 0010-9452/© 2021 Published by Elsevier Ltd. c o r t e x 1 4 1 ( 2 0 2 1 ) 1 1 2 e1 2 7 113 semantic processing. Finally, lower verbal fluency scores correlated with prolonged N400 congruency effects and with reduced pre-word differences in all PD patients (N ¼ 78). This relevantly points to a role of deficits in temporal-dependent mechanisms in PD, besides prototypical frontal dysfunction, in altered semantic anticipation and semantic processing during sentence comprehension. © 2021 Published by Elsevier Ltd. Grisoni, Miller, & Pulvermüller, 2017; for a review, see 1. Introduction Pulvermüller & Grisoni, 2020). Importantly, prior research suggests that linguistic predic- Parkinson's disease (PD) is a chronic, neurodegenerative disor- tion is reduced in populations with limited cognitive resources der that, in addition to motor defects, involves difficulties in a (Federmeier, McLennan, De Ochoa, & Kutas, 2002; Federmeier & variety of cognitive domains, including executive functions, Kutas, 2005; Federmeier, Kutas, & Schul, 2010; Payne & visuospatial skills, memory or language (Kudlicka, Clare, & Federmeier, 2018; Wlotko & Federmeier, 2012; for reviews, see; Hindle, 2011; Muslimovic, Post, Speelman, & Schmand, 2005; Payne & Silcox, 2019; Huettig, 2015). For instance, older adults Pell & Monetta, 2008). These deficits tend to progress in severity seem to take less advantage of semantically rich contexts to and meet criteria for mild cognitive impairment (PD-MCI) in up facilitate subsequent processing (Federmeier & Kutas, 2005) and to 40% of patients within the first 5 years of disease (Aarsland those with lower verbal fluency scores exhibit diminished or et al., 2010; Litvan et al., 2011). The development of PD-MCI is absent ERP effects associated with prediction, suggesting the a robust predictor of further conversion to dementia (PDD). adoption of a ‘wait-and-see’, incremental processing strategy However, the pattern of progression of PD-MCI and the rate of instead (Federmeier et al., 2002). As previously stated, PD in- conversion to PDD is highly variable between patients. Among volves significant impairments in cognitive capacities including the several cognitive phenotypes characterizing PD-MCI, executive functions and working memory (WM), which may changes in visuoperceptive skills, language production and secondarily impact language comprehension in these patients comprehension, and semantic and episodic memory have been (Grossman et al., 2002). Thus, we hypothesized that predictive highlighted to better predict conversion to PDD than deficits language processing could be impaired in PD, especially in those mostly ascribed to frontal-executive dysfunction (Horta-Barba individuals with worse cognitive status, as in the case of those et al., 2020; Martinez-Horta & Kulisevsky, 2019; Lang et al., individuals with PD-MCI. In addition, there is behavioral evi- 2019). Yet, the mechanisms underlying deficits in language dence suggesting that PD patients have difficulties using lexical comprehension in PD and PD-MCI have been scarcely studied. and sentential information to selectively activate appropriate An aspect of language comprehension that has not been meanings (Copland, Chenery & Murdoch, 2000a, 2000b, 2001; evaluated in PD is predictive language processing, which plays Copland, Sefe, Ashley, Hudson, & Chenery, 2009). Also, several an important role in achieving fast and efficient language studies have pointed to slower or delayed lexical and semantic processing (Kutas & Federmeier, 2011). Accordingly, readers activation in PD (Angwin, Chenery, Copland, Murdoch, & and listeners probabilistically infer and pre-activate different Silburn, 2005; Arnott, Chenery, Murdoch, & Silburn, 2001), aspects of upcoming words (e.g., van Berkum, Brown, including research focused on N400 effects (Angwin, Zwitzerlood, Kooijman & Hagoort, 2005; Wicha, Bates, Dissanayaka, & Moorcroft et al., 2017; but see,; Friederici, Kotz, Moreno, & Kutas, 2003). Much evidence of language predic- Werheid, Hein, & von Cramon, 2003; Angwin, Dissanayaka, tion has been derived from the N400 event-related potential McMahon, Silburn, Copland, 2017). These deficits in lexical and (ERP) component, an index of semantic processing (Kutas & semantic processing could also hinder the ability to appropri- Hillyard, 1980). The N400, a negativity peaking ~400 msec ately use sentence contexts to guide and activate representa- after word onset, is reduced for words that are more predict- tions ahead of time during sentence comprehension. able in a given context (N400 context effect) (Kutas & Hillyard, To address this, the current study investigated ERP signa- 1984), which is considered to reflect facilitated semantic pro- tures of semantic prediction and semantic processing in PD with cessing owing to the prediction of the word or some of its se- normal cognition (PD-NC) and in PD-MCI, as well as in a group of mantic features (Federmeier & Kutas, 1999). Most relevantly, healthy control participants matched in terms of age, gender recent research has shown that processing differences arise and education. We presented sentence contexts that were pre- even before the word is presented. In particular, prediction dictive or not of a sentence-final word that was delayed by negative potentials (PNP) e sustained, slow cortical potentials one sec (see Table 2), thus allowing to capture the above- consistent with semantic anticipation e precede sentence- mentioned ERP signatures in the anticipatory and processing final words in predictive contexts, eliciting larger amplitudes stages of prediction. In addition, in predictive contexts, half of before words that are strongly expected in a given context (e.g., the final words were semantically incongruent, thus repre- before the word “ball” in “The goalkeeper managed to catch senting a prediction mismatch. the… ball”) (Leo  n-Cabrera, Rodrı́guez-Fornells, & Morı́s, 2017; Following previous studies in healthy adult population Leo n-Cabrera, Flores, Rodrı́guez-Fornells, & Morı́s, 2019; (Leo n-Cabrera et al., 2017, 2019), we hypothesized that 114 c o r t e x 1 4 1 ( 2 0 2 1 ) 1 1 2 e1 2 7 controls would anticipate the final word after predictive con- Kilford, & Lees, 1992). Motor status and stage of illness were texts, as indexed by the development of a PNP in the antici- assessed by the MDS-UPDRS-III and Hoehn & Yahr scales patory stage. In PD-NC and PD-MCI, potential failure to (Goetz et al., 2008). Demographic variables including age, anticipate would be reflected in a comparatively reduced or gender, years of formal education, disease onset, medication absent PNP. In the processing stage, after the presentation of history, as well as the levodopa equivalent daily dosage (LDD) the final word, we expected controls to exhibit canonical N400 (Tomlinson et al., 2010) were collected for all patients (Table 1). context effects, consistent with facilitated processing of All participants were classified as having either normal congruent words as a result of prediction (Kutas & Federmeier, cognition or MCI based on their score in the Parkinson's Dis- 2011). On the other hand, if PD patients adopt a more incre- ease - Cognitive Rating Scale (PD-CRS) (Pagonabarraga et al., mental strategy for comprehension, they might still exhibit 2008). A cut-off score of 83 was used to classify patients as N400 context effects because, even in the absence of predic- PD-MCI (Ferna ndez de Bobadilla et al., 2013). Participants with tion, congruent words are easier to integrate than incongruent dementia were excluded according to the MDS diagnostic words. However, compared to controls, the magnitude of the criteria for PD with dementia (Emre et al., 2007). In PD patients, effect or its onset latency might be delayed, reflecting more cognition was examined during the ‘on’ state, and all partic- difficult or slower semantic processing compared to controls, ipants were on stable doses of dopaminergic drugs during the who would benefit from prediction instead. Lastly, following 4 weeks before inclusion. prior findings in older adults (Federmeier et al., 2002), we In addition, all participants completed standardized se- explored whether the targeted ERP signatures of linguistic mantic and phonological verbal fluency tests (Pen ~ a-Casanova prediction were reduced or diminished in PD patients with et al., 2009). In these tests, they were asked to either name as lower verbal fluency scores. many animals (semantic verbal fluency) or words beginning with the letter P (phonological verbal fluency) as possible in 1 min. For each participant, total verbal fluency scores were 2. Materials and methods obtained by averaging the direct scores in the semantic and phonological fluency tests (for another study using total ver- No part of the study procedures or analyses was pre-registered bal fluency scores, see Federmeier et al., 2002). prior to the research being conducted. In the following sec- Legal copyright restrictions prevent public archiving of the tions, we report how we determined our sample size, all data neuropsychological tests used, which can be obtained from exclusions, all inclusion/exclusion criteria, whether inclusion/ the copyright holders in the cited references. exclusion criteria were established prior to data analysis, all manipulations, and all measures in the study. 