1. Introduction and background
This experimental study explores how real-time language production in speaking and writing are affected when individuals adopt a deceptive stance during narrative accounts. Regardless of modality, there are two fundamental problems that speakers and writers need to solve simultaneously: determining what to say (i.e. the linguistic content) and how to say it (i.e. the linguistic form) (Hayes & Flower, Reference Hayes, Flower, Gregg and Steinberg1980; Levelt, Reference Levelt1989). Managing these problems relies on limited processing capacity (cf., Baddeley, Reference Baddeley2007; Cowan, Reference Cowan2010), and when multiple demands compete for these resources, moment-to-moment fluctuations in planning and monitoring processes can leave observable traces during language production (Levelt, Reference Levelt1989; McCutchen, Reference McCutchen1996). Such traces often manifest as differences in pausing behaviour in both speaking and writing, as well as in patterns of verbal disfluencies, such as false starts, repetitions and reformulations in speaking (Clark & Wasow, Reference Clark and Wasow1998), and in revision behaviours, such as deletions, insertions and rearrangements of text in writing (Torrance, Reference Torrance, MacArthur, Graham and Fitzgerald2016).
Building on this framework, the present study investigates how adopting a deceptive stance – defined as sustained communicative orientation towards deception – affects speaking and writing processes when producing multi-sentence narrative texts. To explore this, participants were instructed to truthfully and deceitfully retell events depicted in four specially designed elicitation films, whereof one text in each veracity condition was in speaking and one in writing. In the deceptive condition, participants were instructed to deliberately alter a specific part of the narrative – that is, to lie about the events.
1.1. Defining deception
There is a wide literature defining deception, but in this study, deception (or lying) refers to when a person knowingly gives false or misleading information with the purpose of making the receiver believe it to be true, that is, subjective deception that does not account for the factual truth but is instead closely related to the intention of the speakers/writers. Deception can occur either through fabricating parts of a story, inventing something completely novel (DePaulo et al., Reference DePaulo, Kashy, Kirkendol, Wyer and Epstein1996), or by lying through omitting information (Leal et al., Reference Leal, Vrij, Deeb and Fisher2024; Vrij et al., Reference Vrij, Leal, Deeb and Fisher2025). Here, only lies where information is changed or added are of interest since omitting information is not linguistically or cognitively comparable to adding or altering the events of a story.
1.2. Cognitive processes underlying language production
During language production, speakers and writers are engaged in multiple interrelated cognitive processes, such as planning, formulating, monitoring and revising, described in well-established stage theories on speaking (Fromkin, Reference Fromkin1973; Levelt, Reference Levelt1989) and writing (Hayes & Flower, Reference Hayes, Flower, Gregg and Steinberg1980; see also the edited volume on language production in speaking and writing by Hartsuiker & Strijkers, Reference Hartsuiker and Strijkers2023). These processes rely on an interplay between long-term memory and working memory. Long-term memory provides access to stored linguistic knowledge, including lexical, syntactic and genre-related information, which forms the foundation for constructing coherent and contextually appropriate utterances. At the same time, working memory supports the temporary maintenance and manipulation of linguistic and conceptual representations (Baddeley, Reference Baddeley2007; Cowan, Reference Cowan2010), requiring speakers and writers to recruit their limited attentional resources to coordinate the interplay between integrating retrieved information and transforming abstract ideas into linguistic output (Kellogg, Reference Kellogg2001; McCutchen, Reference McCutchen2000). Consequently, language production is an inherently dynamic and effortful cognitive activity that relies on both the activation of pre-existing knowledge and the real-time processing of information. As speakers/writers work to meet communicative goals and adapt to contextual constraints, the demands placed on limited attentional resources contribute to altered processing demands (Clark & Wasow, Reference Clark and Wasow1998; Matsuhashi, Reference Matsuhashi1981).
1.3. Language production in speaking and writing
Most linguists would describe speaking as the primary mode of communication, ontologically as well as for everyday purposes. Speaking is typically used in day-to-day interactions, often in dialogues (Clark, Reference Clark1996; Du Bois & Giora, Reference Du Bois and Giora2014; Linell, Reference Linell2009). Writing, on the other hand, is usually described as a monologic activity, but this traditional view should perhaps be challenged given how writing is often used in everyday computer-mediated communication (Brandt, Reference Brandt2015). However, there are some inherent differences between the modalities that constrain them in different ways.
First, speaking is instantaneous – the listener is typically present when the message is produced – whereas writing is delayed – the reader typically reads the text in a different time and place from when it was produced. Second, speaking is auditory – produced with our vocal organs and perceived auditorily – whereas writing is visual – it is produced with a writing tool external to our bodies. Third, speaking is fast – we need only our vocal organs to speak – whereas writing is slow – it relies on an ability to encode meaning visually by using, for example, pen and paper or computer and keyboard. Lastly, speaking is ephemeral – it disappears when it has been uttered – whereas writing is lasting – it records meaning and lasts over time to be read repeatedly (Chafe & Danielwicz, Reference Chafe and Danielwicz1987; Johansson, Reference Johansson2009).
