Macrolinguistic function in temporal lobe epilepsy: a reinterpretation of circumstantiality

ABSTRACT

Background

While language is frequently explored in stroke and focal lesions, complex language disorders in paroxysmal conditions remain relatively underexplored. Individuals with temporal lobe epilepsy (TLE) often have an impairment of conversational language manifesting as verbosity and attributable to disruption of cognitive-linguistic networks. The micro- and macrolinguistic underpinnings of this disturbance, and the role of epilepsy and cognitive variables, are yet to be explored.

Methods and Procedures

We examined the elicited language output of 16 individuals with TLE and 14 healthy controls under separate monologic discourse tasks: a structured and constrained context, elicited by description of the ‘Cookie Theft’ picture, and an unstructured, unconstrained context, elicited by description of a ‘Typical Day’. We hypothesised that language output in the unstructured context would be characterised by verbosity to a greater extent than language elicited in a structured context.

Outcomes and Results

Following transcription and coding, detailed multi-level discourse analysis suggested that a constrained context gives rise to microlinguistic disturbances in individuals with TLE, reducing fluency, with more pauses and fillers. Under an unconstrained context, as anticipated, classical aspects of verbosity emerge in those with TLE, manifesting as longer speaking time, a longer duration of pauses, and a higher proportion of repetitive or redundant statements. Macrolinguistic elements such as coherence and informativeness are widely impacted, particularly disturbing language formulation. Correlations suggest that microlinguistic disturbances are closely linked with the immediate impact of seizures on cognitive-linguistic function, while macrolinguistic disturbances are more broadly impacted by disorder severity and word retrieval deficits.

Conclusions

These findings suggest that cognitive-linguistic disturbances in TLE produce vulnerability to different psycholinguistic impairments as a function of differing linguistic challenges imposed by constrained and unconstrained discourse contexts. Where constrained, output fluency is primarily affected, while in an unconstrained context disturbed discourse planning and organisation declares itself. We conclude that these patterns reflect a dynamic linguistic system taking shape under specific contextual conditions.

KEYWORDS:

• discourse
• circumstantiality
• verbosity
• language
• context
• temporal lobe epilepsy

Introduction

While the exploration of language in stroke and focal lesions has received much attention, complex language disorders in paroxysmal conditions remain relatively underexplored and poorly understood. Temporal lobe epilepsy (TLE) is an example of one such condition, and represents the most common focal symptomatic epilepsy among adults (Jaimes-Bautista et al., 2015, Tellez-Zenteno & Hernandez-Ronquillo, 2011). Individuals with TLE exhibit characteristically verbose language (Waxman & Geschwind, 1975). This pedantic, repetitive, highly-detailed, and peripheral style is termed ‘circumstantiality’ and was initially considered a personality feature of TLE (Bear & Fedio, 1977; Waxman & Geschwind, 1975). It is now better conceptualised as a non-lateralised neurocognitive phenomenon attributable to subtle interictal disruptions of linguistic function. While language impairments in TLE are commonly identified clinically, proffered as a cognitive complaint, or detected on neuropsychological assessment at a single-word level (Bartha et al., 2005; Dutta et al., 2018), there is limited understanding of how these characteristic impairments relate to dysfunctional discourse.

Circumstantiality in TLE has been proposed to serve as a compensatory mechanism to overcome instances of word-finding difficulties (Brandt et al., 1985; Mayeux et al., 1980). This is disputed by Field et al. (2000) who posit that lexical-processing deficits in left TLE do not account for circumstantiality, suggesting that micro- and macrolinguistic processes are dissociable, being those that relate to lexical-syntactic and suprasentential processes, respectively. The microlinguistic level relates to connections which operate within the level of the sentence and relate to the fluency-lexical access dimension. These functions relate to lexical processing to organise phonological patterns into morphological forms and words, and in turn allow for sentence formation and processing. The macrolinguistic level encapsulates between-utterance organisation and discourse processing to connect utterances cohesively and coherently (Kintsch & van Dijk, 1978). Discourse processing deficits in narrative and conversational contexts are also found in right TLE, and are dissociable from lexical-syntactic impairments (Lomlomdjian et al., 2017). Similar patterns of macrolinguistic dysfunction are present following right hemisphere damage, particularly accompanying anterior lesions (Marini, 2012). While their microlinguistic function appears intact, impairments emerge in coherence, producing more tangential output (Marini, 2012). Right frontal regions, as components of complex networks, may therefore play a role in organising information in narrative discourse (Marini, 2012). Disrupted projections from temporal to frontal regions in TLE (Stretton & Thompson, 2012) might engender similar patterns as those observed by Marini (2012).

Fluency, cohesion, and coherence have been largely overlooked by TLE researchers to date. In this context, fluency, a microlinguistic element, relates to intrasentential or message-level transitions, and describes the rate of language production, rather than psychometrically-evaluated verbal fluency on tasks of orthographic lexical retrieval or semantic fluency (Wood et al., 2001). Individuals with left TLE tend to pause longer than those with right TLE (Howell et al., 1994), and regardless of laterality, they use more noncommunicative fillers, abandon trains of thought, and produce more repetitions than controls (Bell et al., 2003). Despite more frequent fluency disruptions, they are comparable with healthy controls in their rate of output, that is, words per minute (Field et al., 2000; Howell et al., 1994). Cohesion and coherence are macrolinguistic elements. Cohesion describes relatedness of meaning across sentences by making explicit and unambiguous connections to previously introduced content. Cohesive devices include conjunctions and personal referents; for example, Sam was hungry. He made dinner (Halliday & Hasan, 1976). Cohesion relies on simultaneously monitoring output and the listener’s understanding which depends upon working memory (Mozeiko et al., 2011) and lexical retrieval (Coelho et al., 2005). Global coherence relates to discourse organisation above the sentence-level, that is, how content is connected thematically to serve a particular goal or plan (Glosser & Deser, 1991).

Demands on language production are thought to be influenced by type of elicitation (Fergadiotis & Wright, 2011; Stark, 2019). Structured expository elicitation tasks, such as describing a cartoon, provide a clear discourse topic and adequate linguistic scaffolding to facilitate discourse planning and limit disturbances, while narrative tasks are thought to elicit more complex language reliant on memory and macrolinguistic structures including grammar and coherence (Macwhinney et al., 2010). Narrative or conversational tasks are therefore sensitive to evaluating impairments in productivity, vocabulary breadth and content richness (Stark, 2019). In a structured expository elicitation task, participants with TLE were not verbose when describing a six-frame cartoon, and instead were comparable to healthy controls in the number of words and output duration, but were less fluent overall (Bell et al., 2003). There was no evidence to suggest that these disturbances to discourse related to laterality of seizure onset (Bell et al., 2003). Using the structured ‘Cookie Theft’ task (Goodglass & Kaplan, 1972), Hoeppner et al. (1987) reported verbosity in four of nine participants with left-localised focal impaired awareness seizures (FIAS). Rather than verbosity being a broad diagnostic feature in TLE, structure might facilitate discourse processing in some individuals. While a subset were verbose in this structured task, producing more words, dysfluencies, and non-essential details, verbosity might be more widely distributed in an unstructured context, such as eliciting spontaneous output in response to an open-ended question, which likely disturbs macrolinguistic components regarding planning and coherence (Barch & Berenbaum, 1997). These processes rely on disparate aspects of cognition and given network dysfunction in TLE, discourse disturbances are unlikely lateralised (Coelho et al., 2005). Unstructured contexts are more representative of communicative function (Adams & Lloyd, 2005) and are akin to the spontaneous language produced clinically where unusual language features in TLE are anecdotally reported.