2.3. Final samples 2.1. Participants A total of 135 participants completed the baseline EEG recording session. Of those participants, 102 were PD patients Participants were prospectively recruited from a sample of and 33 were healthy controls. The data of some participants outpatients regularly attending the Movement Disorders were excluded because the files were corrupt (n ¼ 3) or their Clinic at Hospital de la Santa Creu i Sant Pau (Barcelona, recordings were excessively noisy (n ¼ 4) as determined by Spain). They were invited to participate in a longitudinal study visual inspection of the raw data, leaving 128 participants involving comprehensive neuropsychological testing and two available for analyses. From the remaining participants, 97 EEG recording sessions (one at baseline and one in a follow-up were PD patients and 31 were controls. Controls were excluded session after 1 year). Healthy adults were also invited to if they had mild cognitive impairment (n ¼ 3) or their cognitive participate in the study to serve as controls. The current study status was not available (n ¼ 2). Lastly, participants with less focused on the data collected in the baseline session. The than 20 available trials in any condition after EEG pre-pro- project was approved by the research ethics committee of the cessing (n ¼ 9) were also excluded from the analyses (see EEG Hospital de la Santa Creu i Sant Pau. Before inclusion, written pre-processing section for more information about the artefact informed consent was obtained from all the participants. detection and rejection procedure). The conditions of our ethics approval do not permit public The final pool of data consisted of 88 PD patients and 26 archiving of anonymized study data. Readers seeking access to controls, from which matched samples were subsequently the data should contact the corresponding authors. Access can obtained to perform the following comparisons according to be granted only to qualified researchers under a collaboration the goals of the present study: 1) PD patients with normal agreement with the corresponding authors and in accordance cognition (PD-NC) (N ¼ 58) with an age-, gender- and with ethical procedures governing the reuse of clinical data, education-matched Control group (N ¼ 24), and 2) a subsam- including the completion of a formal data sharing agreement ple of Parkinson's disease patients with normal cognition and approval of the local ethics committee. (sPD-NC) (N ¼ 19) with an age-, gender-, education- and years of disease evolution-matched PD patients with mild cognitive 2.2. Clinical and cognitive testing impairment (PD-MCI) (N ¼ 20). To avoid the comparison of patients classified as sPD-NC and PD-MCI but having similar PD patients had been diagnosed by a neurologist with exper- PD-CRS total score, we selected participants with PD-CRS tise in movement disorders. All patients accomplished steps 1 scores of <75 for the PD-MCI group and of >85 for the sPD-NC. and 2 of the UKPDSBB criteria, and three or more of the four The demographic and clinical features of the final samples first supportive positive criteria of step 3 (Hughes, Daniel, are reported in Table 1. Controls and PD-NC patients did not c o r t e x 1 4 1 ( 2 0 2 1 ) 1 1 2 e1 2 7 115 Table 1 e Demographics and clinical features of the patient and control samples. Control PD-NC P sPD-NC PD-MCI P N 24 58 19 20 Age (years) 62.85 (6.82) 64.47 (7.61) .37 70.75 (2.70) 72.22 (4.48) .22 [59.9, 65.7] [62.4, 66.4] [69.4, 72] [70.1, 74.3] Education (years) 14.5 (4.04) 13.2 (4.57) .23 12.42 (5.08) 10.4 (5.16) .22 [12.7, 16.2] [12, 14.4] [9.9, 14.8] [7.9, 12.8] Women (%) 41.6 31 .35 42.1 45 .85 Disease duration (years) n/a 5.26 (3.00) e 5.68 (3.06) 6.28 (3.8) .59 [4.4, 6] [4.2, 7.1] [4.4, 8] Levodopa dose (mg/d) n/a 265.25 (249.58) e 252.87 (293.74) 384.4 (215.04) .12 LEDD (mg/d) n/a 456.04 (230.68) e 397.18 (236.08) 532.83 (209.70) .09 PD-CRS total score 106.16 (8.74) 99.72 (10.86) .01 98.2 (9.17) 67.15 (5.39) <.001 [ 83 cutoff] [102.4, 109.85] [96.8, 102.5] [93.7, 102.6] [64.6, 69.6] Total verbal fluency score (DS) 19.66 (3.12) 18.01 (4.18) .08 17.63 (4.32) 10.52 (2.08) <.001 The values correspond to means, standard deviations (in round brackets) and 95% mean confidence intervals (CI) (in square brackets). For simplicity, CIs are shown only for the relevant variables in the group comparisons. p-values were determined with independent samples t-tests between groups for continuous and ordinal variables and with Pearson Chi-Square for nominal variables (Gender). n/a ¼ not applicable; LEDD ¼ Total levodopa equivalent daily dosage; PD-CRS ¼ Parkinson's Disease Cognitive Rating Scale; DS ¼ Direct score. differ in years of age (t(1,80) ¼ .90, p ¼ .37, 95% CI [e5.19, sentences ended with a congruent word and half with a 1.95]), years of education (t(1,80) ¼ 1.20, p ¼ .23, 95% CI [e.84, semantically incongruent word. For the sake of task brevity to 3.43]), gender (X2(1, 82) ¼ .853, p ¼ .35), or verbal fluency scores avoid fatigue, we decided to include the congruency contrast (t(1,80) ¼ 1.73, p ¼ .08, 95% CI [e.23, 3.53]). They differed in PD- (congruent vs incongruent contrast) only in the HC condition. CRS total score (t(1,80) ¼ 2.57, p ¼ .01, 95% CI [e11.41, 1.46]), Sentences were further divided in two lists of 156 sentences, but both groups were above the established score cut-off of 83 each list containing half of the sentences of each condition, for PD-MCI diagnosis. Likewise, sPD-NC and PD-MCI patients and the use of one or another list was pseudo-randomized did not differ in years of age (t(1,37) ¼ 1.23, p ¼ .22, 95% CI across participants. In sum, each participant read the following [e3.89, .94]), years of education (t(1,37) ¼ 1.30, p ¼ .22, 95% CI sentences: 52 HC with congruent endings (HC congruent), 52 [e1.30, 5.34]), gender (X2(1, 39) ¼ .033, p ¼ .855), disease dura- HC with incongruent endings (HC incongruent), and 52 LC with tion (t(1,37) ¼ .53, p ¼ .59, 95% CI [e2.84, 1.65]) or levodopa congruent endings (LC congruent) (Table 2). All final words equivalent daily dose (t(1,37) ¼ 1.74, p ¼ .09, 95% CI [e293.94, were nouns. Congruent ones were the best completions e the 22.64]), but differed in PD-CRS total score (t(1,37) ¼ 12.96, word with the highest cloze-probability (i.e., percentage of in- p < .001, 95% CI [26.2, 35.91]) and total verbal fluency scores dividuals that supply that word as a continuation for that (t(1,37) ¼ 6.59, p < .001, 95% CI [4.92, 9.29]). For all comparisons, sentence) (Taylor, 1953) e and incongruent words were equal variances were assumed based on null results in Lev- selected from the ESPAL database (Duchon, Perea, Sebastia  n- ene's tests. s, Martı́, & Carreiras, 2013) so that they matched the Galle congruent endings in the variables of mean word length, mean 2.4. Materials number of syllables, word frequency, familiarity, imaginability and concreteness. The stimuli and task presentation code can We used the materials from Leo  n-Cabrera et al. (2019) con- be downloaded at https://osf.io/3FSP9/. sisting of high (HC) or low (LC) constraining sentence contexts, establishing a stronger or weaker expectation for an upcoming 2.5. Procedure word (see Table 2 for sentence examples). A total of 312 sen- tences were included, of which 208 had HC sentence contexts Participants were tested individually. After the external elec- (66.6%) and 104 had LC sentence contexts (33.3%). The sen- trodes and the EEG cap were placed, electrode impedances tences were originally created and categorized as either HC were checked. Participants were instructed to keep their gaze (mean cloze probability ¼ 76%, SD ± 17.7%) or LC (mean cloze fixated on the center of the screen and to blink only during probability ¼ 6.1%, SD ± 10.3%) by Mestres-Misse , Rodriguez- the blinking interval. After completing a short training Fornells, & Münte, 2007. Within the HC condition, half of the block (4 sentences), the proper experiment started. They were Table 2 e Example sentences of each condition (original sentences in Spanish in brackets). Condition Sentence context Final word HC congruent The goalkeeper managed to catch the (El portero fue capaz de atrapar la) ball (pelota) HC incongruent The goalkeeper managed to catch the (El portero fue capaz de atrapar la) shore (orilla) LC congruent As a present she gave her son a (Le ha regalado a su hijo una) ball (pelota) 116 c o r t e x 1 4 1 ( 2 0 2 1 ) 1 1 2 e1 2 7 instructed to read the words that would appear at the center of Maris & Schoffelen, 2011), and custom scripts programmed in the screen attentively and silently. MATLAB R2017b. For each participant and condition, the The experiment consisted of 26 blocks of 6 trials each (2 HC continuous data were segmented into the following epochs of congruent, 2 HC incongruent, 2 LC congruent). The order of the interest: 1) from 100 to 1300 msec from the onset of the sentences within each block was randomized. Words were penultimate word (pre-word interval), and 2) from 100 to white on black background (font type: Arial; font size: 24 1000 msec from the target word onset (word interval). Of note, in points). On each trial, a fixation point (a cross) appeared at the a previous study with healthy adult population (Leo  n-Cabrera center of the screen (800 msec). Then, a sentence was pre- et al., 2019), we used a larger pre-word interval that also sented word-by-word (300 msec per