RESOLVING VERBAL INTERFERENCE IN DISRUPTED LEXICAL-SEMANTIC PROCESSING: THE ROLE OF INHIBITORY CONTROL

RESOLVING VERBAL INTERFERENCE IN DISRUPTED

LEXICAL-SEMANTIC PROCESSING:

THE ROLE OF INHIBITORY CONTROL

by Irina Chupina

A Master’s thesis submitted in partial fulfilment of the requirements for the degree of

Master of Science

(Clinical Linguistics)

at the Joint European Erasmus Mundus Master’s Programme in Clinical Linguistics+ (EMCL+)

UNIVERSITY OF GRONINGEN August 1, 2020

RESOLVING VERBAL INTERFERENCE IN DISRUPTED LEXICAL-SEMANTIC PROCESSING: THE ROLE OF INHIBITORY CONTROL

Irina Chupina

Under the supervision of: Vitória Piai, PhD, Radboud University Srdjan Popov, PhD, Universtity of Groningen

ABSTRACT

Goal-consistent lexical selection in word production under distraction is known to be facilitated by inhibitory control. The activity of the control networks gets proportionally enhanced with growing interference, manifesting itself as theta power (4-7 Hz) increases at the mid-frontal electrodes. However, how efficient the top-down networks are at resolving competition when downstream processing is impaired remains unclear. Particularly, the question is whether the networks upregulate their activity to support disrupted lexical selection, a compensatory mechanism suggested in aphasia. In this thesis, previously collected electroencephalographic (EEG) data were analysed to investigate the inhibitory control activity in six healthy older adults and six participants with chronic aphasia, who had lesions overlapping in the middle temporal gyrus, an area associated with conceptually driven lexical retrieval. They performed the picture-word interference task with semantically related, unrelated and congruent distractors. Behaviourally, while controls demonstrated the classic semantic (related vs unrelated distractors) and Stroop (related vs congruent distractors) interference, patients showed higher naming costs and no semantic interference. Electrophysiologically, no interference effects and no effect magnitude differences were found within and between groups, which makes interpreting network activity rather challenging. Possibly attributed to insufficient data and modest sample size, the lack of predicted theta effects as well as the paradoxically high power values in the congruent condition in controls might also be linked to age-related neurophysiological changes. The findings suggest that understanding the role of inhibitory control in disrupted lexical-semantic processing first requires establishing the EEG signatures of conflict control in healthy cognitive ageing.

Keywords: inhibitory control, lexical selection, middle temporal gyrus, picture-word

ACKNOWLEDGEMENTS

I would like to thank Professor Roelien Bastiaanse for conceptualising the EMCL+,

developing it into a diverse and challenging programme with excellent instruction and training and bringing together aspiring students from all over the world. I also express my gratitude to the Erasmus+ for the financial support which enabled me to undertake this Master’s.

I thankfully acknowledge the contributions of the EMCL+ faculty at the universities of Eastern Finland, Groningen and Potsdam. This thesis would not have been possible without their professional expertise, personal support and inspiring academic example. My special thanks go to Professor Stefan Werner, who greeted us with a warm welcome in cold Finland when our cohort embarked on this exciting two-year journey. I would also like to thank my supervisor Dr Srdjan Popov for his helpful feedback and his course on Neuroimaging where my fascination with electrophysiology started.

I express my warmest gratitude and endless appreciation to Dr Vitória Piai for taking me under her wing and masterfully guiding through the internship and thesis writing. At Donders, I learnt many valuable lessons about research, teamwork and, of course, the importance of having fun (and, ideally, your own vegetable garden). It has been an honour to be part of Dr Piai’s amazing Language Function and Dysfunction lab, and to get to know her and all the lovely people there.

And last but not least, I would like to thank my fellow EMCLer Priscila Borba Borges, whose critical eye has helped me improve my academic work and whose friendship has made this programme much more special. We sure walked those walks and talked those talks, both literally and metaphorically.

TABLE OF CONTENTS

ABSTRACT ... iii

ACKNOWLEDGEMENTS ... iv

LIST OF TABLES ... vi

LIST OF FIGURES ... vii

INTRODUCTION Inhibitory control as part of language-related cognitive function……….………… 1

Inhibitory control in disrupted lexical-semantic processing ……….………..………3

Current study………..6

METHOD Participants……….…..………. 7

Design and materials………...……….………10

Procedure and EEG data acquisition ……….…………...…………...………10

Behavioural data analysis………...………..………11

EEG data analyses………...……….………12

RESULTS Error rates………...………..…………14

Response times………...…………..………14

Time-averaged power spectra………...………17

DISCUSSION………..……...………18

LIST OF TABLES

Table 1. Individual patient information, including language testing data from the WAB and present study error rates………...………...……..…..…8 Table 2. Individual control group information, including present study error rates……....…8 Table 3. Error rates and types………..………...….…….14 Table 4. Group-averaged response times across the three conditions and between-group differences in the Stroop and semantic interference effects………...…..…..…15 Table 5. Results of the statistical analyses for response times: fixed and random effects….16

LIST OF FIGURES

Figure 1. Individual lesion and overlap maps of the patients………..………9 Figure 2. Example of the experimental picture-word item in three conditions……..……..10 Figure 3. Biosemi64 electrode montage based on the 10-20 system with marked electrodes of interest………..13 Figure 4. Individual-averaged and group-averaged response times in seconds for the control and patient groups across the experimental conditions………...…..…….15 Figure 5. Time-frequency representations of conditions averaged across trials and participants within groups……….……...…….17 Figure 6. Time-frequency representations of semantic and Stroop interference effects within groups………...………....18 Figure 7. Power spectrum plot in the 350-650ms time window on the average of mid-frontal electrodes for controls and patients……….…….….23

INTRODUCTION

As humans, we constantly engage in complex goal-driven behaviours (Miller & Cohen, 2001). This ability is subserved by extensive neural networks, which regulate processing by coordinating lower-order sensory, memory and motor operations in accordance with our goals. These mechanisms, called cognitive or executive control, are indispensable for supporting a wide array of brain functions such as task performance monitoring, online maintenance of contextual information, error detection and inhibition of irrelevant information (Posner & Petersen, 1990; Botvinick, Braver, Barch, Carter, & Cohen, 2001; Roelofs & Hagoort, 2002). The role of executive control in cognition has been acknowledged in psychology, but despite emerging evidence that domain-general control might be critical for effective language use (for review see Nozari & Novick, 2017), its contribution to language processing remains less clear. Meanwhile, controlled action seems to be particularly vital for speech production (Roelofs & Piai, 2011), since planning both a word and a multi-word utterance engages executive control networks (Roelofs, 2003; Hartsuiker, 2014). Inhibitory control as part of language-related cognitive function

One of the cognitive control functions relevant for language production is inhibitory control. It allows language users to exert top-down control to resolve competition from concurrent information that interferes with goal-driven verbal behaviour. Such interference occurs when similar representations are co-activated in a layer of the language system (e.g. on the level of phonemes or words). Strong activations cause intense competition, which complicates quick automatic selection and generates a conflict signal (Botvinick et al., 2001; Nozari & Novick, 2017). In word planning, this might happen when related lexical-semantic representations get activated in the lexicon. Because both nodes – being from the same semantic category – receive activation of a comparable strength, the goal-compatible node is unable to reach the activation threshold necessary for its selection and further phonological encoding (Levelt, Roelofs & Meyer, 1999; Roelofs, 1992). Here, the top-down inhibitory control steps in and helps resolve the competition by sending signals that amplify the activation of goal-relevant representations (Munakata, Herd, Chatham, Depue, Banich, & O’Reilly, 2011). As a result, the competitor becomes inhibited collaterally (i.e. indirectly), and the relevant word reaches the activation threshold.