The examination of naturalistic output that we will describe here is novel and addresses the paucity of detailed, systematic, and appropriate investigation of language in paroxysmal conditions. This study aimed to characterise language impairments in TLE by examining discourse under structured (Cookie Theft) or unstructured (Typical Day) conditions. To date, spontaneous output in TLE has been largely examined with aphasia batteries which are appropriate to detect focal impairments following stroke, but are not, by design, sensitive to the subtle language changes that characterise TLE (Dutta et al., 2018). Given that microlinguistic processes are language-specific, and underpinned by subtle interictal disturbances (Rao et al., 1992), we hypothesised that individuals with TLE would have microlinguistic disturbances in fluency, marked by more frequent hesitations and false starts, irrespective of contextual demands, and that these impairments would relate to seizure characteristics. On the other hand, given the reliance of macrolinguistic processes on broader aspects of cognition including planning and working memory (Coelho et al., 2005; Shulman, 2000), we anticipated that macrolinguistic disturbances would relate to cognitive disturbances and the broader impact of TLE on cortical dysfunction. We hypothesised that macrolinguistic impairments would emerge when there were fewer constraints on output and therefore greater demands on high-level language systems: individuals with TLE would be comparatively verbose, using more words and speaking for a longer duration, and would additionally exhibit disturbances to cohesion and coherence.

Material and Methods

Participants

This study included 30 participants, 16 with focal unilateral TLE (comprising nine mesial), and 14 age- sex- and education-matched healthy controls. These individuals experience seizures of temporal lobe origin, typically focal aware seizures (FAS) or FIAS and are patients of either The Royal Melbourne Hospital or the Alfred Hospital in Melbourne, Australia. Diagnoses are made as part of a comprehensive team in accordance with International League Against Epilepsy criteria (Engel, 2006). Consensus on the basis of seizure semiology, video electroencephalography, magnetic resonance imaging, positron emission tomography, and inter-ictal single-photon emission computer tomography provided unambiguous diagnoses with a localised seizure focus, 11 of which had left hemisphere involvement.

Inclusion criteria were: English as a first language; a diagnosis of drug-resistant TLE at the time of recruitment (Kwan et al., 2010); no prior neurosurgical resection; full scale IQ >70; no reported history of substance-related and addictive disorders, no formally diagnosed psychiatric disorders, and not currently experiencing a major psychiatric episode (e.g., psychosis). None of these individuals were receiving additional treatments for the control of seizures (e.g., vagal nerve stimulation) and none had a history of developmental language disorder or other neurological condition (e.g., stroke). By assessment, three individuals with TLE reported no seizures in the preceding 12 months on their current anti-seizure medication (ASM) regimen (Kwan et al., 2010). Analyses were performed both with and without these drug-responsive participants to determine whether their inclusion in the sample was a robust choice. We found no difference in key outcome measures and these individuals were subsequently retained in the sample. Their reduced epilepsy burden at that point in time is reflected in the 13-point Seizure Frequency Rating (So et al., 1997); considering seizure frequency, type, and ASM usage. Healthy controls comprised family members or partners of TLE participants, and where necessary were recruited from the community via convenience sampling to ensure the groups were demographically comparable by appropriately age-, education-, and sex-matching to those with TLE.

This multi-site study received ethical approval from the Melbourne Health Human Research Ethics Committee in accordance with ethical standards of the 1964 Declaration of Helsinki. All participants provided written informed consent. This study has not been pre-registered.

Neuropsychological Assessment and Discourse Elicitation

This study involved neuropsychological and language assessment (Table S1), conducted by a single registered psychologist. This evaluated a range of domains including attention, memory, language, executive function, and mood. Given the COVID-19 lockdown conditions in Melbourne, Australia at the time of collection, these assessments were predominantly completed via telehealth (Table 1). Lexical retrieval was assessed using the Boston Naming Test (BNT) (Kaplan et al., 1983), the Controlled Oral Word Association Test (COWAT) (Strauss et al., 2006), the Auditory Naming Test (ANT) (Hamberger & Seidel, 2003) with minor modifications to suit the Australian lexicon, and the Verb Generation Task (VGT) developed to examine verb retrieval (Supplementary Material, Appendix S1). The BNT, ANT, and VGT were evaluated for the number of spontaneous correct responses and response latencies as metrics of lexical retrieval.

Table 1. Sample characteristics for TLE group and healthy controls.

	Control (n = 14)			TLE (n = 16)					p			Effect size
Variable	Median (Q1, Q3)	Range	Median (Q1, Q3)		Range
Age [years]	51.50 (26.00, 55.50)	50	43.50 (32.50, 55.50)		44		0.92			0.03
Education [years]	15.50 (12.00,18.00)	9	14.00 (11.75, 17.00)		15		0.45			0.17
Estimated IQ^a	107.75 (97.63, 111.88)	41	103.75 (95.88, 107.00)		50.50		0.20			0.28
Age at Diagnosis [years]	N/A	N/A	26.50 (21.00, 46.25)		50
Epilepsy duration [years]	N/A	N/A	7.50 (3.50, 16.25)		41.50
Seizure Burden Rating	N/A	N/A	6.50 (2.00, 7.00)		8
Last seizure [weeks]	N/A	N/A	10.00 (4.00, 34.00)		75
Current ASMs	N/A	N/A	2.00 (1.00, 2.00)		3
Laterality, Left [n, %]	N/A	N/A	11 (69)
Mesial focus [n, %]^	N/A	N/A	9 (56)
Sex, Female [n, %]	7 (50)		7 (44)					0.73					0.06
Handedness, Right [n, %]	12 (86)		13 (81)					0.74					0.06
Telehealth [n, %]	14 (100)		15 (94)					0.34					0.17

Note. TLE = Temporal Lobe Epilepsy; ASM = Anti-seizure Medication. For continuous variables, group differences were computed using Mann-Whitney U test, effect size reported is the rank biserial correlation coefficient. ^aSuggests that data does not violate assumptions of normality on Shapiro-Wilk. For discrete variables, percentages of each group reported, group differences computed using X² test of independence, effect size reported is phi-coefficient. Estimated IQ reflects a composite between TOPF scaled score and WASI-II FSIQ-2 score. Seizure frequency calculated on 13-point rating scale (0-12) considering ASM usage, frequency, and type of seizure (So et al., 1997). ^Of these individuals 5 had a diagnosis of left TLE. Non-mesial cases comprise 4 neocortical (3 left, 1 right), and 3 non-lesional.