word, followed by a included the sentence context (from the onset of the sentence 200 msec blank screen). Between the penultimate and the final until the offset of the delay), which importantly revealed that word, the blank screen remained for a period of 1 sec (pre-word differences accrued early and gradually during sentence pro- interval). After the final word (300 msec duration), a blank cessing. However, we discarded this possibility in the current screen was presented (1000 msec) before the blinking signal study because the EEG signal in the sentence context intervals (1 sec; depiction of an eye at the center of the screen). The next contained many artifacts, as preliminarily assessed through trial began after the offset of the blinking signal (see Fig. 1 for a visual inspection. It was not unlikely that participants, espe- simplified display of the trial structure). cially PD patients and older adults, had greater difficulties in To ensure that participants were reading attentively, they refraining from blinking or from moving for a relatively long were asked to judge whether the previous sentence was interval (4.3 sec). We therefore opted to restrict the analysis of congruent or not (“did the previous sentence make sense?”) the anticipatory stage to the 1 sec delay interval following pre- every two blocks. The question remained on screen until they vious reports (Leo  n-Cabrera et al., 2017, 2019). answered ('yes' or 'no') using the keyboard. The next block Artefact detection and rejection were applied on the started immediately after the test except for even-numbered epoched data. To facilitate artefact rejection, we computed the blocks, which were followed by a resting period that could vertical (vEOG) and horizontal (hEOG) bipolar EOG channels by be resumed anytime by the participant. All groups performed subtracting the inferior from the superior vertical EOG chan- above chance in the congruency probe task (see Supplemen- nels and the right from the left horizontal EOG channels, tary Materials for the accuracy of each group). respectively. For the pre-word interval, all epochs with activity ± 85 mV in the ocular channels or ± 200 mV in any other channel 2.6. EEG recording were automatically removed and the remaining epochs were visually inspected and excluded if they contained blinks, 2.6.1. EEG acquisition muscle activity or large drifts. For the word interval, the Continuous electroencephalogram (EEG; sampling concatenated epochs were first subjected to independent rate ¼ 250 Hz) was recorded from 19 electrodes (FP1/2, F3/4, F7/ component analysis (ICA) decomposition to correct blinking 8, Fz, C3/4, Cz, P3/4, Pz, T3/4, T5/6, O1/2) mounted on a cap artifacts because preliminary visual inspection revealed that following the international 10e20 system positions. The EEG many participants had difficulties refraining from blinking signal was amplified on-line (band-pass filter ¼ .016e1000 Hz) until the blinking signal, leading to a large loss of trials in this with a Brain Vision amplifier. To monitor ocular artefacts, interval. ICA was run using the “runica” algorithm as imple- electrooculogram (EOG) electrodes were recorded with elec- mented in EEGlab toolbox v14.0.0, excluding the ocular chan- trodes at bipolar vertical (supraorbital and suborbital ridge of nels from the decomposition. The ocular components were the right eye) and horizontal placements (external ocular detected and rejected based on visual inspection. After canthus of the left and right eyes). All electrode impedances artifact-correction, epochs with activity ± 200 mV in any but were kept below 5 kU. An on-line notch filter (50 Hz) was applied ocular channels were automatically rejected, and those that to attenuate high-frequency electrical noise and data were remained were visually screened for any previously unnoticed referenced to the mean activity of the left and right mastoid artefacts (e.g., muscle activity or large drifts). electrodes. For the Control with PD-NC comparison, an average of 89.8% trials per condition were available for analysis (HC ¼ 85.7 2.6.2. EEG preprocessing trials; LC ¼ 41.9 trials; HC congruent ¼ 49.6 trials; HC EEG data were preprocessed using ERPlab v7.0.0 (Lo  pez- incongruent ¼ 49.6 trials; LC congruent ¼ 49.7 trials; overall Calderon & Luck, 2014) of the EEGlab toolbox v14.0.0 (Delorme average ¼ 46 trials per condition), with no significant differ- & Makeig, 2004), FieldTrip version 20181231 (Oostenweld, Fries, ences in the percentage of available trials between groups (all Fig. 1 e Depiction of the structure of a trial. Sentences were presented one word a time. Each word of the sentence context (w1 to w7) was displayed for 300 msec followed by a 200 msec inter-word interval, except for the last word (w7), which was followed by a 1000 msec interval (pre-word interval) leading up to the presentation of the final word (w8). c o r t e x 1 4 1 ( 2 0 2 1 ) 1 1 2 e1 2 7 117 p-values > .3). For the sPD-NC with PD-MCI comparison, an congruent, as well as for LC congruent relative to HC congruent average of 87.4% trials per condition were included (HC ¼ 79.2 (Kutas & Federmeier, 2011). In turn, we chose two-tailed testing trials; LC ¼ 39.2 trials; HC congruent ¼ 49.3 trials; HC for the between-subject comparisons, as we did not have a priori incongruent ¼ 49.5 trials; LC congruent ¼ 49.6 trials; overall hypotheses of the direction of the effects in this case. average ¼ 45 trials per condition), with no significant differ- ences in the percentage of remaining trials between groups (all 2.7.3. Correlations between ERP effects and verbal fluency p-values > .1). After examining ERP effects, we conducted non-parametrical SpearmaneBrown correlations to evaluate a potential linear 2.7. Statistical analyses relationship between the ERP effects and verbal fluency performance. All PD patients were included in this analysis 2.7.1. Event-related potential analyses (N ¼ 78), that is, PD patients with normal cognition (N ¼ 58) as Before performing statistical analyses, data were baseline- well as PD patients with MCI (N ¼ 20). For each participant, corrected to the 100 msec pre-stimulus period and low-pass the total verbal fluency score was obtained by averaging the filtered (as implemented in ERPlab v7.0.0). For the pre-word direct scores in the semantic and phonological fluency sub- interval, a 5 Hz low-pass Butterworth filter (12 dB/oct roll- tests (see Supplementary Materials for correlations with each off) was applied to the data to focus exclusively on slow ac- subtest individually). ERP measures were labelled and tivity (Brunia, Hackley, van Boxtel, Kotani, & Ohgami, 2011). quantified as follows: 1) PNP (LC minus HC mean amplitude For the word interval, in which the N400 component was difference in the 600 msec before pre-word onset at a left targeted, a 30 Hz low-pass Butterworth filter (12 dB/oct roll-off) anterior cluster including electrodes FP1, F3, Fz, F7); 2) N400 was used. For the figures presented, the data were filtered congruency effect (HC congruent minus HC incongruent mean using a 12 Hz low-pass Butterworth filter (12 dB/oct roll-off). amplitude difference from 300 to 500 msec post-word onset); 3) N400 constraint effect (LC congruent minus HC congruent 2.7.2. Cluster-based permutation analyses mean amplitude difference from 300 to 500 msec post-word Non-parametrical cluster-based permutation tests were used to onset); and, based on the ERP results, 4) prolonged N400 con- assess the differences (Maris & Oostenveld, 2007) (as imple- gruency effect (HC congruent minus HC incongruent mean mented in Fieldtrip version 20181231 under MATLAB R2017b), a amplitude difference from 600 to 800 msec post-word onset). method that exerts proper control for the increased probability of All effects tied to the N400 were tested on a centrally- false positives in the context of many comparisons. Broadly, it distributed cluster (averaged electrodes: Cz, C3, C4, Pz, P3, identifies adjacent time points and channels with similar dif- P4). All values were normalized before performing correla- ferences between conditions. The test worked as follows in the tions. All p-values were corrected with the Holm-Bonferroni present study. For within-subject comparisons, every sample correction (Holm, 1979). (channel  time) was compared between two conditions (e.g., HC vs LC) through a dependent t-test. For between-subject compar- isons, an independent t-test compared the effect (e.g., HC minus 3. Results LC) between groups at each sample (channel  time). Next, adjacent samples were clustered based on a t-value threshold 3.1. Event-related potentials (pre-determined from a one-tailed t-distribution with an alpha level of .05 and N-1 degrees of freedom) and the cluster with the 3.1.1. Controls versus PD patients with normal cognition largest sum of t-values was selected (cluster-level t-value). To Fig. 2A and B shows the grand-average ERPs at the F7 electrode determine whether effects were significant, the Monte Carlo during the pre-word interval in the Control group and the method was used to construct a null distribution from the group of PD patients with normal cognition, respectively. The cluster-level t-values of random partitions obtained by randomly ERPs were computed in the interval from the onset of the swapping the samples (between conditions and within partici- penultimate word to