The neural sources of this competitive inhibition, associated with various interference paradigms, have been localised to the dorsolateral prefrontal cortex (dLPFC, Miller & Cohen, 2001; January, Trueswell, & Thompson-Schill, 2009), midsuperior frontal cortex (Piai, Roelofs, Jensen, Schoffelen, & Bonnefond, 2014) and parts of the limbic system such as the dorsolateral anterior cingulate cortex (dACC, Bush, Luu, & Posner, 2000; Piai,

Roelofs, Acheson, & Takashima, 2013). According to Dosenbach and colleagues (2007, 2008), the dLPFC, being part of the fronto-parietal control network together with the inferior parietal lobule, dorsal frontal cortex, intraparietal sulcus, precuneus and middle cingulate cortex, can represent and maintain abstract information such as goals, task specifications and contexts. This allows it to aid adjustment and implementation of cognitive control in response to feedback from other task-relevant neocortical regions. In case of lexical-semantic processing, the pyramidal neurons, abundantly found in the PFC, are able to directly excite remote processing areas to support representations (Cohen, Dunbar, & McClelland, 1990; Munakata et al., 2011). The dACC, which tends to upregulate its activity when the task is more demanding or error-prone compared to when the task is easy (Roelofs & Hagoort, 2002), belongs to the cingulo-opercular control network that also involves the medial superior frontal cortex, anterior PFC, anterior insula, frontal operculum regions and thalamus (Dosenbach et al., 2008). This network supports stable set-maintenance throughout the task, engaging in parallel processing with the fronto-parietal network: dLPFC and dACC demonstrate strong connections at the structural level via fibre tracts (Goldman-Rakic, 1987) and at the functional level through oscillatory phase coupling (Hanslmayr, Pastötter, Bäuml, Gruber, Wimber, & Klimesch, 2008).

While the debate on the anatomy of cognitive control and the particular function of the network nodes is ongoing (Botvinick et al., 2001; Roelofs, van Turennout, & Coles, 2006; Nozari & Pinet, 2020), the literature on the electrophysiological correlates of executive control is less controversial. Interference studies have repeatedly reported a robust theta band power increase (4-7 Hz) over midline frontal regions, which is widely acknowledged as the marker of cognitive control (Cavanagh & Frank, 2014; Cohen, 2014). Strong power increases or decreases occurring in a narrow frequency band are thought to reflect oscillatory activity in the brain and serve as its proxy, since neuronal oscillations cannot be measured directly (Luck, 2014). The ability of neuronal populations to oscillate collectively in different frequencies is well documented, and modern theories suggest that, by synchronising their activity, neurons organise information processing through cyclic temporal reference frames. This mechanism enables dynamic communication within and across the neural networks, thus supporting brain function and, in particular, enabling effective integration of information processed by distributed regions (Fries, 2005; Buzsáki,

Logothetis, & Singer, 2013). Top-down inhibitory control exerted in response to conflict has been linked to a specific type of theta increases that produce very similar spatial-temporal-frequency signatures across experimental paradigms and are differentiable from other

prefrontal thetas associated with working memory load and error monitoring (Cohen, 2014). Thus, neurons, oscillating in the theta band, establish and support a time window for monitoring and adjusting sequenced actions by phase-locking mid-frontal areas with the task-relevant regions such as the lateral prefrontal, parietal or motor cortices where the lower-level processing takes place (Fries, 2005).

Theta increases marking response control in verbal interference have been elicited with paradigms that use distracting stimuli to tap regulatory processes, including the classic Stroop task (Stroop, 1935) and the more linguistically diverse picture-word interference task (PWI, Rosinski, Golinkoff, & Kukish, 1975). In PWI, the goal is to name the target picture presented with a distractor word superimposed on top of it. The degree of interference induced by each picture-word pair is determined by the type of semantic relations the distractor has to the target. This relation modulates the amount of conflict produced at the lexical selection stage, which imposes varying demands on top-down control as reflected in the theta power magnitude (Piai et al., 2014). Theta band activity is measured in relation to the baseline, typically the unrelated condition created by pairing the target picture with a semantically and phonologically unrelated distractor word (CHAIR with picture pig, Fig. 2). Compared to the baseline, naming with a distractor that comes from the same semantic category (COW with picture pig, category ‘animals’) is more challenging due to amplified lexical competition. Behaviourally, this related condition elicits longer response times (RTs), as naming requires additional resources to override the activation of the irrelevant word – the phenomenon called semantic interference (Glaser & Düngelhoff, 1984). It puts extra demands on the frontal inhibitory control networks, which upregulate their activity in the theta band in the 350-650ms time window post-stimulus onset (Piai et al., 2014; Krott, Medaglia, & Porcaro, 2019). Interference from semantically related words can be further highlighted by contrasting it with the congruent condition, which produces the Stroop interference effect (MacLeod, 1991). Naming with a congruent distractor word (PIG with picture pig) elicits the fastest RTs, since in this case the distractor facilitates naming by prompting the target word. The effect is also present electrophysiologically (Shitova, Roelofs, Schriefers, Bastiaansen, & Schoffelen, 2017) and is typically more prominent than semantic interference (Piai et al., 2014).

Inhibitory control in disrupted lexical-semantic processing

Compared to our knowledge about inhibitory control in neurotypical populations, data on how it is recruited in atypical language use are much scarcer. Meanwhile, the fact that speech generation requires attention at the stage of word planning, as well as constant monitoring

and adjustment in response to sensory interference (Roelofs & Piai, 2011; Roelofs & Hagoort, 2002), implies that cognitive control impairments and speech production deficits can be interdependent. This puts clinical populations such as people with aphasia, who have limited processing resources and disrupted functional and structural neural connectivity, in a particularly vulnerable position. Naming difficulties are, indeed, prevalent among patients with various lesion profiles and severity of impairment (Goodglass, 1980; Kohn & Goodglass, 1985; Ardila & Rosselli, 1993). But while it is known that attention deficits do lead to deteriorated language performance (Murray, 1999), what kind of impact a lesion to the speech-related regions produces on the functioning of the structurally intact cognitive control networks remains an open question.