Discourse-level language was examined via two forms of elicitation: one being unstructured and unconstrained (Typical Day) by prompting participants “tell me about a typical day in your life”; and the other being narrative-based, structured, and highly constrained in its content (Cookie Theft) of the Boston Diagnostic Aphasia Examination (Goodglass & Kaplan, 1972). The cartoon image was presented on an A4 sheet of paper at the bedside, and via screen sharing for telehealth assessments and remained visible throughout their description to minimise any memory contribution. Participants were encouraged to take enough time to understand the cartoon and were prompted to “tell me everything you see happening in this scene”. There was no time limit imposed for either task and the order of presentation was identical for all participants. The researcher minimised verbal and non-verbal participation and participants were required to make a definitive statement to conclude the task e.g., “I’m finished”. Participants were prompted “is there anything else?” where a particularly brief, simplified description reasonably indicated a misunderstanding of task requirements, or where greater than five seconds of silence had elapsed without self-disclosing their completion.

Recording and transcription

Language samples were audio recorded for subsequent transcription and analyses. A Yeti microphone and Audacity® software was used at the bedside, while telehealth output was recorded from the Zoom session (Zoom Video Communications Inc, 2016). All audio files were then manually transcribed verbatim and segmented by a single researcher within four weeks of the file being obtained. Statement segmentation aligns with methodology applied by Stein and Glenn (1979) and Trabasso and van den Broek (1985), where a single statement refers to a predicate and its corresponding arguments. This promotes consistent proposition-based extraction of content (Davis et al., 1997), rather than by communicative unit (C-unit). Pause lengths were manually extracted from Audacity® software. Sample lengths refer to the total number of completed words, excluding words filling pauses such as “um” “uh” (Nicholas & Brookshire, 1993; Stockbridge et al., 2021).

Discourse variables and coding practices

Based on models of discourse production and discourse examination in clinical populations, and previous literature regarding impairment in TLE, select discourse variables were used as the basis for the discourse analysis. These relate to key linguistic components at both the micro- and macrolinguistic level, being lexical-syntactic and suprasentential, respectively and relate to output volume, fluency, cohesion, and coherence. Composite metrics were developed to reflect these components. Fluency disturbances are commonly reported in TLE and were investigated using a measure of disruption (comprising hesitations, word finding difficulties, false starts, and clarity disruptors). Cohesion is reliant on working memory which is presumably more limited where planning demands are higher. A composite metric of cohesion disruption (comprising ‘ands’, other referents i.e., missing or attempted ties) was used to evaluate how capacity limitations in TLE might impact cohesion. Examining coherence was core to this investigation in order to explore previous reports of repetitive content and extraneous details in TLE, this was evaluated by extracting the number of novel units and non-progression errors in language samples.

A description of all discourse features coding is presented in Appendix 1. Transcript coding was completed using NVivo 12 software by a single researcher who was blind to participant characteristics, other than data acquisition date. An example of a coded transcript is in Supplementary Materials, Figure S1. Each transcript was coded twice by the same reviewer, with an intra-rater agreement of 94% across coding occasions. Where discrepancies were identified, the researcher re-considered the criteria to produce a final decision. Any ambiguities in the description of criteria for coding features were clarified.

Coding agreement

A second expert researcher was assigned to blindly code a random subset of transcripts in NVivo. In line with other similar discourse analyses, this was determined to be 12.5% (Sherratt, 2007) and included five transcripts of each task. Inter-rater agreement was determined on a point-by-point basis for each discourse feature to allocate as well as appropriate statement segmentation. The second researcher had access to the complete, disambiguated code-book for this process (Appendix 1). For discourse features, percentage of agreement was 90% overall (ranging from 82%-100% for individual features), while for the segmentation of statements, that is, determining each predicate and its corresponding arguments, agreement was 92%, surpassing the minimal accepted requirement level of 80% (Kazdin, 1982). Coding discrepancies were discussed among the reviewers and resolved via consensus. Once again, any ambiguities in discourse feature descriptions were resolved and their coding was updated (Appendix 1).

Statistical Analyses

Jamovi software (Jamovi, 2021) was used for all analyses. For sample characteristics, neuropsychological measures, and discourse variables group-specific measures of central tendency and group-differences were computed. Many of the data were skewed and to be conservative, non-parametric tests (Mann-Whitney U test) have been applied across all analyses for the purpose of uniformity. Data that did not violate assumptions of normality are indicated in table notes. Contingency tables (X² test of independence) were used for categorical variables. The rank biserial correlation (RBC) was used as a non-parametric estimate of effect size, reflected as small 0.1 < 0.3 < 0.5 large. Micro- and macrolinguistic discourse variables that significantly distinguished individuals with TLE from controls were then subject to correlation analyses, to examine their relationship to clinical and cognitive characteristics, and are reported as Spearman’s rank correlation coefficients. To ensure that the analyses were not sensitive to an artifact such as group, we ran the analyses with left and right TLE separately and then together. These analyses suggested no difference in discourse outcomes between left and right TLE or relative to healthy controls when individuals with TLE were considered as separate left and right groups or a single group. Consistent with the notion that high-level language functions are not lateralised, all individuals with TLE were subsequently treated as a single group. This methodological point is further addressed in the discussion. To account for Type I error, a false discovery rate (FDR) of 0.05 was applied to primary analyses (Benjamini & Hochberg, 1995).

Results

Sample characteristics

Individuals with TLE and healthy controls were comparable across most demographic characteristics and many aspects of neuropsychological function (Table 1 and Table S1). Those with TLE reported higher rates of depressive symptomatology than healthy controls. A lexical retrieval deficit broadly emerged in TLE, and they reported more frequent and distressing word finding difficulties than controls. Rather than total raw or scaled scores for the number of correct items, delays in word retrieval for BNT and ANT appeared to be more sensitive to failures of lexical retrieval—increased mean response time latencies and increased tip-of-the-tongue (TOT) states, i.e., responding >2000ms post-stimulus or requiring phonemic prompting. Individuals with TLE also demonstrated longer latencies on VGT. Relative to controls, individuals with TLE also produced fewer words within a semantic category. See Table S1.

Cookie Theft: Constrained Output

The groups were comparable in most aspects regarding output volume and coherence, including sample length, duration of output, duration excluding pauses, total number of statements, total core units, proportion of non-progression to novel statements, and syntactic simplicity. There was the strongest statistical evidence for group differences in fluency (see Table 2), as these relationships held when FDR correction was applied. This included individuals with TLE producing significantly more fluency disruptions, Mann-Whitney U = 43.50, p = .005, r = 0.61, which include pausing more frequently at non-grammatical junctures when compared to healthy controls, Mann-Whitney U = 39.50, p = .003, r = 0.65. Refer to Table 2 for complete results of the analysis for key discourse features. The relative ranking of effect sizes are presented in Figure 1. Examples of this output are provided in Appendix 2, Table 1.