the onset of the final word (1300 msec pants for within-subject comparisons; and between conditions duration, the first 300 msec correspond to the penultimate and between groups for between-subject comparisons) (5000 word, and the last 1000 msec to the delay interval), with a randomizations). Only observed clusters with cluster-level t- 100 msec pre-stimulus baseline. values within the 2.5th percentiles (alpha level of .05) of the null Both groups exhibited similar patterns, namely, an N100 distribution were considered significant. associated with the processing of the penultimate word, fol- Choosing one-tailed testing is an optimal methodological lowed by a positivity, and the later development of a negativity decision when there are a priori hypotheses about the direction that became progressively larger as the presentation of the of the differences (Lakens, 2016). It allows to improve statistical final word approached. As expected, based on previous find- sensitivity, which is particularly important when working with ings in healthy adult population (Leo  n-Cabrera et al., 2019), the clinical samples. Following this, we chose one-tailed testing for negativity was more prominent in the HC condition (relative to the within-subject comparisons, as we had a priori hypotheses the LC condition), that is, when the final word could be strongly both for the pre- and post-word interval. In the pre-word in- predicted from the prior context. This difference was terval, we expected more negative amplitudes for HC than LC confirmed by a significant cluster in the Control group (p ¼ .01) (Leo n-Cabrera et al., 2017, 2019) In the post-word interval, in (Fig. 2A) as well as in PD individuals with normal cognition which N400 context effects were targeted, we expected more (p ¼ .01) (Fig. 2B). The cluster showed a similar temporal profile negative amplitudes (larger N400 responses) for unexpected in both groups. It started approximately 800 msec before the than for expected words, that is, for HC incongruent than for HC final word and continued until the word appeared. In regard to 118 c o r t e x 1 4 1 ( 2 0 2 1 ) 1 1 2 e1 2 7 Fig. 2 e ERP results for the Control and the PD-NC group. A and B) Grand-averaged ERPs at the F7 electrode during the pre- word interval for both conditions for (A) Controls and (B) PD-NC patients. ERPs are time-locked to the presentation of the penultimate word (100 pre-stimulus baseline). The horizontal bar highlights (grey) the time interval in which significant differences were obtained. Below the ERPs, topographical maps indicating the electrodes (white asterisks) that showed a significant effect of constraint (HC minus LC) in the 600 msec prior to the critical word (in steps of 200 msec). C and D) Grand- averaged ERPs at Cz electrode after the presentation of the sentence-final word (100 pre-stimulus baseline) for (C) Controls and (D) PD-NC patients. The horizontal bars highlight the time interval of the N400 effects of Congruency (green) and Constraint (purple). Topographical maps with electrodes (white asterisks) that showed effects in the time interval in which N400 effects were expected to be maximal (300e500 msec post-word onset). the spatial distribution, the cluster encompassed mainly The effect of constraint (HC congruent vs LC congruent) frontal electrodes, with a slight left-lateralization. There were was captured by significant clusters in the Control group no significant differences between groups. (p ¼ .001) (Fig. 2C) and in the PD group (p < .001) (Fig. 2D), We then turned to the interval after word presentation. As whereby words presented in less predictive contexts (LC can be observed in Fig. 2C and D, for both groups, semantically congruent) showed more negative amplitudes than words in unexpected words (HC incongruent and LC congruent) elicited strongly predictive contexts (HC congruent). There were no a larger negativity peaking about 400e600 msec (relative to differences between groups. The cluster started about expected words; HC congruent) that is consistent with the 250e300 msec after word onset and resolved at about canonical features of the N400. These differences were re- 700 msec, and it exhibited a widespread distribution over the flected in the significant clusters described below. scalp. In fact, the cluster was significant at all sites. On the c o r t e x 1 4 1 ( 2 0 2 1 ) 1 1 2 e1 2 7 119 other hand, the effect of congruency (HC congruent vs HC 3.1.2. Subset of PD patients with normal cognition versus PD incongruent) was reflected in a shorter-lived cluster in the with MCI Control group (p ¼ .007) (Fig. 2C) and in the PD group (p < .001) The mean voltage pattern elicited during the pre-word interval (Fig. 2D), in this case confirming more negative amplitudes for (Fig. 3A and Fig. 3B) exhibited the normative N100 component, incongruent than congruent words in strongly predictive followed by a transient positivity and a subsequent negativity contexts. Again, there were no significant differences be- in the last part of the interval that evidenced the hypothesized tween groups. The effect was widespread with a seemingly tendency towards more negative amplitudes for HC than LC. slight rightward focus in magnitude. However, in this case, there were no statistically significant Fig. 3 e ERP results for the subgroup of PD patients wit normal cognition (PD-NC) and the group of PD-MCI patients. A and B) Grand-averaged ERPs at the F7 electrode during the pre-word interval for (A) sPD-NC and (B) PD-MCI patients. ERPs are time- locked to the presentation of the penultimate word (100 pre-stimulus baseline). The grey horizontal bar highlights the time interval in which significant differences were obtained. Below the ERPs, the topographical maps indicate the electrodes (white asterisks) that showed a significant effect of constraint (HC minus LC) in the 600 msec prior to the critical word (in steps of 200 msec). C and D) Grand-averaged ERPs at the Cz electrode after the presentation of the sentence-final word (100 pre-stimulus baseline) for (C) sPD-NC and (D) PD-MCI patients. The horizontal bars highlight the time interval of the N400 effect of Congruency (green) and Constraint (purple). The topographical maps show which electrodes (white asterisks) showed effects in the time interval in which N400 effects were expected to be maximal (300e500 msec post-word onset). 120 c o r t e x 1 4 1 ( 2 0 2 1 ) 1 1 2 e1 2 7 differences between conditions, nor any differences between predictive contexts elicited a larger N400 than words in the two groups. strongly predictive contexts, a difference that encompassed After the presentation of the final word (Fig. 3C and D) there the interval between about 300 and 700 msec post word onset. was an observable larger negativity for unexpected (HC The scalp distribution of the effect covered a wide set of incongruent and LC congruent) relative to expected words (HC electrodes over frontal and central regions of the scalp, with a congruent) in both groups, in line with classical N400 context slight frontward amplitude maximum. No differences be- effects. These observable differences were confirmed by sig- tween groups were found. nificant clusters as specified next. As for the effect of congruency in predictive contexts (HC There were significant effects of constraint (HC congruent congruent vs HC incongruent), again as expected, significant vs LC congruent) both for the group of PD patients without MCI clusters reflected that incongruent words had more negative (p < .001) (Fig. 3C) and the group of PD patients with MCI amplitudes than congruent words in patients without MCI (p < .001) (Fig. 3D). That is, as expected, words in less (p ¼ .005) (Fig. 3C), and also in PD patients with MCI (p ¼ .001) Fig. 4 e A) Topographical maps showing the temporal evolution and scalp distribution of the congruency contrast (HC incongruent minus HC congruent) during final word processing, showing a prolonged N400 congruency effect in the group of individuals with PD and MCI. B) Grand-averaged ERPs of the difference waveforms (HC incongruent and HC congruent) of both groups, averaged over the set of electrodes that were part of the significant cluster of differences between the groups (highlighted in white in the topographical map). The grey-colored area shows the timepoints included in the cluster. The topographical map shows the scalp distribution of the averaged t-values within the cluster (cluster t-value ± 2.02 for an alpha level of .05). c o r t e x 1 4 1 ( 2 0 2 1 ) 1 1 2 e1 2 7 121 (Fig. 3D). Relevantly, in this case, there were significant dif- robust to the presence of outliers. This indicates that a larger PNP ferences between the groups (Fig. 4). The difference waveforms (more negative amplitudes prior to words in predictive than in (HC incongruent minus HC congruent) of each group were unpredictive contexts) is associated with better verbal fluency contrasted in a between-group cluster-based permutation test performance in patients with PD. In turn, the prolongation of the including all electrodes and time points (from 0 to 1000 msec). N400 congruency effect (more negative amplitudes for incon- The test yielded a significant cluster (p ¼ .008) that revealed a gruent than for congruent words 600e800 msec after word longer-lasting effect of congruency for the group of patients onset) correlates with worse mean