Naming breaks down in multiple ways, and based on the patient’s verbal behaviour and lesion locus it might be possible to determine the functional locus of the deficit, i.e. which stage of speech production is impaired (Kohn & Goodglass, 1985; Garrett, 1992). Baldo and colleagues (2013), who used the voxel-based lesion symptom mapping (VLSM) technique, analysed picture naming performance of people with aphasia and correlated the results with their lesion profiles. Consistent with the observation that speech difficulties are common for people with a variety of lesions, they found that naming deficits were associated with damage to large areas of the left anterior, lateral and posterior temporal regions, as well as parts of the inferior parietal cortex. However, after controlling for the visual-perceptual processing and speech fluency deficits, the region critical for successful name retrieval was confined to the left mid-posterior middle temporal gyrus (MTG) and the adjacent white matter.

The MTG has been repeatedly implicated in lexical-semantic processing during picture naming by neuroimaging studies conducted with healthy adults and is considered crucial for conceptually driven lexical access (Indefrey, 2011; Roelofs, 2014). Increased semantic error rates and hesitations associated with mid-posterior MTG lesions in impaired naming corroborate this interpretation (e.g., behavioural studies, Ardila & Rosselli, 1993;

VLSM studies, Schwartz et al., 2009; virtual lesion studies, Krieg, Sollmann, Tanigawa, Foerschler, Meyer, & Ringel, 2016). Although incorrect lexical-semantic selection may be associated with other functional impairments, such as phonological encoding issues (Dell, Schwartz, Martin, Saffran & Gagnon, 1997), hesitations seem to have a more direct link to difficulties with lexical-semantic retrieval (Goldman-Eisler, 1968). This indicates that naming problems in people with MTG lesions, at least to some extent, emerge due to lemma retrieval disruptions. Owing to the fundamental role of word retrieval in production (Levelt

et al., 1999; Dell et al., 1997) and to the fact that strokes caused by the left middle cerebral artery occlusion frequently result in damage to the temporal lobe (Hillis, 2007), it is important to understand how cognitive control modulates verbal behaviour observed in people with disrupted lexical-semantic processing when they are naming under distraction.

Assuming that damage to the MTG causes suboptimal activation and/or processing of lexical-semantic representations, Piai and Knight (2018) suggested that presence of competitors further complicates node selection, possibly, by allocating a similar degree of relevance to both the target and the distractor. This lesion-related noise might explain inflated error rates and slower response latencies, as well as bigger facilitation effects in the congruent condition when compared to neurotypical naming (Piai & Knight, 2018; Python, Glize, & Laganaro, 2018). Consistent with the similar degree of relevance idea are the findings of Janssen and colleagues (under review), who used behavioural and computational modelling approaches to investigate lexical-semantic competition in individuals with primary progressive aphasia. They found that patients with temporal lobe degeneration, which typically affects lexical-semantic processing, were overall slower than controls in the related compared to neutral condition (XXX on top of the picture). Furthermore, interference magnitude in the patient group depended on the activation strength of representations and the integrity of the ventral fibre tracts. When representations were weaker and the tracts were more damaged (i.e. supposedly not able to propagate the signal properly), the interference from the related distractor was significantly reduced, leading to relatively faster RTs.

Since cognitive control requires lower-level representations to operate on, it is unclear whether the higher-order mechanisms are able to stay computationally efficient when they receive noisy or conflicting information from the MTG. Previous research points to the possibility that the cognitive control networks might upregulate their activity to enhance impaired downstream lexical-semantic processing (Geranmayeh, Brownsett & Wise, 2014). However, as only a handful of studies have investigated naming with distractors in participants with well-defined temporal lesions, this assumption is rather speculative. None of these studies have directly tackled the issue of how noisy competing representations interact with top-down regulation, focusing instead on semantic priming effects in PWI (Python et al., 2018)and the contribution of frontal and temporal areas to the resolution of competitive lexical selection (Piai & Knight, 2018). Most importantly, to the best of our knowledge, no study has yet investigated how inhibitory control manages interference from noisy representations at the electrophysiological level.

Current study

The aim of the current study is to fill the gap in the literature by exploring activation patterns of inhibitory control networks in response to noisy informational flow from lower-level lexical-semantic representations under various degrees of interference. The main question is whether intact frontal cognitive control networks will upregulate their activity to facilitate disrupted lexical selection. In order to answer the question, previously collected electroencephalography (EEG) and behavioural data from six people with temporal lesions and six healthy controls were analysed. EEG was recorded while participants performed the picture-word interference task that included the semantically related, unrelated and congruent conditions. The target picture and the distractor word were presented simultaneously, and all distractors served as target pictures on other trials to increase interference (Piai, Roelofs, & Schriefers, 2012). Interference was further magnified by using the congruent condition as the baseline: introducing a word reading dimension to naming simplified the task and, therefore, highlighted the naming costs in related and unrelated conditions (Lowe & Mitterer, 1982).

The RTs and mid-frontal theta power activity were explored with two contrasts: related vs congruent condition (Stroop interference), and related vs unrelated condition (semantic interference). Error rates and RTs were reported, the latter analysed statistically to determine the presence of interference effects at the behavioural level. EEG data were analysed in the frequency domain to investigate power increases in the theta band (4-7 Hz) over the temporal window of interest (350-650 ms) within each participant group. To examine theta activity differences between patients and controls, the magnitudes of theta interference effects were compared between the groups. Analysing RT patterns together with power increases was crucial for interpreting theta activity as a function of interference. In young healthy adults, stronger interference leads to stronger conflict, which manifests itself as longer latencies, which, in turn, are associated with larger theta increases (Piai et al., 2014). Since people with aphasia have previously demonstrated inconsistent behavioural interference and facilitation patterns (Hashimoto & Thompson, 2010; Python et al., 2018, Piai & Knight, 2018) and no studies so far have confirmed that in older neurotypical populations conflict theta activity patterns follow those observed in young adults, determining interference magnitudes through RTs as a proxy and then establishing the presence of the electrophysiological Stroop and semantic effects within groups was a critical step for conducting the following between-group effect magnitudes analyses in a meaningful way.

In line with prior research (Shitova et al., 2017; Krott et al., 2019), controls were hypothesised to demonstrate the RT and electrophysiological response hierarchy related >

unrelated > congruent, with longest latencies and highest theta power increases in the related

condition and both Stroop and semantic interference effects present. For patients, however, related and unrelated conditions were predicted to produce similar RTs (i.e. no semantic interference) reflecting the similar degree of relevance given to the target and both types of distractors due to disrupted lexical-semantic processing (Piai & Knight, 2018; Janssen et al., under review). Stroop interference was hypothesised to be exaggerated compared to controls, as the identity distractor would facilitate naming by offsetting word retrieval deficits. As theta increases are driven by the degree of conflict, based on the similar RTs signalling potentially similar interference effects in the related and unrelated conditions, the same pattern related/unrelated > congruent was predicted at the electrophysiological level for the patients.

In terms of the between-group network activity differences, the magnitude of the mid-frontal theta effects in both Stroop and semantic interference were predicted to be significantly larger for controls. Patients were expected to show smaller power differences between conditions based on the findings that cognitive control networks upregulate their activity in people with aphasia when they perform tasks that do not demand increased executive control from neurotypical participants (Geranmayeh et al., 2014; for auditory language comprehension see Brownsett, Warren, Geranmayeh, Woodhead, Leech, & Wise, 2014). Thus, the compensatory overactivation in easier conditions was predicted to drive both unrelated and congruent conditions closer to the most challenging related condition, i.e. to decrease semantic and Stroop effect magnitudes compared to controls.