Figure 1. Mean differences on discourse variables between TLE and Healthy Controls for Cookie Theft task, represented as absolute value effect size (r, small 0.1 < 0.3 < 0.5 large), * = significant difference, † = difference holds on FDR correction. Microlinguistic features are represented in black, while macrolinguistic features are grey.

Table 2. Group differences in discourse features in constrained output: Cookie Theft task.

	Control				TLE		Mean Difference	Statistic	p	r
Discourse Measure	Median (Q1, Q3)	Range			Median (Q1, Q3)	Range
Sample length	90.00 (76.00, 136.75)	124.00			99.00 (76.25, 115.25)	147.00	-1.00	111.00	0.98	0.01
Spontaneous duration (seconds)^a	41.46 (34.40, 57.68)	65.59			54.98 (39.71, 62.26)	67.47	-7.72	82.00	0.22	0.27
Pause duration	12.34 (6.61, 16.07)	14.68			17.07 (11.88, 27.49)	48.65	-6.56	56.00	0.02*	0.50
Duration excluding pauses (seconds)^a	30.33 (23.78, 40.50)	51.08			33.11 (25.56, 40.98)	42.81	-1.09	109.00	0.92	0.03
Production rate (words/second)^a	2.33 (2.13, 2.47)	1.08			2.04 (1.60, 2.22)	1.32	0.38	53.50	0.02*	0.52
Total statements	13.00 (10.00, 15.75)	14.00			13.00 (11.00, 18.75)	18.00	-1.00	94.00	0.46	0.16
Fluency disruptors^^a	0.28 (0.21, 0.35)	0.26			0.38 (0.32, 0.41)	0.42	-0.11	43.50	0.005**^†	0.61
Pauses (non-grammatical)^^a	0.08 (0.06, 0.11)	0.13			0.14 (0.11, 0.17)	0.17	-0.06	39.50	0.003**^†	0.65
Disrupted Cohesion	1.00 (0.00, 2.00)	3.00			3.00 (0.75, 4.00)	8.00	-2.00	58.50	0.02*	0.48
Cookie core units^a	16.00 (14.00, 19.25)	15.00			16.50 (15.00, 18.25)	14.00	-1.00	102.00	0.69	0.09
Proportion non-progression to task-on novel units	0.33 (0.26, 0.43)	0.49			0.39 (0.27, 0.64)	2.40	-0.28	86.00	0.29	0.23
Syntactic Simplicity^	0.04 (0.03, 0.08)	0.11			0.06 (0.05, 0.09)	0.20	-0.02	77.00	0.15	0.31

Note. Metrics marked ^ are expressed relative to sample length. Group differences computed using Mann-Whitney U test, ^a denotes that assumptions of normality held for these analyses. Effect sizes (r) expressed as the rank biserial correlation (RBC) coefficient, where * = p <.05, ** = p <.01. † = significance holds on false discovery rate (FDR) correction.

Typical Day: Unconstrained Output

Group differences emerged in coherence, informativeness, and circumstantiality of output (see Table 3). The relative ranking of effect sizes can be visualised in Figure 2. Examples are provided in Appendix 2, Table 2. There was the strongest statistical evidence for group differences in spontaneous duration, pause duration, and proportion of non-progression to novel units, which remained significant on FDR correction. Compared to healthy controls, individuals with TLE spoke for longer, Mann-Whitney U = 50.00, p = .01, r = 0.52, and had a longer duration of pauses, Mann-Whitney U = 34, p = .001, r = 0.68. With regards to informativeness, those with TLE had a higher proportion of non-progression statements to task-on novel statements than healthy controls, that is, they produced more statements that did not progress content than statements that did, Mann-Whitney U = 45.00, p = .009, r = 0.57. Refer to Table 3 for complete results of the analysis for key discourse features.

Figure 2. Mean differences on discourse variables between TLE and Healthy Controls for Typical Day task, represented as absolute value effect size (r, small 0.1 < 0.3 < 0.5 large), * = significant difference, † = difference holds on FDR correction. Microlinguistic features are represented in black, while macrolinguistic features are grey.

Table 3. Group differences in discourse features in unconstrained output: Typical Day task.

	Control				TLE		Mean Difference	Statistic	p	r
Discourse Measure	Median (Q1, Q3)	Range			Median (Q1, Q3)	Range
Sample length	134.50 (92.00, 203.25)	298.00			188.00 (143.00, 453.00)	546.00	-73.50	59.00	0.04*	0.44
Spontaneous duration (seconds)^a	57.43 (43.65, 85.15)	89.13			89.26 (64.83, 191.66)	211.04	-42.38	50.00	0.01*^†	0.52
Pause duration^a	14.64 (10.57, 18.96)	26.19			28.23 (19.06, 53.53)	54.76	-13.64	34.00	0.001**^†	0.68
Duration excluding pauses (seconds)^a	40.74 (31.70, 62.30)	84.52			62.22 (45.76, 134.73)	158.67	-29.10	54.00	0.03*	0.49
Production rate (words/second)^a	2.33 (2.04, 2.68)	1.24			2.30 (2.15, 2.50)	1.28	0.13	93.50	0.63	0.11
Total statements	25.00 (17.25, 31.50)	50.00			29.00 (24.50, 70.0)	29.00	-11.00	65.50	0.09	0.38
Fluency disruptors^^a	0.34 (0.27, 0.36)	0.41			0.36 (0.28, 0.42)	0.36	-0.05	79.00	0.27	0.25
Pauses (non-grammatical)^	0.10 (0.08, 0.12)	0.16			0.14 (0.11, 0.16)	0.14	-0.03	51.00	0.02*	0.51
Disrupted Cohesion	1.00 (0.00, 6.75)	13.00			4.00 (3.00, 6.50)	15.00	-2.00	68.50	0.11	0.35
Proportion non-progression to task-on novel units	0.39 (0.24, 0.51)	0.95			0.78 (0.51, 0.91)	3.17	-0.35	45.00	0.009**^†	0.57
Syntactic Simplicity^^a	0.11 (0.03, 0.13)	0.20			0.10 (0.06, 0.11)	0.17	0.01	98.50	0.79	0.06

Note. Metrics marked ^ are expressed relative to sample length. Group differences computed using Mann-Whitney U test, ^a denotes that assumptions of normality held for these analyses. Effect sizes (r) expressed as the rank biserial correlation (RBC) coefficient, where * = p <.05, ** = p <.01. † = significance holds on false discovery rate (FDR) correction.