verbal fluency scores with MCI, which can be visualized in Fig. 4A. More specifically, (r(78) ¼ .358, pHB ¼ .005). No other correlations were statisti- the cluster spanned approximately from 630 to 760 msec after cally significant. word onset at centro-parietal sites (significant electrodes: C3, Cz, C4, P3, Pz, P4, T5, T4, T6, O2) (see the scalp distribution of t- values in Fig. 4B). This result suggests a prolonged N400 con- 4. Discussion gruency effect in the group of PD patients with MCI relative to the group of PD patients without MCI. Motivated by previous evidence of altered predictive language processing in populations with low cognitive resources 3.2. Correlations between ERP effects and verbal fluency (Federmeier et al., 2002, 2005, 2010), the present study scores assessed linguistic prediction in PD with normal cognition and PD with MCI. To this end, ERP modulations were examined After establishing the ERP effects, we examined the association before and after encountering words that could or could not be between ERP measures and verbal fluency scores in all PD pa- predicted from context, thus capturing both the postulated tients (N ¼ 78) (for further specification of this analysis, see anticipatory and processing stage of prediction. Specifically, Section 2.7.3.) by means of non-parametric Spearman correla- we focused on PNPs as signatures of semantic anticipation, tions between direct scores in verbal fluency tests, and four ERP and post-word N400 context effects as indices of processing measures, namely, 1) PNP (LC minus HC), 2) N400 constraint facilitation by pre-activation. Relative to controls, PD patients effect (LC congruent minus HC congruent), 3) N400 congruency with normal cognitive status exhibited the expected ERP sig- effect (HC congruent minus HC incongruent), and 4) prolonged natures, pointing to preserved predictive language processing. N400 congruency effect (HC congruent minus HC incongruent). In turn, PD patients with MCI showed absent PNPs and a sig- Note that, in these analyses, the direction of the condition nificant prolongation of N400 congruency effects in predictive subtraction is reversed, such that positive values represent contexts, suggesting altered mechanisms tied to linguistic differences in the expected direction of the effect (e.g., in the prediction. Interestingly, correlational analyses revealed that case of the PNP, positive values indicate larger amplitudes for worse verbal fluency performance was associated with the HC than for LC). All correlations were corrected using Holm- presence of the ERP pattern observed in the group of PD pa- Bonferroni correction (indicated as pHB in the text). tients with MCI. The results of the correlational analyses are presented in Table 3. After correcting for multiple comparisons, two of the 4.1. Neural correlates of predictive sentence processing ERP measures showed a significant correlation with verbal are preserved in PD-NC fluency scores: the PNP and the prolonged N400 congruency ef- fect (Fig. 5). In particular, the PNP showed a significant positive Consistent with prediction, PD patients with normal cognition correlation with verbal fluency (r(77) ¼ .326, pHB ¼ .010). Note exhibited the canonical reduction of N400 amplitude for ex- that we excluded one participant who had outlying scores in the pected (HC congruent) compared to unexpected words (LC PNP (z-score ¼ 3.91), although the correlation was unaffected congruent and HC incongruent) (Kutas & Hillyard, 1984). This by its inclusion (r(78) ¼ .326, pHB ¼ .01), which is normal when effect is taken to reflect facilitated lexical and semantic acti- using non-parametric Spearman correlations which are very vation of expected words as a result of their pre-activation (Federmeier & Kutas, 1999; Lau, Phillips, & Poeppel, 2008). In line with this interpretation, in predictive contexts (HC), PD Table 3 e Results of correlational analyses between ERP patients with normal cognition exhibited a frontally-distrib- measures and verbal fluency scores in PD patients. uted negative PNP that developed progressively prior to the presentation of the final word (Leo  n-Cabrera et al., 2019; for Test ERP effect r p-value (corrected) similar results, see, Grisoni, Miller & Pullvermüller, 2017; Li, Zhang, Xia, & Swaab, 2017). Studies in healthy population Verbal fluency PNP .326 .010* suggest that such PNPs might capture anticipation of semantic N400 constraint .157 .336 N400 congruency .073 .525 aspects of the upcoming word (for a recent review, see Prolonged N400 congruency e.358 .005* Pulvermüller & Grisoni, 2020). More specifically, they might reflect retrieval or maintenance of pre-activated lexical and/or PNP effect (LC minus LC); N400 constraint effect (LC congruent minus HC congruent); N400 congruency effect (HC congruent semantic representations (Leo  n-Cabrera et al., 2019; Li et al., minus HC incongruent); Prolonged N400 congruency effect (HC 2017). Altogether, PD patients with normal cognition seem to congruent minus HC incongruent). make proper use of sentence contexts to predict upcoming Holm-Bonferroni correction was applied to adjust p-values. Sta- linguistic information during comprehension. tistically significant correlations are highlighted with asterisks. More broadly, the present findings are in line with previous studies showing intact N400 effects in sentence processing in 122 c o r t e x 1 4 1 ( 2 0 2 1 ) 1 1 2 e1 2 7 Fig. 5 e Scatterplots and scalp maps of the significant correlations between ERP measures and verbal fluency scores in all PD patients (PD with and without MCI) (N ¼ 78). A) Scatterplot of the positive correlation (r ¼ .326, pHB ¼ .010) between mean verbal fluency scores and the PNP (more negative amplitudes prior to words in HC than in LC contexts) and B) of the negative correlation (r ¼ ¡.358, pHB ¼ .005) between verbal fluency scores and the prolongation of the N400 congruency effect (more negative amplitudes for incongruent than for congruent words 600e800 msec post-word). All values were normalized. Below the scatter diagrams, the r-values (Spearman rho) for all 19 electrodes are plotted on an idealized head. Darker shading indicates larger r-values in the direction of the correlation. Electrodes included in the ROI used to perform the reported correlations with verbal fluency (shown in the scatter diagrams) are highlighted in white. R-values between electrodes were estimated by spherical spline interpolation. PD patients with normal cognition (Angwin, Dissanayaka, & has also been linked with anticipatory processes in PD. Spe- McMahon, et al., 2017; Friederici et al., 2003), and, in turn, cifically, PD patients with normal cognition off medication contrasts with behavioral and electrophysiological evidence show reduced amplitudes of the stimulus-preceding negativity, pointing to altered lexical and semantic processing in these suggesting that the DA system is implicated in the anticipation patients (Angwin, Chenery, Copland & Arnott et al., 2004, of motivationally salient and rewarding stimuli (Mattox Valle- Angwin, Chenery, Copland & Murdoch et al., 2004, 2006, 2009, Incla n & Hackley, 2006). Future studies testing patients off Angwin, Dissanayaka, & Moorcroft et al., 2017; Arnott et al, medication would help untangle to what extent dopaminergic 2001, 2010; Copland et al., 2009; Kutas et al., 2013). Of note, compensation contributed to the preservation of otherwise most studies showing abnormal lexical and semantic pro- altered mechanisms. cessing employed tasks that consisted of single-word contexts, instead of sentence contexts. Importantly, sentence process- 4.2. PD-MCI impacts late semantic processing in ing involves the construction of a high-level meaning repre- predictive contexts sentation (Graesser, Singer, & Trabasso, 1994), which serves to constrain in a top-down manner towards the appropriate To further investigate the impact of cognitive impairment, we lexical and semantic representation, perhaps compensating also examined neural responses in a smaller sample of PD-MCI for activation problems within semantic networks in these patients. Compared to a matched subgroup of PD patients with individuals. Nonetheless, note that all patients in the current normal cognition, PD patients with MCI also exhibited ex- study were on dopaminergic medication. PD patients on pected N400 amplitude modulations within the normal onset dopaminergic medication perform better in semantic priming latency (i.e., about 300e400 msec post-word onset). However, tasks than PD patients off medication (with normal cognition) most remarkably, PD patients with MCI showed a significantly (Angwin, Chenery, Copland & Arnott et al., 2004, 2009), sug- prolonged N400 congruency effect in predictive contexts, gesting that dopamine (DA) mediates semantic activation, extending up to 700 msec after word onset. Critically, this was arguably through the calibration of the spread and focus of not observed in the N400 constraint contrast (LC congruent activation within semantic networks (Kischka et al., 1996). DA vs HC congruent), which suggests that the prolongation c o r t e x 1 4 1 ( 2 0 2 1 ) 1 1 2 e1 2 7 123 stemmed from the ERP response to incongruent words, rather dysfunction is common with the progression of fronto-striatal than that to congruent words (i.e., from HC incongruent, deterioration in PD with and without MCI (Aarsland et al., instead of HC congruent). With that in mind, it is important to 2010; Kudlicka et al., 2011; Monchi et al., 2004) and has been note that N400 context effects do not solely capture prediction proposed to indirectly hinder language processing in PD effects, but also integration demands (Van Berkum, Hagoort, & (Grossman et al., 2002, 2003). However, verbal fluency perfor- Brown, 1999). Interestingly, a recent study showed that pre- mance depends not only on frontal lobe function, but also on diction effects dominate the earlier portion of the N400 lexical and semantic retrieval processes that are dependent on context effect (starting as early as 200 post-word onset), temporo-parietal structures (Unsworth, Spillers, & Brewer, whereas integration effects began later and continued until 2011). In fact, recent findings point to a greater weight of about 650 msec after word onset (Nieuwland et al., 2020; see language-related processes than executive function in verbal also,; Brothers, Swaab, & Traxler, 2015; Lau, Namyst, Fogel, & fluency (Whiteside et al., 2016). Recent meta-analyses and Delgado, 2016). As such, prolonged N400 congruency effects lesion-based studies have also shown that verbal fluency is may reflect abnormally sustained efforts to integrate the supported by standard language networks and underlying incongruent word with the previous context (Nieuwland et al., white-matter connectivity (Baldo, Schwartz, Wilkins, & 2020). Curiously, a previous study found that patients with Dronkers, 2006; Costafreda et al., 2006; Griffis, Nenert, bilateral basal ganglia lesions exhibited similarly prolonged Allendorfer, & Szaflarski, 2017). Therefore, the ERP pattern N400 effects in sentence processing (up to 700 msec post- associated with lower verbal fluency e suggestive of less effi- word) (Kotz, Frisch, Von Cramon, & Friederici, 2003). In PD, cient predictive processing mechanisms, as previously dis- the worsening of the condition from PD with normal cognition cussed e may reflect not only executive dysfunction, but also to PD with MCI may lead to altered late integrational semantic disruption in semantic networks, in line with recent findings of processing as well. difficulties in temporal-dependent functions in PD with and More recently, N400 effects have postulated to reflect the without MCI (Horta-Barba et al., 2020; Martinez-Horta & ‘prediction error’ (albeit with nuances in their definition, see Kulisevsky, 2019; Lang et al., 2019). Willems, Frank, Nijhof, Hagoort, & Van den Bosch, 2016; Rabovsky, Hansen, & McClelland, 2018; Kuperberg, Brothers, & 4.4. Concluding remarks and future directions Wlotko, 2020). In predictive processing frameworks (Clark, 2013), predictions are sent top-down from higher to lower Overall, the current results suggest preserved predictive lan- level regions to ‘explain away’ the incoming sensory input and, guage processing in PD patients with normal cognitive status. then, the ‘prediction error’ e the portion of information that In turn, in PD patients with MCI, greater cognitive limitations remains unexplained e is sent back and used as a learning might hinder mechanisms associated with semantic prediction signal to update internal models and improve future pre- in normal circumstances. While these limitations may not dictions. Thus, prolonged N400 congruency effects may indi- prevent the pre-activation of relevant representations, as cate difficulties in this ‘learning’ phase, leading to less efficient indexed by canonical N400 amplitude modulations, they might internal model updating. negatively affect sentence processing due to poorer semantic Furthermore, also in line with less efficient predictive lan- anticipation and less efficient semantic processing when con- guage processing in PD patients with MCI, there were no ef- texts are predictive. Furthermore, both executive dysfunction fects in the anticipatory stage in this group, pointing to absent and damage within semantic networks may underlie these semantic anticipation. Note, however, that there were no ef- difficulties, as suggested by their association with low verbal fects in the matched subgroup of PD patients with normal fluency scores. Along with recent research (Horta-Barba et al., cognition either. In fact, there were no differences between the 2020; Martinez-Horta & Kulisevsky, 2019; Lang et al., 2019), two groups. Therefore, the absence or extreme reduction of the these findings emphasize the value of examining cognitive PNP in these two groups might be associated with factors other changes in PD patients with and without MCI in other domains than MCI, such as the older age of the patients in these samples beyond executive functions, like the language domain. Inter- (compared to the whole sample of PD with normal cognition), estingly, altered N400 effects are associated with a higher risk of or insufficient statistical power to detect differences, as sample transiting to dementia in individuals with amnesic MCI size was smaller in this case. (Olichney et al., 2008). Therefore, future longitudinal studies in PD could evaluate changes in N400 congruency effects as po- 4.3. Worse verbal fluency performance is associated tential markers of the transition to PD with MCI and to PD with with altered signatures of predictive language processing in dementia (for interested readers, see Supplementary Materials PD for a hierarchical regression analysis further attesting that the N400 congruency effect is a good predictor of global cognitive Interestingly, correlational analyses revealed that lower verbal decline in PD). Finally, as previously highlighted, future studies fluency scores were associated with reduced PNP and prolonged testing patients off medication could assess the exact role of DA N400 congruency effects. Similarly, previous evidence has in predictive sentence comprehension. Yet, the clinical value of shown that verbal fluency scores are a good predictor of the evaluating PD patients on stable doses of dopaminergic medi- status of predictive language processing in healthy older popu- cation must be emphasized, as it is the most representative of lation (Federmeier et al., 2002). Worse performance in verbal their habitual cognitive status and thus has the potential to fluency is generally associated with executive dysfunction, uncover limitations that may impact the everyday functioning such as difficulties in rule switching or inhibition. Executive of individuals with PD. 124 c o r t e x 1 4 1 ( 2 0 2 1 ) 1 1 2 e1 2 7 impairment in Parkinson disease: A multicenter pooled Funding analysis. Neurology, 75, 1062e1069. https://doi.org/10.1212/ WNL.0b013e3181f39d0e This study was funded by grants from the Fundacio  la Angwin, A. J., Arnott, W. L., Copland, D. A., Haire, M. P. L., Marato  de TV3 2014/U/477 and 20142910, and by financial Murdoch, B. E., Silburn, P. A., et al. (2009). Semantic activation support from the Center for Biomedical Research and Neuro- in Parkinson's disease patients on and off levodopa. Cortex, 45, 950e959. https://doi.org/10.1016/j.cortex.2009.02.012 degenerative Resources (CIBERNED). PLC was funded with a Angwin, A. J., Chenery, H. J., Copland, D. A., Arnott, W. L., pre-doctoral grant FPU15/05554 (FPU "Ayudas para la For- Grattan, R., Murdoch, B. E., et al. (2006). Priming of semantic macio  n de Profesorado Universitario") of the Spanish Ministry features in Parkinson's disease. Brain and Language, 99(1e2), of Education, Culture and Sport. JP was funded by PERIS 102e103. https://doi.org/10.1016/j.bandl.2006.06.060 (expedient number: SLT008/18/00088) of Generalitat de Cata- Angwin, A. J., Chenery, H. J., Copland, D. A., Arnott, W. L., lunya. JK and HB-K were funded by FIS PI18/01717. HB -K was Murdoch, B. E., & Silburn, P. A. (2004). Dopamine and semantic funded by Rı́o Hortega CM17/00209 of Instituto de Salud Carlos activation: An investigation of masked direct and indirect priming. Journal of the International Neuropsychological Society: III (ISCIII) (Spain). JM was funded by Rı́o Hortega CM15/00071 of JINS, 10(1), 15e25. https://doi.org/10.1017/S1355617704101033 Instituto de Salud Carlos III (ISCIII) (Spain). None of these Angwin, A. J., Chenery, H. J., Copland, D. A., Murdoch, B. E., & funding sources had any involvement in the conduct of the Silburn, P. A. (2004). The time course of semantic activation in research or the preparation of the article. Parkinson's disease. Brain and Language, 91, 145e146. https:// doi.org/10.1016/j.bandl.2004.06.076 Angwin, A. J., Chenery, H. J., Copland, D. A., Murdoch, B. E., & CRediT authorship contribution statement Silburn, P. A. (2005). Summation of semantic priming and complex sentence comprehension in Parkinson's disease. Cognitive Brain Research, 25(1), 78e89. https://doi.org/10.1016/ PLC: Conceptualization, Formal analysis, Visualization, j.cogbrainres.2005.04.008 Writing - Original Draft, and Writing - Review & Editing; JP: Angwin, A. J., Dissanayaka, N. N., McMahon, K. L., Silburn, P. A., & Conceptualization, Supervision, Methodology, Writing - Re- Copland, D. A. (2017). Lexical ambiguity resolution during view & Editing, Resources, and Funding acquisition; JM: sentence processing in Parkinson's disease: An event-related Conceptualization, Supervision, Methodology and Writing - potential study. Plos One, 12(5), Article e0176281. https:// Review & Editing; SMH: Investigation and