METHOD

The protocol was approved by the University of California, Berkeley Committee for Protection of Human Subjects, following the declaration of Helsinki. All participants signed written informed consent and received monetary compensation.

Participants

Two participant groups were formed from a subset of people who participated in the study by Piai & Knight (2018). All participants were native speakers of American English, without a history of psychiatric events, substance abuse or dementia. Only behavioural results have been previously published as part of a bigger sample (Piai & Knight, 2018).

In the current study, six participants with left lateral-temporal cortex lesions formed the patient group. Patients were 50 to 74 years old (mean age = 65, sd = 10), one was female. At the time of testing, months post-stroke onset ranged from 23 to 310 (mean time = 134

months, sd =112). All patients were premorbidly right-handed, with damage centred on the middle temporal gyrus (MTG, 100% overlap). Importantly, all six had spared medial superior frontal and dorsal anterior cingulate cortices. Individual lesion and overlap maps are shown in Fig. 1 (for lesion volume information see Table 1). All but one patient (P1) had been clinically diagnosed and tested on the Western Aphasia Battery (WAB; Kertesz, 1982). As reported by the researchers, P1, who had the smallest lesion in the group, retained language function after stroke and was able to maintain professional activity that included academic teaching. Only P6 had a relatively lower composite Aphasia quotient score (AQ=59.9) and demonstrated poorer performance on the naming subtest (Naming = 4.3) when compared to other patients (AQ ≥ 78, Naming ≥ 8.3). P6 was also the only patient showing language deficits associated with Wernicke’s aphasia (Table 1). The control group comprised six healthy right-handed participants, who were matched with the patients on age and years of education (four females; age range 54-74, mean age = 63.5, sd = 7.5; with 17.6 mean years of education, see Table 2).

Table 1

Individual patient information, including language testing data from the WAB and present study error rates.

m = male; f = female; Naming = WAB Naming and Word Finding score (maximum = 10); Aphasia Quotient (AQ, maximum = 100); WNL = within normal limit; MPO = months post stroke onset; NA = not assessed; MTG = middle temporal gyrus; STG = superior temporal gyrus; IFG = inferior frontal gyrus; MFG = middle frontal gyrus; SFG = superior frontal gyrus.

* This patient continued performing his occupation without problems, which included academic teaching amongst other tasks.

Table 2

Individual control group information, including present study error rates. age at the time of testing gender years of education Error rate, % C1 67 f NA 3.17 C2 68 f 16 2.38 C3 54 m 16 1.59 C4 57 m 22 0.79 C5 74 f 22 1.59 C6 61 f 14 4.76 m = male; f = female. age at testing gender MPO at testing MPO at WAB Aphasia type AQ (max. 100) Naming (max. 10)

Lesion volume, mm3 Error rate, % Total volume Temporal damage volume Frontal damage volume

P1 66 m 114 114 NA* NA* NA* 18.32 MTG 23.6

STG 34 IFG 0 MFG 0 SFG 0 2.38 P2 50 m 23 23 Conduction 77,9 8,6 93.75 MTG 50.4 STG 87.9 IFG 0 MFG 0 SFG 0 16.67 P3 56 m 72 72 Anomic 87,7 8,3 103.17 MTG 17.6 STG 33.7 IFG 21.4 MFG 3.9 SFG 0 26.98 P4 73 m 310 310 Anomic 92,9 9,5 85.82 MTG 82.6 STG 88.6 IFG 0 MFG 0 SFG 0 29.37 P5 74 f 230 230 WNL 94 8,6 36.95 MTG 56.3 STG 22.3 IFG 0 MFG 0 SFG 0 23.02 P6 70 m 55 55 Wernicke 59,9 4,3 79.68 MTG 76.9 STG 94.6 IFG 0.15 MFG 0 SFG 0 53.97

Figure 1

Individual lesion and overlap maps of the patients. (a) Individual lesion maps of the six left temporal cortex

patients (P1-P6). Each pair of images depicts a 3D model of the brain with individual lesion rendered on top (left), and then the axial view slice with cross hairs (right), marking the MTG (MNI coordinates -63, -29, 0).

(b) Lesion overlap maps. The colour scale shows the degree of overlap in the lesion location, with turquoise

indicating 0% overlap (i.e. only one patient had a lesion in the region) and red indicating 100% overlap across all the patients. The cross hairs on the coronal (top left) and axial view slices (bottom left) indicate the MTG with the same coordinates as in (a). MTG = middle temporal gyrus; L= left; R= right.

(a)

Design and materials

Fifty-six colour pictures were selected from the BOSS database (Brodeur, Dionne-Dostie, Montreuil, & Lepage, 2010). They belonged to fourteen semantic categories (e.g., tools, animals, fruit, body parts, musical instruments, etc.), each category represented by four objects. The same fifty-six picture label words were used as distractors. In the congruent condition, the distractor word was identical to the picture name (PIG on the picture pig, Fig. 2). In the related condition, one of the other three picture label words from the same semantic category served as a distractor (COW with pig). Finally, in the unrelated condition, distractors were words from other semantic categories in the response set that were semantically and phonologically unrelated to the target (CHAIR with pig). Each picture was shown to the participant once in each condition, thus making up the total of 168 trials, 56 per condition. The trials were randomized using Mix (van Casteren & Davis, 2006), with one unique list per participant. Participants were given instructions to name the picture and ignore the distractor word, with the emphasis on both speed and accuracy.

Figure 2

Example of the experimental picture-word item in three conditions. Left to right: picture-word items used for

congruent (distractor word PIG matching the picture), related (distractor word COW from the same semantic category) and unrelated conditions (distractor word CHAIR from a different semantic category).

Procedure and EEG data acquisition

The presentation of the experimental stimuli and the recording of responses were controlled by Presentation Software (Neurobehavioral Systems). Vocal responses were recorded with a microphone and evaluated online by the experimenter. Participants were seated in front of a computer monitor. Each trial began with a fixation cross presented for 1s, followed by the presentation of a picture-word item for 2s. The stimulus appeared on a white background in the centre of the screen with the distractor word superimposed on top of it in white. The inter-trial interval was programmed to vary between 1.25 and 2s, and participants were instructed to blink if necessary during this time after the trial. Due to the concerns that

patients might have different memory capacity that could confound the results, no familiarisation phase was introduced.

EEG was recorded from 64 Ag/AgCl active scalp electrodes with the Biosemi system (BioSemi, Amsterdam, the Netherlands) at the sampling rate of 1024 Hz. The electrodes were mounted in an elastic cap according to the extended 10-20 system. The horizontal electrooculogram (EOG) was recorded from electrodes on the left and right temples, and the vertical EOG from Fp1 and the electrode positioned below the left eye. To register muscle movement, electromyogram was recorded from the orbicularis oris muscle, with one electrode placed on the left upper and the other on the right lower corner of the mouth. Behavioural data analysis

Fourteen out of fifty-six pictures were excluded from the analyses due to low naming agreement or confusion with other target words. This yielded 42 trials per participant per condition, 126 trials in total. Dysfluent and erroneous responses were coded online as incorrect, and their corresponding trials were excluded from the response time (RT) and electrophysiological analyses. The following responses were classified as errors: (1) naming the distractor word, (2) hesitations (starting the response with filled pauses like ‘hum’ or poorly articulated initial phonemes), (3) no response, (4) phonological paraphasias, (5) semantic paraphasias, and (6) using another name for the picture (e.g., dish instead of bowl).