Correlations

The relationships between these discourse variables surviving FDR correction and demographic, epilepsy-specific, and cognitive variables are reported in Table 4.

Table 4. Correlations between key discourse variables, demographic, clinical, and cognitive characteristics.

Variables of interest	Cookie Theft			Typical Day
	Fluency disruption	Pauses (non-grammatical)		Spontaneous Duration	Pause Duration	Non-progression to novel units
Age at assessment	-0.01	-0.08		-0.05	-0.13	0.14
Years of Education	-0.15	0.08		-0.37	-0.23	-0.33
Disease duration	0.50**^†	0.47**		0.43*	0.55**^†	0.57**^†
Time since last seizure	0.58***^†	0.56**^†		0.41*	0.53**^†	0.48**^†
Seizure frequency rating	0.40*	0.47**		0.46*^†	0.57**^†	0.46*^†
Number of ASMs	0.44*	0.48**		0.48*^†	0.58**^†	0.63**^†
IQ composite	-0.25	-0.00		-0.43*	-0.42**	-0.25
Victoria Stroop Dots	-0.15	-0.09		-0.11	-0.03	0.35
WAIS-IV Digits Backwards	0.19	0.19		-0.37*	-0.21	-0.16
Animals	-0.22	-0.20		-0.20	-0.36	-0.25
OLR	-0.30	-0.09		-0.24	-0.31	-0.01
Lexical retrieval TOT	0.40*	0.33		0.40*	0.44*	0.50**^†
Lexical retrieval latency	0.20	0.21		0.38*	0.41*	0.32
HADS depression	0.13	0.27		0.19	0.31	0.38*

Note. ASM = Anti-seizure Medication; IQ = Intelligence Quotient; WAIS-IV = Wechsler Adult Intelligence Scale, Fourth Edition; WMS-IV = Wechsler Memory Scale, Fourth Edition; OLR (COWAT) = Orthographic Lexical Retrieval (Controlled Oral Word Association Test); TOT = tip of the tongue state; HADS = Hospital Anxiety and Depression Scale. Reported as Spearman rank correlation coefficients. *p < .05, ** p < .01, *** p < .001. ^† = hold on FDR correction

Cookie Theft Correlations

Microlinguistic components distinguished individuals with TLE from healthy controls and relate closely to seizure characteristics. Fluency disruptions correlated significantly with epilepsy variables including disease duration and time since last seizure. The number of non-grammatical pauses also significantly correlated with the time since their last seizure.

Typical Day Correlations:

All three variables of interest, including micro- and macrolinguistic elements, correlate strongly with epilepsy-specific variables relating to seizure frequency and the number of ASMs. Pause duration and the proportion of non-progression to novel units are also related to disease duration and the time since last seizure. The increased proportion of non-progression units correlates with more TOT states.

Discussion

Our findings suggest that individuals with TLE are not more impaired in one communicative context than another, but rather distinct linguistic impairments emerge and are hypothesised to reflect task demands. One task is highly constrained, structured, visually prompted, and contains a focused and finite set of elements to communicate, while the other is highly unconstrained, unstructured, and therefore particularly vulnerable to the effects of verbosity.

Under conditions of constrained output, there is a specific focus on providing a complete description of the visual stimulus. Individuals are limited in their lexical choices and direction of output to meet task demands, and therefore have less agency and less variability in their output. According to our findings, the restricted scope of discourse poses a linguistic challenge to individuals with TLE, who might otherwise change direction when faced with a lexical retrieval deficit. In this context their fluency is predominantly impacted, and there is no overall pattern of circumstantiality. The strongest evidence was for disturbed fluency, including more frequent non-grammatical hesitations than in healthy controls, reflecting a disturbance overall output quality. In this context, dysfluencies plausibly represent compensations for lexical retrieval deficits. These findings are consistent with those of both Hoeppner et al. (1987) who found greater dysfluencies on this Cookie Theft task across all individuals with FIAS, and in Bell et al. (2003) on a narrative task. At this stage we do not have robust evidence to suggest any difference in disturbed output rate relative to healthy controls, consistent with previous findings (Field et al., 2000; Howell et al., 1994). While Hoeppner et al. (1987) reported a verbose subset of participants with left-localised FIAS, Bell et al. (2003) suggested that individuals with TLE were similar to healthy controls in speaking time and total number of words, and were therefore not verbose. These are consistent with the present finding that there was no evidence of circumstantiality among individuals with TLE in this context.

The overall hypothesis was that verbosity, and ultimately macrolinguistic dysfunction, were more likely to be featured when output was spontaneous and unconstrained, and is supported by our findings. Individuals with TLE produced more microlinguistic errors in total, although these errors are proportional to those in healthy controls when considering sample length. While unstructured contexts have higher planning demands (Howell et al., 1994), in the current unconstrained context, content is familiar and therefore individuals are likely to have a more specific and refined lexicon for their personal lives, within a known semantic space. Presumably this familiarity limits the impact of lexical retrieval deficits on fluency. By virtue of saying more, their output does not seem as disrupted in its fluency as we might anticipate. With the freedom of spontaneous discourse, a different set of linguistic challenges emerge where macrolinguistic deficits prevail among those with TLE, impacting suprasentential mechanisms. According to our findings, they tend to deviate more from the topic of conversation; producing longer output, pausing for longer duration, and being less concise and informative in the content they communicate, that is, they say more about less. Producing more repetitive and redundant units that do not contribute to content progression disrupts coherence and represents a communicative issue. At a non-linguistic level, it has also been hypothesised that verbosity in TLE might have a social desirability component. In attempting to maintain social contact and be understood they become particularly repetitive and loquacious (Rao et al., 1992). A constrained, impersonal task such as the Cookie Theft might not elicit verbose output given a weaker social-motivation component, potentially because the task is straight forward, largely invariable in its overall portrayal, not reliant on self-representation, and not followed by questions about its content. In contrast, the unstructured questioning regarding a Typical Day is more likely to elicit this social need to maintain face via elaboration, clarification, and self-disclosures irrelevant to task goals. While the contents of a typical day might differ for those with TLE compared to healthy controls, that is, those with poor seizure control could be more limited in their daily activities, this is not presumed to affect the manner in which individuals deal linguistically with the content. We would anticipate disturbances to declare themselves regardless of the content. The presence of circumstantiality despite more restricted daily activities support the linguistic basis of this effect. That is, saying more about a limited range of content reflects impairment in conciseness and therefore coherence.