Writing - Review & doi.org/10.1371/journal.pone.0176281 Angwin, A. J., Dissanayaka, N. N., Moorcroft, A., McMahon, K. L., Editing; JML: Investigation and Writing - Review & Editing; Silburn, P. A., & Copland, D. A. (2017). A neurophysiological AHB: Investigation; HBK: Investigation; JK: Resources, study of semantic processing in Parkinson's disease. Journal of Writing - Review & Editing, and Funding acquisition; ARF: the International Neuropsychological Society: JINS, 23(1), 78e89. Conceptualization, Supervision, Methodology, and Writing - https://doi.org/10.1017/S1355617716000953 Review & Editing. Arnott, W. L., Chenery, H. J., Angwin, A. J., Murdoch, B. E., Silburn, P. A., & Copland, D. A. (2010). Decreased semantic competitive inhibition in Parkinson's disease: Evidence from an investigation of word search performance. International Declaration of competing interests journal of speech-language pathology, 12(5), 437e445. https:// doi.org/10.3109/17549507.2010.492875 The authors declare no disclosure of financial interests and Arnott, W. L., Chenery, H. J., Murdoch, B. E., & Silburn, P. A. (2001). potential conflict of interest. Semantic priming in Parkinsons disease: Evidence for delayed spreading activation. Journal of Clinical and Experimental Neuropsychology, 23(4), 502e519. https://doi.org/10.1076/ jcen.23.4.502.1224 Acknowledgments Baldo, J. V., Schwartz, S., Wilkins, D., & Dronkers, N. F. (2006). Role of frontal versus temporal cortex in verbal fluency as revealed In honor and loving memory of Dr. Jordi Riba, who passed by voxel-based lesion symptom mapping. Journal of the away in the process of preparing this manuscript. We would International Neuropsychological Society: JINS, 12(6), 896e900.  like to thank Dr. Ion Iarritu, Dr. Laura Ferreri, Miriam Cornella https://doi.org/10.1017/S1355617706061078 Brothers, T., Swaab, T. Y., & Traxler, M. J. (2015). Effects of and Fernando Martı́nez for their help with data collection and prediction and contextual support on lexical processing: technical support. Prediction takes precedence. Cognition, 136, 135e149. https:// doi.org/10.1016/j.cognition.2014.10.017 Brunia, C. H., Hackley, S. A., van Boxtel, G. J., Kotani, Y., & Supplementary data Ohgami, Y. (2011). Waiting to perceive: Reward or punishment? Clinical Neurophysiology, 122(5), 858e868. https:// Supplementary data to this article can be found online at doi.org/10.1016/j.clinph.2010.12.039 https://doi.org/10.1016/j.cortex.2021.03.032. Clark, A. (2013). Whatever next? Predictive brains, situated agents, and the future of cognitive science. The Behavioral and Brain Sciences, 36(3), 181e204. https://doi.org/10.1017/ references S0140525X12000477 Copland, D. A., Chenery, H. J., & Murdoch, B. E. (2000a). Impaired meaning selection for polysemous words in individuals with Aarsland, D., Bronnick, K., Williams-Gray, C., Weintraub, D., nonthalamic subcortical lesions and Parkinson's disease. Brain Marder, K., Kulisevsky, J., et al. (2010). Mild cognitive and Language, 74(3), 360e363. c o r t e x 1 4 1 ( 2 0 2 1 ) 1 1 2 e1 2 7 125 Copland, D. A., Chenery, H. J., & Murdoch, B. E. (2000b). Reviews, 101, 371e395. https://doi.org/10.1037/0033- Understanding ambiguous words in biased sentences: 295X.101.3.371 Evidence of transient contextual effects in individuals with Griffis, J. C., Nenert, R., Allendorfer, J. B., & Szaflarski, J. P. (2017). nonthalamic subcortical lesions and Parkinson's disease. Damage to white matter bottlenecks contributes to language Cortex, 36(5), 601e622. https://doi.org/10.1016/S0010-9452(08) impairments after left hemispheric stroke. NeuroImage: 70541-0 Clinical, 14, 552e565. https://doi.org/10.1016/j.nicl.2017.02.019 Copland, D. A., Chenery, H. J., & Murdoch, B. E. (2001). Discourse Grisoni, L., Miller, T. M., & Pulvermüller, F. (2017). Neural priming of homophones in individuals with dominant correlates of semantic prediction and resolution in sentence nonthalamic subcortical lesions, cortical lesions and Parkinsons processing. Journal of Neuroscience, 2800e2816. https://doi.org/ disease. Journal of Clinical and Experimental Neuropsychology, 23(4), 10.1523/JNEUROSCI.2800-16.2017 538e556. https://doi.org/10.1076/jcen.23.4.538.1233 Grossman, M., Cooke, A. C., Devita, C., Alsop, D., Detre, J., Chen, W., Copland, D. A., Sefe, G., Ashley, J., Hudson, C., & Chenery, H. J. et al. (2002). Age-related changes in working memory during (2009). Impaired semantic inhibition during lexical ambiguity sentence comprehension: An fMRI study. Neuroimage, 15, repetition in Parkinson's disease. Cortex, 45(8), 943e949. 302e317. https://doi.org/10.1006/nimg.2001.0971 https://doi.org/10.1016/j.cortex.2009.02.023 Grossman, M., Cooke, A., Devita, C., Lee, C., Alsop, D., Detre, J., Costafreda, S. G., Fu, C. H., Lee, L., Everitt, B., Brammer, M. J., & et al. (2003). Grammatical and resource components of David, A. S. (2006). A systematic review and quantitative sentence processing in Parkinson's disease: An fMRI study. appraisal of fMRI studies of verbal fluency: Role of the left Neurology, 60(5), 775e781. https://doi.org/10.1212/ inferior frontal gyrus. Human Brain Mapping, 27, 799e810. 01.WNL.0000044398.73241.13 https://doi.org/10.1002/hbm.20221 Holm, S. (1979). A simple sequentially rejective multiple test Delorme, A., & Makeig, S. (2004). EEGLAB: An open source toolbox procedure. Scandinavian Journal of Statistics, 6, 65e70. for analysis of single-trial EEG dynamics including independent Horta-Barba, A., Pagonabarraga, J., Martı́nez-Horta, S., Marı́n- component analysis. Journal of neuroscience methods, 134(1), 9e21. Lahoz, J., Sampedro, F., Ferna  ndez-Bobadilla, R., et al. (2020). https://doi.org/10.1016/j.jneumeth.2003.10.009 The free and cued selective reminding test in Parkinson's Duchon, A., Perea, M., Sebastia  n-Galle s, N., Martı́, A., & disease mild cognitive impairment: Discriminative accuracy Carreiras, M. (2013). Espal: One-stop shopping for Spanish and neural correlates. Frontiers in Neurology, 11, 240. https:// word properties. Behavior Research Methods, 45(4), 1246e1258. doi.org/10.3389/fneur.2020.00240 https://doi.org/10.3758/s13428-013-0326-1 Huettig, F. (2015). Four central questions about prediction in Emre, M., Aarsland, D., Brown, R., Burn, D. J., Duyckaerts, C., & language processing. Brain research, 1626, 118e135. https:// Mizuno, Y. (2007). Clinical diagnostic criteria for dementia doi.org/10.1016/j.brainres.2015.02.014 associated with Parkinson's disease. Movement Disorders, 22, Hughes, A. J., Daniel, S. E., Kilford, L., & Lees, A. J. (1992). Accuracy 1689e1707. https://doi.org/10.1002/mds.21507, 2007. of clinical diagnosis of idiopathic Parkinson's disease: A Federmeier, K. D., & Kutas, M. (1999). A rose by any other name: clinico-pathological study of 100 cases. Journal of Neurological Long-term memory structure and sentence processing. Journal and Neurosurgical Psychiatry, 55, 181e184. https://doi.org/ of Memory and Language, 41(4), 469e495. https://doi.org/ 10.1136/jnnp.55.3.181 10.1006/jmla.1999.2660 Kischka, U., Kammer, T. H., Maier, S., Weisbrod, M., Thimm, M., & Federmeier, K. D., & Kutas, M. (2005). Aging in context: Age- Spitzer, M. (1996). Dopaminergic modulation of semantic related changes in context use during language network activation. Neuropsychologia, 34(11), 1107e1113. comprehension. Psychophysiology, 42(2), 133e141. https:// https://doi.org/10.1016/0028-3932(96)00024-3 doi.org/10.1111/j.1469-8986.2005.00274.x Kotz, S. A., Frisch, S., Von Cramon, D. Y., & Friederici, A. D. (2003). Federmeier, K. D., Kutas, M., & Schul, R. (2010). Age-related and Syntactic Language Processing: ERP lesion data on the role of the individual differences in the use of prediction during language basal ganglia. Journal of the International Neuropsychological Society: comprehension. Brain and Language, 115(3), 149e161. https:// JINS, 9(7), 1053e1060. https://doi.org/10.1017/S1355617703970093 doi.org/10.1016/j.bandl.2010.07.006 Kudlicka, A., Clare, L., & Hindle, J. V. (2011). Executive functions in Federmeier, K. D., McLennan, D. B., De Ochoa, E., & Kutas, M. Parkinson's disease: Systematic review and meta-analysis. (2002). The impact of semantic memory organization and Movement disorders, 26(13), 2305e2315. https://doi.org/10.1002/ sentence context information on spoken language processing mds.23868 by younger and older adults: An ERP study. Psychophysiology, Kuperberg, G. R., Brothers, T., & Wlotko, E. W. (2020). A Tale of 39(2), 133e146. https://doi.org/10.1111/1469-8986.3920133 Two Positivities and the N400: Distinct neural signatures are Ferna  ndez de Bobadilla, R., Pagonabarraga, J., Martı́nez-Horta, S., evoked by confirmed and violated predictions at different Pascual-Sedano, B., Campolongo, A., & Kulisevsky, J. (2013). levels of representation. Journal of Cognitive Neuroscience, 32(1), Parkinson’s disease-cognitive rating scale: Psychometrics for 12e35. https://doi.org/10.1162/jocn_a_01465 mild cognitive impairment. Movement Disorders, 28(10), Kutas, M., & Federmeier, K. D. (2011). Thirty years and counting: 1376e1383. https://doi.org/10.1002/mds.25568 Finding meaning in the N400 component of the event-related Friederici, A. D., Kotz, S. A., Werheid, K., Hein, G., & von brain potential (ERP). Annual Review of Psychology, 62, 621e647. Cramon, D. Y. (2003). Syntactic processes in Parkinson's https://doi.org/10.1146/annurev.psych.093008.131123 