Naming RTs were calculated manually using Praat (Boersma & Weenink, 2013) before separating trials by condition. Single-trial RTs were analysed with linear mixed-effects models (Baayen, Davidson, & Bates, 2008), which were fitted with the lme4 package (version 1.1-19; Bates, Maechler, Bolker, & Walker, 2015) in R (RCoreTeam, 2015). RTs were log-transformed to meet the normality of residuals requirement. The effects for “group” (controls and patients), “condition” (related, unrelated, congruent) and their interaction were included as fixed terms, with random intercepts for participants and items. The final model was:

log(RTs) ~ group * condition + (1 | participant) + (1 | item).

More complex models with random slopes failed to converge or produced a singular fit. Factor levels were successively tested against each other with repeated contrast coding (package MASS, Venables & Ripley, 2002): (a) related with congruent condition, i.e. main effect of Stroop interference; (b) related with unrelated condition, i.e. main effect of semantic interference; (c) controls with patients, i.e. main effect of group; (d) interaction between groups regarding the Stroop interference; (e) interaction between groups regarding the semantic interference. Significance of effects was calculated using the Satterthwaite

approximation (package lmerTest, version 3.1.1, Kuznetsova, Brockhoff, & Christensen, 2016). To investigate differences between conditions within groups, a post-hoc Tukey test was conducted (package multcomp, version 1.4-13, Hothorn, Bretz & Westfall, 2008). EEG data analyses

The analyses were conducted in MATLAB 2019b using FieldTrip version 20200220 (Oostenveld, Fries, Maris, & Schoffelen, 2011). The data were segmented into epochs time-locked to the picture-word item presentation, defined from 500ms pre- to 800ms post-stimulus onset when the participants were about to start articulation. Electrodes were re-referenced offline to the average of the mastoids and a baseline correction was applied, with the entire pre-stimulus 500ms interval subtracted from the signal. Only accurate trials were processed further. Based on previous conflict theta findings (Shitova et al., 2017; Krott et al., 2017), the region of interest (ROI) comprised the electrodes F1, Fz, F2, FC1, FCz, FC2, C1, Cz and C2. The channels surrounding the ROI were also included in the artefact cleaning procedure to be later used for channel repair, see below for more information (Fig. 3).

Since a significant amount of data contained eye blinks, independent component analysis (ICA) was implemented to avoid substantial data loss (Jung et al., 2000; as implemented in FieldTrip). First, all 64 channels were manually inspected to exclude noisy or broken electrodes pre-ICA. The independent components for eye blinks, eye movement and muscle activity were removed. Next, noisy and drifting electrodes in the ROI were interpolated across trials for several participants. Finally, all trials with remaining excessive noise, with electrodes that could not be repaired and trials from P1 and C3 with eye blinks (which did not undergo ICA as less than 5% of their data contained blinks) were discarded. As a result, artifact-free datasets for controls and P1 included 111 - 122 trials, distributed nearly equally between three experimental conditions (about 40 per condition, ranging from 35 to 42). In the patient group, P6 was a notable exception with only 53 trials available for analysis (7 related, 13 unrelated and 33 congruent). Other patient datasets contained 95 trials on average (21-30 related, 24-32 unrelated and 40-44 congruent trials).

The within-group analyses of differences between conditions and the between-group analyses of differences between effect magnitudes were conducted on the time-averaged power spectra in the window of interest (250-650ms). Power spectra between 1 and 20 Hz were calculated for each condition averaged over trials per participant within the groups using the fast Fourier transform with the Hanning taper to control for spectral leakage. Next, keeping the individual data for the repeated measures within-group statistical analysis, the spectra were averaged per condition within each participant group. For the between-group

analyses, the magnitudes of the Stroop and semantic interference effects were calculated by subtracting the power values of congruent and unrelated conditions from the related condition power values separately for patients and controls.

Figure 3

Biosemi64 electrode montage based on the 10-20 system with marked electrodes of interest. The channels

marked in light-blue and green are the electrodes forming the region of interest for the mid-frontal conflict theta activity. The channels marked in yellow were used for channel repair.

Within- and between-group statistical tests were performed using a non-parametric cluster-based permutation approach (Maris & Oostenveld, 2007). The tests were performed for the frequency range of 4 to 10 Hz. For the within-group analysis, the two-dimensional samples (frequency x channel) were scanned for frequencies and channels showing a difference between two conditions with two-tailed dependent-samples t-tests (threshold alpha = .05). The samples with significant p-values were clustered on the basis of spatial-frequency adjacency. Then, a cluster-level statistic was calculated by taking the sum of the t-values within each cluster. The statistical significance of these clusters was estimated with a Monte Carlo method. The algorithm created a permutation distribution by: 1) repeatedly randomly dividing the collected data into two conditions; 2) performing the clustering procedure described above for each partition; and 3) entering the cluster with the largest summed t-values into the permutation distribution. The number of repetitions of the random partitioning procedure was set to 1000 to allow the biggest possible number of draws for producing the maximum number of unique permutations. Finally, both positive and negative (two-tailed) cluster-level statistics from the observed data were compared to the resulting random permutation distribution. The Monte Carlo p-value estimate with a critical alpha level of .05 was the proportion of random partitions with the cluster test-statistic more extreme than that in the observed data. A Monte Carlo p-value smaller than alpha would mean that the compared conditions significantly differed from each other. The differences

of effect magnitudes between groups were analysed analogously, only using the independent samples t-test as the magnitude values were coming from two separate groups.

RESULTS

Error rates

Controls made few errors (18/756 trials, or 2.38 % of total responses), while the patient group had considerably higher error rates (192/756 trials, or 25.4% of total responses, see Table 3). Just as for controls, the congruent condition was the least challenging (5% of errors), while 53% and 42% of incorrect trials fell on the related and unrelated condition, respectively. It should be noted that P1, who had the smallest lesion load, committed very few errors, performing at the level of controls. Conversely, P6, who demonstrated most serious functional limitations, made almost a third of the errors in the patient group, which is considerably more than the other patients (21-37 errors). Error types were relatively equally spread across the categories in the control group, whereas the patient group gravitated towards specific types: most frequently to hesitations (42%) and non-responses (19%). Patients produced hesitations, non-responses and paraphasias mostly in the related (47/80) and unrelated (30/80) conditions.