Discourse production is cognitively demanding (Ulatowska et al., 1990). Importantly, microlinguistic processes are considered language-specific, and underpinned by subtle interictal disturbances (Rao et al., 1992). The present findings regarding microlinguistic disturbances align closely with our understanding of cognitive and linguistic systems, where we see differences in aspects of function that are presumed to be directly modified and compromised by seizures generally. For example, seizure burden, ASM usage, and their impact on processing speed (Englot et al., 2010; Kwan & Brodie, 2001) have ramifications for disrupting fluency, particularly in relation to non-grammatical pauses and pause duration, which are more highly determined by fundamental neurolinguistic systems, and correlate significantly with disease duration and recent seizures. These disturbances are disproportionately observed among individuals with TLE, regardless of context. On the other hand, syntactic complexity is presumed and found to be comparable between groups as it is socio-linguistically determined (Labov, 1972) and therefore remote from the effects of seizure activity. Instead, macrolinguistic processes rely on broader aspects of cognition (Coelho et al., 2005). As a network disorder, these systems are vulnerable to disturbances in TLE with seizure focus and propagation into distal regions (Shulman, 2000). The increased demands on linguistic and non-linguistic cognitive systems involved in planning and organising spontaneous discourse potentially account for the observation that macrolinguistic impairments become more overt when task constraints decrease, but might not be sufficiently captured by the limited scope of psychometric measures. These macrolinguistic disturbances are more vulnerable to the broader syndromal effects specific to TLE, including neurotoxicity and the impact of seizure activity and subsequent widespread cortical dysfunction which confer a degree of compromise in high-level language systems. This is mirrored in significant correlations between spontaneous duration and proportion of non-progression units with seizure frequency and number of ASMs, as indicators of disorder severity, as well as increase in TOT states being correlated with proportionally higher non-progression units.

The sample size is in keeping with similar studies and is appropriate for the scope of a discourse analysis which focuses on thoroughly characterising language use (Potter & Wetherell, 1987). Participants were well-matched on demographic characteristics. Of note, each discourse type was elicited using a single task and it is difficult to discern whether something inherent in these particular tasks might have given rise to these findings. However, examining errors relative to sample length, rather than raw values, has adequately equated these tasks to make appropriate comparisons. Future research could consider including multiple tasks with similar demands to confirm these findings. This study is concerned with high-level linguistic function which is not represented in brain tissue in a modular or focal fashion. The right and left hemisphere have been posited to play different roles in language processing and its impairment (van Lancker Sidtis, 2012); however, the reported laterality-specific impairments in TLE (Lomlomdjian et al., 2017) relied on structured aphasia assessment tools and limited categorical scoring systems which do not fully characterise high-level language. While more fundamental aspects of cognitive-linguistic function can be localised or lateralised, this is not the expectation for higher-order language, as a higher cortical function. Functions as complex as this are seldom as localised as lateralised as one might think, and instead relate to widely distributed networks. In light of this, right and left TLE participants have been considered as a single diagnostic group. While this might sound counterintuitive, there was no sense in which the groups were distinguishable (see Appendix 2 Tables 3 and 4). This approach was supported by our analyses which indicated that there were no differences in discourse outcomes when considered as separate groups or collectively, and ultimately that treating them as a single group was a robust choice. While we identified no evidence of a difference between left and right groups, it is possible that in larger samples these effects could be significant. Future research with larger and more evenly distributed groups might be useful to confirm the veracity of our findings or to determine possible subtle differences between these groups at the discourse level. This preliminary research is the first of its kind in TLE to address naturalistic output with detailed discourse analysis which generated a very large corpus of data for each participant, rather than dealing with single, disparate data points. This allowed the examination of complex functions. Additionally, accounting for language dominance, atypical activation, and reorganisation is an important future direction. Procedures for lateralisation of language are typically only conducted as part of surgical decision-making, and as such this information was not uniformly available for this sample. Lastly, most assessments were completed via telehealth given COVID-19 lockdown conditions, where fewer rapport-building opportunities and technical disruptions (Frye et al., 2021) might promote brevity. Individual differences in responses to this format might have impacted their output and ultimately conflated findings, although these conditions were comparable among both groups.

These distinct patterns of impairment reflect a dynamic linguistic system taking shape under specific contextual conditions. In this functional system, micro- and macrolinguistic discourse components among individuals with TLE are differentially affected depending on task demands. Discourse in unconstrained conditions aligns closely with the clinician’s anecdotal reporting of verbosity in TLE which often does not map onto standardised assessment. This speaks to the demands of naturalistic clinical contexts where patients are asked open-ended questions. While this study has considered contextual demands on discourse, further research is essential to fully characterise the nuances of the neurolinguistic impairments in TLE beyond the single-word level. A future focus on circumstantiality in other constrained and unconstrained contexts is a viable avenue. Our findings suggest that there are multiple components of discourse with moderate to large effect sizes that might distinguish TLE from neurologically normal individuals but did not survive FDR correction. These include measures of output volume (sample length and duration excluding pauses) as well as production rate, pause duration, and cohesion disruption. Narrowing the scope of examinable discourse features to these components of volume, fluency, and cohesion, along with our reported differences could maximise data extraction in a reduced coding time. Understanding language in naturalistic contexts provides better opportunities to consider how disturbances to language in TLE impact daily communication demands, and ultimately social and occupational functioning. This work highlights the need to deeply explore and understand complex language disorders in paroxysmal conditions.

Disclosure Statement

The authors report there are no competing interests to declare.

Additional information

Funding

This work was financially supported by the Australian Government Research Training Program Scholarship (Stipend and Fee offset) awarded by the Australian Commonwealth Government and the University of Melbourne to the first author.

Notes on contributors

Fiore D’Aprano

Fiore D’Aprano Conceptualisation (equal); Methodology (equal); Investigation (lead); Data Curation (lead); Formal Analysis (lead); Project Administration (lead); Visualisation (lead); Writing – Original Draft Preparation (lead); Writing – Review & Editing (equal).

Charles B. Malpas

Charles B. Malpas Conceptualisation (equal); Methodology (equal); Formal Analysis (supporting); Resources (equal); Writing – Review & Editing (equal), Supervision (equal).

Stefanie Roberts

Stefanie Roberts Data curation (supporting); Methodology (equal); Validation (lead).

Michael M. Saling

Michael M. Saling Conceptualisation (equal); Methodology (equal); Resources (equal); Writing – Review & Editing (equal), Supervision (equal).