disease: A dissociation between early automatic and late Kutas, M., & Hillyard, S. A. (1980). Reading senseless sentences: integrational processes. Neuropsychology, 17(1), 133e142. Brain potentials reflect semantic incongruity. Science, https://doi.org/10.1037/0894-4105.17.1.133 207(4427), 203e205. https://doi.org/10.1126/science.7350657 Goetz, C. G., Tilley, B. C., Shaftman, S. R., Stebbins, G. T., Fahn, S., Kutas, M., & Hillyard, S. A. (1984). Brain potentials during reading Martinez-Martin, P., et al. (2008). Movement Disorder Society- reflect word expectancy and semantic association. Nature, Sponsored Revision of the Unified Parkinson's Disease Rating 307(5947), 161. https://doi.org/10.1038/307161a0 Scale (MDS-UPDRS): Scale presentation and clinimetric testing Kutas, M., Iragui, V. J., Niu, Y. Q., D'Avanzo, T. J., Yang, J. C., results. Movement Disorders, 23, 2129e2170. https://doi.org/ Salmon, D. P., et al. (2013). Altered N400 congruity effects in 10.1002/mds.22340 Parkinson's disease without dementia. Cognitive Graesser, A. C., Singer, M., & Trabasso, T. (1994). Constructing Electrophysiology of Attention: Signals of the Mind, 254e267. inferences during narrative text comprehension. Psychological https://doi.org/10.1016/B978-0-12-398451-7.00020-8 126 c o r t e x 1 4 1 ( 2 0 2 1 ) 1 1 2 e1 2 7 Lakens, D. (2016). Will knowledge about more efficient study designs Olichney, J. M., Taylor, J. R., Gatherwright, J., Salmon, D. P., increase the willingness to pre-register?. https://doi.org/10.31222/ Bressler, A. J., Kutas, M., et al. (2008). Patients with MCI and osf.io/svzyc. Retrieved from osf.io/uy8x5. N400 or P600 abnormalities are at very high risk for conversion Lang, S., Hanganu, A., Gan, L. S., Kibreab, M., Auclair-Ouellet, N., to dementia. Neurology, 70(19 Part 2), 1763e1770. https:// Alrazi, T., et al. (2019). Network basis of the dysexecutive and doi.org/10.1212/01.wnl.0000281689.28759.ab posterior cortical cognitive profiles in Parkinson's disease. Oostenveld, R., Fries, P., Maris, E., & Schoffelen, J. M. (2011). Movement Disorders, 34(6), 893e902. https://doi.org/10.1002/ Fieldtrip: Open source software for advanced analysis of MEG, mds.27674. Jun. EEG, and invasive electrophysiological data. Computational Lau, E. F., Namyst, A., Fogel, A., & Delgado, T. (2016). A direct Intelligence and Neuroscience, 1. https://doi.org/10.1155/2011/ comparison of N400 effects of predictability and incongruity in 156869, 2011. adjective-noun combination. Collabra, 2(1). https://doi.org/ Pagonabarraga, J., Kulisevsky, J., Llebaria, G., Garcı́a-Sa  nchez, C., 10.1525/collabra.40 Pascual-Sedano, B., & Gironell, A. (2008). Parkinson's disease- Lau, E. F., Phillips, C., & Poeppel, D. (2008). A cortical network for cognitive rating scale: A new cognitive scale specific for semantics:(de) constructing the N400. Nature Reviews Parkinson's disease. Movement disorders: Official Journal of the Neuroscience, 9(12), 920e933. https://doi.org/10.1038/nrn2532 Movement Disorder Society, 23(7), 998e1005. https://doi.org/ Leo n-Cabrera, P., Flores, A., Rodrı́guez-Fornells, A., & Morı́s, J. 10.1002/mds.22007 (2019). Ahead of time: Early sentence slow cortical Payne, B. R., & Federmeier, K. D. (2018). Contextual constraints on modulations associated to semantic prediction. Neuroimage, lexico-semantic processing in aging: Evidence from single- 189, 192e201. https://doi.org/10.1016/ word event-related brain potentials. Brain Research, 1687, j.neuroimage.2019.01.005 117e128. https://doi.org/10.1016/j.brainres.2018.02.021 Leo n-Cabrera, P., Rodrı́guez-Fornells, A., & Morı́s, J. (2017). Payne, B. R., & Silcox, J. W. (2019). Aging, context processing, and Electrophysiological correlates of semantic anticipation comprehension. In , Vol. 71. Psychology of learning and during speech comprehension. Neuropsychologia, 99, 326e334. motivation (pp. 215e264). Academic Press. https://doi.org/ https://doi.org/10.1016/j.neuropsychologia.2017.02.026 10.1016/bs.plm.2019.07.001. Litvan, I., Aarsland, D., Adler, C. H., Goldman, J. G., Kulisevsky, J., Pell, M., & Monetta, L. (2008). How Parkinson's disease affects Mollenhauer, B., et al. (2011). MDS task force on mild cognitive non-verbal communication and language processing. impairment in Parkinson's disease: Critical review of PD-MCI. Language and Linguistics Compass, 2(5), 739e759. https://doi.org/ Movement disorders, 26(10), 1814e1824. https://doi.org/10.1002/ 10.1111/j.1749-818X.2008.00074.x mds.23823 ~ a-Casanova, J., Quin Pen ~ ones-Ubeda, S., Gramunt-Fombuena, N., Li, X., Zhang, Y., Xia, J., & Swaab, T. Y. (2017). Internal Quintana-Aparicio, M., Aguilar, M., Badenes, D., et al. (2009). mechanisms underlying anticipatory language processing: Spanish multicenter normative studies (NEURONORMA project): Evidence from event-related-potentials and neural Norms for verbal fluency tests. Archives of Clinical Neuropsychology, oscillations. Neuropsychologia, 102, 70e81. https://doi.org/ 24, 395e411. https://doi.org/10.1093/arclin/acp042 10.1016/j.neuropsychologia.2017.05.017 Pulvermüller, F., & Grisoni, L. (2020). Semantic prediction in brain  pez-Calderon, J., & Luck, S. J. (2014). ERPLAB: An open-source Lo and mind. Trends in Cognitive Sciences, 24(10), 781e784. https:// toolbox for the analysis of event-related potentials. Frontiers in doi.org/10.1093/arclin/acp042 Human Neuroscience, 8, 213. https://doi.org/10.3389/ Rabovsky, M., Hansen, S. S., & McClelland, J. L. (2018). Modelling fnhum.2014.00213 the N400 brain potential as change in a probabilistic Maris, E., & Oostenveld, R. (2007). Nonparametric statistical testing representation of meaning. Nature Human Behaviour, 2(9), of EEG-and MEG-data. Journal of Neuroscience Methods, 164(1), 693e705. https://doi.org/10.1038/s41562-018-0406-4 177e190. https://doi.org/10.1016/j.jneumeth.2007.03.024 Taylor, W. I. (1953). Cloze procedure: A new tool for measuring Martinez-Horta, & S., & Kulisevsky, J. (2019). Mild cognitive readability. Journal and Mass Communication Quartely, 30, 415. impairment in Parkinson's disease. Journal of Neural https://doi.org/10.1177/107769905303000401 Transmission, 126(7), 897e904. https://doi.org/10.1007/s00702- Tomlinson, C. L., Stowe, R., Patel, S., Rick, C., Gray, R., & 019-02003-1 Clarke, C. E. (2010). Systematic review of levodopa dose Mattox, S. T., Valle-Incla  n, F., & Hackley, S. A. (2006). equivalency reporting in Parkinson's disease. Movement Psychophysiological evidence for impaired reward anticipation Disorders, 25, 2649e2653. https://doi.org/10.1002/mds.23429 in Parkinson's disease. Clinical Neurophysiology, 117(10), Unsworth, N., Spillers, G. J., & Brewer, G. A. (2011). Variation in verbal 2144e2153. https://doi.org/10.1016/j.clinph.2006.05.026 fluency: A latent variable analysis of clustering, switching, and Mestres-Misse , A., Rodriguez-Fornells, A., & Münte, T. F. (2007). overall performance. Quarterly Journal of Experimental Psychology, Watching the brain during meaning acquisition. Cerebral 64(3), 447e466. https://doi.org/10.1080/17470218.2010.505292 Cortex, 17(8), 1858e1866. https://doi.org/10.1093/cercor/bhl094 Van Berkum, J. J., Brown, C. M., Zwitserlood, P., Kooijman, V., & Monchi, O., Petrides, M., Doyon, J., Postuma, R. B., Worsley, K., & Hagoort, P. (2005). Anticipating upcoming words in discourse: Dagher, A. (2004). Neural bases of set-shifting deficits in Evidence from ERPs and reading times. Journal of Experimental Parkinson's disease. Journal of Neuroscience, 24, 702e710. Psychology: Learning, Memory, and Cognition, 31(3), 443. https:// https://doi.org/10.1523/JNEUROSCI.4860-03.2004 doi.org/10.1037/0278-7393.31.3.443 Muslimovic , D., Post, B., Speelman, J. D., & Schmand, B. (2005). Van Berkum, J. J., Hagoort, P., & Brown, C. M. (1999). Semantic Cognitive profile of patients with newly diagnosed Parkinson integration in sentences and discourse: Evidence from the disease. Neurology, 65(8), 1239e1245. https://doi.org/10.1212/ N400. Journal of Cognitive Neuroscience, 11(6), 657e671. https:// 01.wnl.0000180516.69442.95 doi.org/10.1162/089892999563724 Nieuwland, M. S., Barr, D. J., Bartolozzi, F., Busch-Moreno, S., Whiteside, D. M., Kealey, T., Semla, M., Luu, H., Rice, L., Darley, E., Donaldson, D. I., et al. (2020). Dissociable effects of Basso, M. R., et al. (2016). Verbal fluency: Language or prediction and integration during language comprehension: Evidence executive function measure? Applied Neuropsychology: Adult, from a large-scale study using brain potentials. Philosophical 23(1), 29e34. https://doi.org/10.1080/23279095.2015.1004574 Transactions of the Royal Society. https://doi.org/10.1098/ Wicha, N. Y., Bates, E. A., Moreno, E. M., & Kutas, M. (2003). Potato rstb.2018.0522. B37520180522. not pope: Human brain potentials to gender expectation and c o r t e x 1 4 1 ( 2 0 2 1 ) 1 1 2 e1 2 7 127 agreement in Spanish spoken sentences. Neuroscience Letters, Wlotko, E. W., & Federmeier, K. D. (2012). Age-related changes in 346(3), 165e168. https://doi.org/10.1016/S0304-3940(03)00599-8 the impact of contextual strength on multiple aspects of Willems, R. M., Frank, S. L., Nijhof, A. D., Hagoort, P., & Van den sentence comprehension. Psychophysiology, 49(6), 770e785. Bosch, A. (2016). Prediction during natural language https://doi.org/10.1111/j.1469-8986.2012.01366.x comprehension. Cerebral Cortex, 26(6), 2506e2516. https:// doi.org/10.1093/cercor/bhv075