Table 3

Error rates and types. (a) The distribution of error rates across conditions. The number of errors produced in

each condition out of 756 trials in the patient and control groups. Error rates are provided in percentages. (b) The distribution of error types across groups calculated as percentages from the total amount of errors.

amount of errors (error rates from total responses, % )

(a) related unrelated congruent Total error

rate controls 8 1% 6 0.8% 4 0.5% 18 / 756 2.38% patients 102 13.5% 80 10.6% 10 1.3% 192 / 756 25.4% (b) Alternative name Naming distractor word Hesitations No response given Perseverance Phonological paraphasia Semantic paraphasia controls 27.8% 16.7% 27.8% 11.1% - - 16.7% patients 0.5% 8.9% 41.7% 18.8% 1% 16.1% 13% Response times

Descriptively, classic Stroop and semantic interference patterns were present in both groups (related > unrelated > congruent, Fig. 4). Compared to controls, the magnitude of both the semantic and Stroop interference effects for patients was twice as large (Table 4). Even after

controlling for the general slowdown of 139ms in the patient group, facilitative effects of the congruent condition in relation to naming with an unrelated distractor were three times bigger for patients. Controls showed gradual increase of RTs in relation to the congruent baseline, on average being 7% and 14% slower in the unrelated and related conditions, respectively. Patients, on the other hand, were almost equally slower in the unrelated (24%) and related (26%) conditions.

Figure 4

Individual-averaged and group-averaged response times in seconds for the control and patient groups across the experimental conditions. Individual-averaged in grey and group-averaged in black. CONG = congruent;

REL = related; UNREL = unrelated.

Table 4

Group-averaged response times across the three conditions and between-group differences in the Stroop and semantic interference effects. Naming cost shows the facilitative effect of having a congruent distractor

compared to naming pictures in the unrelated condition.

mean RTs (sd)

related unrelated congruent Stroop

interference Semantic interference Naming cost controls 1188ms (294) 1094ms (220) 1019ms (251) 169ms 94ms 75ms response slowdown 14% 7% baseline patients 1555ms (361) 1516ms (379) 1158ms (268) 397ms 39ms 358ms response slowdown 26% 24% baseline*

* Naming in congruent condition in the patient group was 139ms slower than in the control group. RTs = response times; sd = standard deviation, ms= milliseconds.

Results of the statistical analyses revealed a significant main effect of group (t(10)= 3.57, p=.005), with patients being overall slower; as well as main effects for Stroop and semantic interference (t(1249)=19.6, p< .001 and t(1247) = 4.37, p< .001). Interaction effects between group and condition were found to be significant for Stroop interference (t(1248)= -6.97, p<.001) but not for semantic interference (t(1246)=1.26, p=.2). The post-hoc Tukey test revealed that while the Stroop and semantic interference effects were significant for

controls (z=3.57 and z= 4.28, respectively, p-values < .001), patients showed the Stroop effect (z=3.8, p<.001) but no significant differences between related and unrelated conditions (see Table 5).

Due to a considerable misbalance in RT data as patients made more errors, particularly in the related condition, maximum likelihood ratio tests were conducted to check for significance of effects using the anova function (lme4 package, Bates et al., 2015). Predictor variables and their interactions were manually coded using repeated contrasts, and the full linear mixed model was compared to five null models. In each null model, one fixed effect of interest was omitted, thus allowing to estimate whether its inclusion significantly improved the full model. The omitted terms were the main effects of (1) group, (2) Stroop interference, and (3) semantic interference, as well as the interaction effects of (4) Stroop and (5) semantic interference. The likelihood ratio tests confirmed all significant effects, including the interaction effect for semantic interference (Χ² =19.04, p< .001) which was not

reported as significant in the lmer output.

Table 5

Results of the statistical analyses for response times: fixed and random effects Fixed effects main and interaction effects β SE t(df) p Group (patients vs controls) 0.254 0.07 3.57 (10) 0.005 Semantic interference (related vs unrelated) 0.055 0.013 4.37 (1247) < 0.001 Stroop interference (related vs congruent) 0.24 0.012 19.57 (1249) <0.001 Semantic interference (between-group difference) -0.031 0.025 1.26 (1246) 0.2 * Stroop interference (between-group difference) -0.17 0.024 -6.97 (1248) <0.001 within-group effects ** β SE z p Controls: semantic interference 0.309 0.07 4.28 < 0.001 Controls: Stroop interference 0.253 0.07 3.57 < 0.001 Patients: Semantic interference 0.17 0.08 2.06 0.13 Patients: Stroop interference 0.285 0.075 3.8 < 0.001 Random effects Variance SD Item Intercept 0.002 0.042 Subject Intercept 0.014 0.121 Residual 0.03 0.125

β = slope; SE = standard error; t = t-value; df = degrees of freedom; p = p-value; z = z-value; SD = standard deviation.

* Due to misbalance in the data, the linear mixed model output was checked with the maximum likelihood ratio tests, which did reveal significant semantic interference differences between the groups.

Time-averaged spectra results

Although theta activity was present across all conditions (Fig. 5), cluster-based permutation tests conducted on the power spectra within groups did not reveal any statistically significant differences between theta increases in the conditions in either group (Fig. 6). Analogously, between-group analyses conducted on the power spectra of differences between conditions, i.e. semantic and Stroop interference magnitudes, did not show any statistically significant effects. It should be noted, however, that the test comparing the Stroop interference magnitudes between controls and patients returned only 7% of permutations more extreme than the observed data values, thus approaching the critical alpha level of .05 and suggesting that Stroop interference was borderline bigger in the control group.

Figure 5

frequency representations of conditions averaged across trials and participants within groups.

Time-frequency representations (TFRs) for each condition within groups on the average of 9 mid-frontal electrodes show the presence of theta activity in the time window of interest. The TFRs were calculated by convolving the EEG data with a complex wavelet using the Hanning taper at the frequency range 1 - 30 Hz with a sliding time-window of three cycles’s length, advanced in steps of 50ms in the time and 1 Hz in the frequency dimensions. Relative baseline correction was applied using the entire 500ms pre-stimulus interval (i.e. the baseline values were subtracted from the duration of the trial, then divided by the same baseline activity).

Figure 6

Time-frequency representations of semantic and Stroop interference effects within groups. Time-frequency

representations on the average of the mid-frontal electrodes for the relative differences (in % change) between related and unrelated conditions (semantic interference) and between related and congruent conditions (Stroop interference) within groups. No baseline correction was applied.

DISCUSSION

The aim of the present study was to investigate activation patterns of inhibitory control networks under interference of various strength, in particular, in disrupted lexical-semantic processing caused by temporal-lobe lesions. Participants from control and patient groups named pictures with lexical distractors in three conditions. The main hypothesis was that, in patients, intact frontal cognitive control networks will upregulate their activity in order to enable efficient processing of noisy or weak lexical-semantic representations (Geranmayeh et al., 2014). Interference magnitudes were inferred from response times (RTs), while network activity was measured via midline frontal conflict theta power increases (4-7 Hz). In line with previous research (Python et al., 2018), patients with middle temporal gyrus (MTG) lesions committed considerably more errors than controls, with the related condition of the picture-word interference task being most challenging. Hesitations were the most common error in the patient group, supporting the observation that damage to the MTG results in lexical-semantic retrieval deficits (Goldman-Eisler, 1968; Indefrey, 2011).