Appendices Appendix 1

Discourse features of interest and their coding criteria

Table

Discourse Feature	Description
“and”	Any instance of the word “and”, the semantically least marked conjunction in place of other conjunctions. To be counted separately and then included in count of total connectives.
Abandonment	Discarding all of the interrupted utterance and replacing it entirely
Attempted/incomplete ties	The number of cohesive ties which are either ambiguous or incorrect (e.g. number, gender, meaning) are tallied and the ratio to the total number of cohesive ties is calculated.
Clarity Disruptors	Composite metric includes deictic terms, circumlocution, empty phrases, and indefinite terms. Deictic terms are a word or phrase (e.g., this, that, these, those, now, then, here) that points to the time, place, or situation in which a speaker is speaking. Empty phrases are common idioms or idiosyncratic fillers not contributing any content to the discourse (e.g. “and so on and so forth”, “you know”, “like”, “but yeah”). Indefinite terms are highly non-specific nouns or terms, (e.g. “thing”, “something”, “stuff”)
Clausal embedding	Count of number of embedded clauses, as either left- or right-branching. In left-branching clauses, the embedded/dependent clause interrupts the main clause (comes first), e.g., “When I saw the cars on the road, I was really frightened.” In right-branching structures, the main clause is followed by the embedded clause, e.g., “I was really frightened when I saw the cars on the road”
Cohesion Disruptors	A composite metric including count of ‘ands’, other referents (missing or attempted ties), referent repetition errors - being those where a referent or its pronoun is used inconsistently for the same object or character.
Conjunctions	A cohesive relation occurring between clauses and specifies the way in which what is to follow is systematically connected to what has gone before, e.g. but, or, so, because. (excluding and)
Connectives	Incorporating instances of conjunctions and “ands”, thereby representing all sentence connections
Cookie core units	Total number of novel core content units as per Yorkston and Beukelman (1977)
Deictic terms	a word or phrase (such as this, that, these, those, now, then, here) that points to the time, place, or situation in which a speaker is speaking (excludes the use of “that” as a relative pronoun).
Empty phrases	Common idioms or idiosyncratic fillers not contributing any content to the discourse (e.g., “and so on and so forth”, “you know”, “like”)
False start	A phrase or sentence which is revised to modify, correct or clarify what the speaker has already said and may include an editing expression such as “I mean”, etc. E.g “I went to town on Thursday I mean on Friday”. Includes abandonments, incomplete mazes, word-choice corrections, part-word productions, and repetitions.
Filler	Any aspect of speech which is non-communicative and serves no addition to content (e.g., “um”, “ah”, “er”)
Fluency disruptors	Composite metric of clarity disruptors, false starts and repairs (including abandonment, incomplete mazes, part-word production, repetitions), hesitations (pauses and fillers) and word finding difficulties
Incomplete mazes	A phrase or sentence which is begun but abandoned, leaving the thought incomplete and signifying a shift in thought, e.g. “After going home.. well first I went shopping”
Indefinite words/terms	Highly non-specific nouns (e.g., “thing”, “something”, “stuff”)
Non-progression error	Instances where information that has already been presented is repeated or redundant, or where content is irrelevant to the task goals
Other referents	A composite metric including attempted/incomplete ties and missing referents - being those where the referent or subject is absent or omitted from the statement.
Part word production	The production or repetition of syllables and sounds within a word, e.g. “He went ho- home first”
Pause (unfilled)	Any duration of silence (>250ms)
Pause duration	Collective length of time spent pausing
Postponement	Interrupting an utterance so additional material may be inserted, with a return to the original utterance later on, e.g. “and they thought- there was pig’s blood in it and they thought that there was somebody hurt. But it was the pig”.
Production rate	Words spoken per second
Reference correction	Replacing a noun or pronoun so as to specify a referent more exactly, for the purpose of disambiguation, e.g. “Gerald and Kevin threw a snowball at me and I got him right in the face- Kevin right in the face”, or where a pronominal reference may be replaced by a noun phrase, e.g. “we went to- uh me and Jess went to Aunt Carol’s”
Referential ties	Instances where an individual refers to another object/person using pronouns. Code anaphoric and cataphoric referential ties separately. Anaphoric = noun usage precedes pronoun introduction. Cataphoric = pronoun usage precedes introduction of noun.
Repetition (single and multi-word)	Instances of repetition of a previous word or words of the utterance before carrying on, e.g. “I have- I have new shoes on”. Single vs multi-word to be coded separately, as well as utterance-initial and mid-utterance occurrences.
Repetition errors	Errors associated with the semantic relatedness of different parts of a narrative, and are marked by the inconsistent used of pronouns, or referents for the same object or character. These correspond with instances of inconsistent referential ties.
Sample length	Count of words spoken for a given task or trial. Includes all completed words uttered (excluding part words and fillers, although including repetitions)
Spontaneous duration	Length of time producing output for a given task or trial
Syntactic complexity	Composite metric: clausal embedding, postponement, conjunctions, syntactic corrections
Syntactic correction	Refers to instances of replacing a content word with the same word of a different morphological form, rearranging word order, inserting appropriate function words, altering noun-pronoun agreement, or any other repair that improves the syntactic quality of the utterance. E.g. “He felt like there’s bells in his ears and he can’t- couldn’t hear”.
Syntactic simplicity	Composite metric: missing referents, ‘ands’, and syntax errors
Syntax error	Count of errors to syntax including incomplete utterances, missing definite article, subject, or referent
Total statements	Number of statements, where single statement refers to a predicate and its corresponding arguments
Word choice correction	Replacing a content word with another of the same morphological form (e.g., “yesterday I leant- borrowed my friend’s car”)
Word finding difficulties	Instances where an individual is unable to produce the correct word, and either uses circumlocutory language or non-specific language to attempt to describe it

Appendix 2

Table A1. Description and example of TLE output in constrained output (Cookie Theft).

Cookie Theft measure	TLE impairment	TLE example output
Pause duration	Spend a longer duration pausing
Production rate (words/second)	Produce fewer words per second. Where the production rate is slower, language output is considered to be less fluent.
Fluency disruptors	Produce more disruptions of fluency than healthy controls. These include empty phrases, indefinite terms, deictic terms, false starts and repairs	“standard kitchen benches and all of that” (empty phrase) “fall and break his neck (0.60) but anyway (0.31) that’s just the way it is”(empty phrase) “trying to get a cookie as well while the boy has (0.26) been going on (0.26) this (0.87) chair”(deictic term) he’s going to fall off that chair there (deictic term) “um (1.60) couple of- couple of things sitting out on the counter (0.25) to be washed” (indefinite term)
Pauses (non- grammatical)	Produce more instances of pauses in non-grammatical positions, i.e., occurring in the middle of a clause rather than at a natural grammatical juncture. This creates disruption in the fluency of output	“and then we’ve got (0.62) a little girl” “and his (0.25) sister or a little girl is (0.39) um (0.60) putting her (1.12) yeah putting her hand up”
Disrupted Cohesion	Have significantly more elements that disrupt the overall cohesion of their language output, i.e., there are more disruptions to interconnectedness between sentences. This includes the use of other referents that are ambiguous, incomplete, or missing and errors in the consistency of referents. Whereas controls tend to avoid the use of referent repetition errors, by providing consistent referents where necessary and otherwise using anaphoric referents such as “he” and “she” to complete referential ties.	“um just having a really good talk” (missing referent) “a brother and sister are trying to eat cookies but maybe the sister’s like I’m gonna- (1.05) I’m gonna (0.25) tell (0.27) her if you don’t give me one” (ambiguous referent) “boy” “young boy” “fella” used interchangeably in same elicitation instead of single referent and subsequent ties such as “he”(referent repetition error)

Table A2. Description and example of TLE output in unconstrained output (Typical Day).