Descriptively, lesion volume in the superior and middle temporal gyri was connected to error rates: the patient with the smallest load performed at the same level of controls, while those with biggest load made the most errors.

In Stroop-like paradigms, RTs serve as a proxy for measuring the strength of competition elicited by various distractor types: the more interference at the lexical-sematic stage of word planning, the more time required for resolving it and selecting the word consistent with communicative goals (Glaser & Düngelhoff, 1984). Overall, patients were significantly slower than controls across conditions, including the congruent one (for similar findings in aphasic patients, see Hashimoto & Thompson, 2010; Piai & Knight, 2018), likely owing to lesion-related cognitive processing disruptions linked to grey or white matter damage (Turken, Whitfield-Gabrieli, Bammer, Baldo, Dronkers, & Gabrieli, 2008; Price et al., 2016). Significant Stroop and semantic interference effects were present in the control group, demonstrating that distractor manipulations produced interference which gradually increased from the congruent to unrelated to related conditions. The patients, on the other hand,showed Stroop but not semantic interference: unlike naming in the congruent condition prompted by identity distractors, naming in both related and unrelated conditions was similarly difficult for people with MTG lesions. One way to account for these results is by adopting the executive control approach. Based on the naming cost, i.e. the difference between the congruent and unrelated conditions, which was exaggerated threefold in the patient group, it is reasonable to hypothesise that as both these conditions were equally more difficult for the patients, they both required more inhibitory control, with longer selection process leading to slower RTs. Alternatively, instead of being a by-product of cognitive control processing, longer latencies can be attributed solely to lesion impact, i.e. stem from the local processing slowdown. This implies that, while interference control still remained effective, slower naming was caused by lexical-semantic retrieval deficits. As both retrieval and executive control issues lead to longer RTs and exaggerated naming costs, at this point, disentangling one from the other and saying with certainty which underlying process levelled the two conditions is unfeasible.

General word retrieval deficits arising from weaker or noisier representations might mean that lemma retrieval becomes equally unstable for both targets and all types of distractor words. Consequently, a similar degree of relevance allocated to targets and distractors does not only slow down naming due to increased difficulty at the selection stage (Piai & Knight, 2018), but is also likely to erase the interference magnitude differences between conditions. This idea is consistent with Janssen and colleagues’ (under review) results from patients with primary progressive aphasia (PPA) which is linked to temporal lobe neurodegeneration. They found that interference magnitude in the related condition depended both on the integrity of representations (potentially affecting the strength of

activation, represented by input strength in the model) and integrity of the ventral fibre tracts (affecting the signal propagation capacity, represented by input duration in the model). The interaction of weaker activation and inability of the tracts to properly propagate the signal led to reduced interference from the semantically related distractors and faster RTs. Unfortunately, there is no information on the ventral fibre tract integrity of the patients from the current study, as well as no data from Janssen and colleagues on the unrelated and congruent conditions which were not included in the experiment. Future studies should consider adding various distractors as well as a measure for white matter integrity to shed more light on the relations between the activation strength, fibre tract integrity and intensity of competition at the lexical selection stage. Furthermore, it is worth mentioning that PPA is typically associated with atrophy to a number of regions such as the anterior temporal lobe, posterior temporal and inferior parietal structures (Gorno-Tempini et al., 2011), which means that comparisons between people with PPA and acquired aphasia require caution.

Another source of atypical interference, apart from the input and signal propagation strength, might be the conflict monitoring/signalling nodes of the cognitive control networks, registering either upregulated interference from the unrelated distractors, attenuated interference from the related ones, or even not detecting their presence at all. For instance, lack of semantic interference has been previously reported in patients with damage to the prefrontal cortex (PFC,Piai, Riès, & Swick, 2016; Piai & Knight, 2018), the area considered critical for supporting goal-directed behaviour (Miller & Cohen, 2001). Another potentially relevant node in the control network is the inferior parietal lobule (IPL). In adults, it has strong functional connections with the PFC and, when attentional demands grow, it gets activated just before the frontal cortex. Based on these observations, some suggest that it is the parietal areas that detect and signal conflict to the PFC which then exerts top-down control (Dosenbach et al., 2007; Liston, Matalon, Hare, Davidson, & Casey, 2006). If, indeed, damage to the IPL disrupts conflict monitoring, interference may stay undetected, thus, levelling the related and unrelated conditions. It might be this mechanism that has driven the lack of semantic interference in the current study as half of the patients had lesions spreading into the parietal lobe. It should be mentioned, however, that lack of semantic interference in people with PFC and IPL lesions is likely to be of a different origin as these two nodes seem to be serving dissociable functions (Liston et al., 2006).

Further investigations of lexical interference should focus on understanding interactions between the nodes of the cognitive control network and simultaneously damaged temporal regions associated with lexical-semantic processing. In particular, the potential

differential influence of temporal and temporo-parietal lesions on behavioural performance should be taken into account. Moreover, careful experimental design is required to determine whether the interference irregularities are directly contributed to by the fronto-parietal network damage (such as IPL lesions leading to inability to detect conflict or PFC lesions disrupting effective control implementation), or, alternatively, arise from the activation and signal propagation deficits at the lower processing level in the temporal lobe (MTG and ventral fibre tracts lesions).

With regard to electrophysiological Stroop and semantic interference effects, although theta power increases in the 350-650ms window of interest were present upon visual inspection in both groups (Fig. 5), the activation differences between conditions within groups were not statistically significant. Lack of significance may be attributed, first of all, to insufficient data that made finding robust effects problematic. EEG studies that found theta power differences analysed over a hundred trials per condition per participant (Shitova et al., 2017; Krott et al., 2019), whereas in the current study there were forty or fewer trials per person per condition, especially few in the related condition which is crucial for both interference effects. Related to the insufficient number of trials is a rather modest sample size. Since power is strongly affected by a number of factors such as skull thickness, cerebrospinal fluid volume, distance between electrodes or degree of arousal (Klimesch, 1999), the small sample size made accounting for the variance in the EEG signal and detecting the effect more difficult.

Secondly, the effects might have been attenuated by individual variation in spectral boundaries. Klimesch (1999) argued against the use of conventionally determined frequency bins and proposed adjusting the window for theta separately for each participant by anchoring individual alpha peak frequency (IAPF) at the average frequency of highest power between 6-13 Hz. The IAPF seems to be a stable, heritable characteristic, that increases from early childhood to adulthood and then starts declining, with averages of 9.8-10.5 Hz in young adults and 8.5-9.7 Hz in participants over sixty (Grandy, Werkle-Bergner, Chicherio, Schmiedek, Lövdén, & Lindenberger, 2013; Posthuma, Neale, Boomsma, & De Geus, 2001; Dustman, Shearer, & Emmerson, 1993). Individual alpha peak modifies the transition frequency between theta and alpha bins, meaning that maximal power in theta frequency varies as a function of alpha (e.g., Hogan, Swanwick, Kaiser, & Rowan, 2003, calculated the theta band to be 3-5 Hz for older participants; see Fig. 7 for power spectrum plots from the present study). Connected to the IAPF might be the concept of background activity slowing, considered by some a neurophysiological sign of healthy ageing (Rossini, Rossi, Babiloni,

& Polich, 2007): resting state delta, theta and alpha activity are often found in lower frequency bins in older participants (Woodruff & Kramer, 1979; Duffy, Albert, McAnulty, & Garvey, 1984). However, it is worth noting that this slowing has also been argued to reflect hidden health issues such as cardiovascular disease or diabetes (Duffy et al., 1984).