Measure Typical Day	TLE impairment	TLE example output
Sample length (total words)	Produce output containing more words overall
Spontaneous duration (seconds)	Produce output that has a longer duration overall
Pause duration	Spend a longer duration pausing
Duration excluding pauses	Even when excluding the amount of time spent pausing, those with TLE produce a longer duration of output
Pauses (non-grammatical)	Produce more instances of pauses in non-grammatical positions, i.e., occurring in the middle of a clause rather than at a natural grammatical juncture. This creates disruption in the fluency of output	“and um I uh (0.49) make myself (0.25) um some breakfast” “so I’m expecting a call from (0.25) mechanics to say (0.47) uh this car needs that done this car needs that done”
Proportion Non-Progression to Task-On Novel Units (non-progression includes task-off novel, task-off repeated, and task-on repeated-redundant units)	Produce a significantly higher proportion of statements that do not progress content relative to those that are novel and related to the task. As such, those with TLE produce more statements that are vague, repetitive, or unrelated to the task at hand. The output is less informative overall	“um (0.66) a typical day in my life is that I wake up at around about seven thirty (0.31) and I feed my dog (0.89) and uh (0.25) he’s uh champing at the bit to get fed ‘coz he only gets fed once a day (task-off novel) (0.41) so I feed him”(task-on repeated-redundant)

Table A3. Group differences between left and right TLE in discourse features for Cookie Theft task.

Discourse Measure	Left TLE (n=11)			Right TLE (n=5)		Mean Difference	Statistic	p	d/r	N rep
	Median (Q1, Q3)	Range		Median (Q1, Q3)	Range
Sample length	104.00 (85.00, 143.00)	124.00		68.00 (54.00, 110.00)	68.00	37.00	15.00	0.17	1/0.45	36
Spontaneous duration (seconds)	60.01 (50.73, 74.39)	66.00		37.41 (29.82, 48.32)	29.46	22.60	8.00	0.03*	1.23/0.71	24
Pause duration	23.37 (13.52, 28.67)	48.61		11.78 (11.68, 17.94)	8.67	7.90	12.00	0.09	0.84/0.56	50
Duration excluding pauses (seconds)	33.44 (30.10, 45.47)	41.38		19.47 (18.14, 34.75)	22.58	13.04	14.00	0.15	1.09/0.49	30
Production rate (words/second)	2.06 (1.53, 2.19)	1.32		2.03 (1.71, 2.28)	0.84	-0.06	25.00	0.82	0.08/0.09	5140
Total statements	13.00 (12.00, 21.00)	16.00		12.00 (11.00, 14.00)	10.00	3.00	17.50	0.28	0.73/0.36	64
Fluency disruptors^	0.38 (0.34, 0.47)	0.42		0.38 (0.31, 0.40)	0.13	0.01	23.00	0.65	0.46/0.16	158
Pauses (non-grammatical)^	0.14 (0.13, 0.17)	0.17		0.13 (0.10, 0.13)	0.10	0.02	19.50	0.39	0.43/0.29	180
Disrupted Cohesion	3.00 (1.50, 3.50)	8.00		3.00 (0.00, 4.00)	4.00	0.00	26.50	0.95	0.24/0.04	574
Cookie core units	18.00 (16.00, 20.00)	14.00		15.00 (13.00, 16.00)	7.00	3.00	10.00	0.052	1.05/0.64	32
Proportion non-progression to task-on novel units	0.40 (0.24, 0.60)	2.40		0.33 (0.33, 0.71)	1.15	0.03	26.50	0.96	0.15/0.04	1464
Syntactic Simplicity^	8.00 (4.50, 9.00)	16.00		5.00 (4.00, 7.00)	9.00	-0.01	23.50	0.69	0.50/0.15	134

Note. Metrics marked ^ are expressed relative to sample length. Group differences computed using Mann-Whitney U test. Effect sizes (d) reflects Cohen’s d, (r) expressed as the rank biserial correlation (RBC) coefficient, where * = p <.05, ** = p <.01. † = significance holds on false discovery rate (FDR) correction. N_{rep =} number for planned replication, computed as Mann-Whitney U test at 80% power, assuming equal group sizes and 5% alpha, two-tailed.

Table A4. Group differences between left and right TLE in discourse features for Typical Day task.

	Left TLE (n=11)		Right TLE (n=5)		MeanDifference	Statistic	p	d/r	Nrep
Discourse Measure	Median (Q1, Q3)	Range	Median (Q1, Q3)	Range
Sample length	170.00 (127.75, 456.25)	546.00	188.00 (175.00, 444.00)	362.00	-37.96	19.00	0.50	0.10/0.24	3290
Spontaneous duration (seconds)	97.08 (63.06, 185.15)	211.04	89.26 (79.83, 192.91)	123.66	-15.00	19.00	0.51	0.07/0.24	2952
Pause duration	29.28 (18.71, 59.69)	54.76	28.23 (19.96, 38.79)	38.42	1.46	24.00	0.95	0.22/0.04	612
Duration excluding pauses (seconds)	71.95 (42.06, 123.67)	158.67	62.22 (61.03, 144.03)	97.68	-16.46	18.00	0.44	0.17/0.28	1024
Production rate (words/second)	2.33 (1.99, 2.55)	1.28	2.20 (2.19, 2.30)	0.54	0.08	20.50	0.63	0.16/0.18	1154
Total statements	27.00 (24.25, 68.75)	24.25	31.00 (29.00, 68.00)	29.00	-2.57	21.50	0.73	0.07/0.14	6022
Fluency disruptors^	0.40 (0.34, 0.52)	0.32	0.26 (0.25, 0.36)	0.21	0.12	11.50	0.11	1.07/0.54	28
Pauses (non-grammatical)^	0.14 (0.12, 0.17)	0.12	0.11 (0.11, 0.14)	0.05	0.03	11.00	0.09	1.01/0.56	32
Disrupted Cohesion	6.00 (4.25, 7.75)	15.00	3.00 (2.00, 4.00)	2.00	3.00	8.50	0.049*	0.99/0.66	32
Proportion non-progression to task-on novel units	0.73 (0.51, 0.86)	1.90	0.89 (0.74, 1.00)	2.98	-0.21	16.00	0.31	0.65/0.36	72
Syntactic Simplicity^	25.50 (15.75, 28.50)	15.75	19.00 (19.00, 23.00)	19.00	2.67	21.00	0.67	0.05/0.16	11800

Note. Metrics marked ^ are expressed relative to sample length. Group differences computed using Mann-Whitney U test. Effect sizes (d) reflects Cohen’s d, (r) expressed as the rank biserial correlation (RBC) coefficient, where * = p <.05, ** = p <.01. † = significance holds on false discovery rate (FDR) correction. N_{rep =} number for planned replication, computed as Mann-Whitney U test at 80% power, assuming equal group sizes and 5% alpha, two-tailed.