There are other possible age-related explanations for the lack of predicted effects. Most interference-related theta activity literature is based on young adults, and there are reasons to believe that neurophysiology of conflict control might differ for older people. Healthy cognitive ageing has been associated with reorganisation of the frontal brain regions, which undergo neurochemical, anatomical and functional alterations (Cabeza, Daselaar, Dolcos, Prince, Budde, & Nyberg, 2004; Bäckman, Lindenberger, Li, & Nyberg, 2010), including significant hypometabolism in the ACC and medial PFC regions (Pardo et al., 2007). Moreover, some functions seem to become subserved by more distributed networks, possibly to support processing in the ageing brain. High-performing older adults often recruit areas that are not engaged by their low-performing peers and young participants, such as right hemisphere homologues for cognitive control augmentation or additional temporo-parietal semantic regions for word naming (Cabeza, Anderson, Locantore, & McIntosh, 2002; Hoyau et al., 2017). Changed functional and structural organisation of the cognitive control networks in older participants may lead to consequent alterations in the neurophysiological circuitry and/or the ways underlying computations are performed. These processes are likely to affect the electrophysiological signatures of the effect of interest, differentiating them from those of younger adults in terms of power, both in raw microvolts and relative to other conditions.

Regarding absolute event-related theta power1, older participants showed attenuated fronto-parietal theta increases in comparison to younger people on the auditory memory task, possibly due to decreased efficiency of memory encoding (Karrasch, Laine, Rapinoja, & Krause, 2004), and on delayed choice reaction time tasks, hypothesised to reflect weaker involvement of hippocampal-cortical and thalamo-cortical circuits, traditionally linked to cognitive performance (Babiloni et al., 2004). Mid-frontal memory-related absolute theta also demonstrated interesting behaviour across experimental conditions: while young people showed gradual increase of theta activity with increasing memory load, highest in the hardest condition, older adults had highest power in conditions with less memory load (McEvoy,

1_{As most research on age-related theta changes was conducted in the memory domain and there have}

been no analogous studies of conflict response theta, the papers cited below mainly concern the working and episodic memory.

Pellouchoud, Smith, & Gevins, 2001; Kardos, Toth, Boha, File, & Molnar, 2014). Widagdo and colleagues (1998), who investigated mid-frontal and central theta power increases during arithmetic tasks, found even more radical changes in the older group who showed not power increases but decreases in relation to the resting state baseline. Similar paradoxical patterns were observed in the control group in the current study: upon visual inspection of the power spectrum plots, theta power increases at the group level were most prominent in the congruent condition, followed by smaller power values in the unrelated condition and even smaller in the related one (Fig. 7). This might indicate that cognitive networks in healthy older participants show a differential pattern of activation not only in memory tasks but also while resolving interference.

Figure 7

Power spectrum plot in the 350-650ms time window on the average of mid-frontal electrodes for controls and patients The absolute power values are displayed in microvolts squared on the y-axis, the frequency bins are

on the x-axis. Unrel = unrelated; cong = congruent; cond = condition.

Nevertheless, there are some limitations to this reasoning. Firstly, returning to the factors that affect power such as scalp density mentioned by Klimesch (1999), unless recruiting a considerable number of participants, using absolute power values for between-group comparisons might be misleading as they may be reflecting the impact of individual variability and not the experimental effects per se. Secondly, the studies which found attenuated theta in older participants (apart from Hogan et al., 2003) did not use IAPF to adjust the frequency band boundaries, thus, making it plausible that decreased power was the consequence of imprecise frequency window choice. Finally, the findings from memory-related theta should not be directly used for interpreting the current results, as the conflict theta associated with verbal interference is known to be supported by separate neural circuits (Cohen, 2014). Moreover, while memory processing is known to become worse with age

(Dixon, Wahlin, Maitland, Hultsch, Hetzog, & Bäckman, 2004; Mattay et al., 2006), leading to limited capacity for goal selection and maintenance in working memory (for review see

De Jong, 2001), whether inhibitory control is necessarily impaired remains an open question. Performance of older participants often depends on the task and type of inhibition involved. For instance, age-related changes were found to negatively affect inhibition of automatic but not of dominant responses (Andrés & Van der Linden, 2000). This suggests that some but not all inhibitory mechanisms may, indeed, be subject to cognitive ageing – a factor that should be considered in interpreting theta activity results.

Concerning between-group differences in cognitive control activity, to avoid using potentially unreliable direct raw power comparisons, instead, the magnitudes of power differences between conditions (i.e. electrophysiological semantic and Stroop effect sizes) were contrasted. The effect sizes between patients and controls did not prove to be statistically different, the result likely driven by lack of robust effects within groups in the first place. Interestingly, the Stroop effect in controls was borderline bigger than in patients, with only 7% of permutations more extreme than the observed data values. This points to the probability of the presence of activation differences between controls and patients with MTG lesions, which should be further investigated with a more robust participant sample size and increased number of trials per condition. In addition, future studies should address the challenge of between-group comparisons of electrophysiological activity. Although previous research mostly compared absolute power values directly (e.g. between older and younger participants: McEvoy et al., 2001; between healthy older participants and people with Alzheimer’s disease: Hogan et al., 2003), development of more optimal approaches is desirable, since variability in raw power is often attributable to biological and technical instead of factors of interest (Klimesch, 1999). Effect magnitudes used in the present study or, alternatively, magnitude ratios which describe the effects within groups in relative terms, might serve as a less confounded measure for network activation comparisons between groups.

Future research examining inhibitory control in older populations should take into account a number of age-related neurological changes such as the oscillatory slowing in lower frequency bands, the impact of individual alpha peak frequencies on band boundaries, the more distributed nature of cognitive control and the possible consequent alterations of its electrophysiological signatures. The major challenge is understanding the conflict theta patterns in neurotypical older participants and linking them to behavioural performance and

interference effects. Apart from that, better approaches to comparing electrophysiological network activity between groups of participants should be adopted.

In conclusion, despite a number of limitations, this study has been the first to report electrophysiological signatures of cognitive control in older neurotypical people and people with acquired aphasia naming pictures under conditions of interference. Participants demonstrated patterns different from those in younger adults, which indicates that conflict control mechanisms may be changing with age. Moreover, the study has raised some theoretically and practically relevant questions about the contribution of top-down control and lower-level processing to the efficiency of word production, particularly in cases when lexical-semantic selection is disrupted. In order to understand these mechanisms, in future, precise experimental design and bigger participant samples are required. Additionally, the study has highlighted the links between cognitive ageing and aphasiology research and the importance of collecting multidimensional data to enable meaningful interpretations. As the interest towards the role of domain-general systems in language use seems to be increasing, both the results and lack thereof in the current thesis may provide some ideas for further investigations.