These variances give rise to differences in how speakers and writers manage their cognitive resources during production. The fleeting nature of speaking allows for only a small window of opportunity for corrections or additions (Christiansen & Chater, Reference Christiansen and Chater2016; Clark & Wasow, Reference Clark and Wasow1998), whereas the lasting, visual nature of writing provides writers with the possibly to visually attend to their text: delete, add or move different parts. Importantly, writers are not constrained by the immediate reception of their linguistic output, which allows for jotting down ideas and revise them later in a way that is not possible in speaking. Such revisions in writing will normally not leave traces in the final written text. Contrarily, speakers typically operate under pressing time constraints and are often observed by listeners in real time, making pausing and reformulation publicly available. Further, as Clark (Reference Clark1996) has shown, pauses serve a functional role in speaking; it allows for planning while also facilitating the listener’s comprehension, aiding the perceptive processing of the message. No such joint function exists in written communication, where instead pauses can be assumed to serve the writers’ needs. Importantly, despite the communicative function of pauses in speaking, both silent and filled pauses are still present even in contexts with no present listener. This indicates that dialogic features are characteristic of speaking independent of the listeners’ needs, and the pressing time constraint of speaking remains even without a listener present; what has been uttered cannot be edited (Christiansen & Chater, Reference Christiansen and Chater2016).
1.4. Verbal deceit
In forensic settings, methods for assessing credibility of, for example, witness statements have been developed, such as criteria-based content analysis and statement validity assessment (cf. e.g. Amado et al., Reference Amado, Arce, Fariña and Vilariño2016; Steller & Köhnken, Reference Steller, Köhnken and Raskin1989). However, as argued by Nicklaus and Stein (Reference Nicklaus and Stein2020), linguistic perspectives can supplement these findings, as shown by Adams and Jarvis (Reference Adams and Jarvis2006) who examined longer written statements made to police from a forensic linguistic perspective. The methods to assess creditability through language includes, for example, examinations of deceptive texts to find out if lies show certain syntactic structures (Newman et al., Reference Newman, Pennebaker, Berry and Richards2003), specific vocabulary (Johansson et al., Reference Johansson, Gullberg and Johansson2025; Zuckerman et al., Reference Zuckerman, DePaulo and Rosenthal1981) or other discerning phenomena. Such studies have mainly focused on investigating finished texts rather than speaking and writing processes. From a cognitive perspective, beyond finished texts, deception is expected to lead to increased processing demands since liars must remember the truth while coming up with a lie (Granhag et al., Reference Granhag, Vrij and Verschuere2015). Research has tried to identify processing cues for deception, for example, through response time studies showing that answering autobiographical yes/no questions take longer when lying than when telling the truth (Suchotzki et al., Reference Suchotzki, Verschuere, Van Bockstaele, Ben-Shakhar and Crombez2017; Walczyk et al., Reference Walczyk, Roper, Seemann and Humphrey2003; Williams et al., Reference Williams, Bott, Patrick and Lewis2013). This increase in response time has been attributed to response inhibition, as liars must suppress true facts and generate a conflicting response (Williams et al., Reference Williams, Bott, Patrick and Lewis2013), reflecting the involvement of cognitive control and increased processing demands.
Further, Mann et al. (Reference Mann, Vrij and Bull2002) examined cues of deception in interviews by police officers and revealed a significant increase in the number of pauses in lies compared to truths. However, in a meta-analysis of deception cues by DePaulo et al. (Reference DePaulo, Lindsay, Malone, Muhlenbruck, Charlton and Cooper2003), the 116 reports did not reveal a difference in disfluencies in the form of pausing or hesitation in lies compared to truths, but an increase in verbal disfluencies, such as word repetitions, was reported in deceptive texts. Oftentimes, deception studies have examined disfluencies in speech in a wider perspective, not differentiating between verbal (i.e. word repetitions and reformulations) and non-verbal (pausing behaviour, including both silent and filled pauses) disfluencies. For example, Bond et al. (Reference Bond, Kahler and Paolicelli1985) and DePaulo et al. (Reference DePaulo, Rosenthal, Rosenkrantz and Green1982) used the wider concept of speech hesitations, including both repetitions and filled pauses in the same category, and found that hesitations increased in lies. The findings in relation to disfluencies in spoken deception are inconsistent, and part of this is probably explained by different operationalisations of disfluencies in speech. In writing, Banerjee et al. (Reference Banerjee, Feng, Kang, Choi, Moschotti, Pang and Daelemans2014) examined real-time writing of short deceptive and truthful reviews and showed an increase in the number of pauses and edits during deception.
Deception has also been studied across modalities, for instance, Leins et al. (Reference Leins, Fisher and Vrij2012) asked participants to report the same information across different report modes (e.g. verbal and pictorial) and found that liars were less consistent across the modes, suggesting that this was due to an increase in cognitive load when lying. Similarly, Vrij et al. (Reference Vrij, Mann, Fisher, Leal, Milne and Bull2008) examined the effect of increasing cognitive load during deception by instructing participants to recount events in reverse order, showing that signs of heightened cognitive load increased more in lies. Likewise, Vrij et al. (Reference Vrij, Mann, Leal and Fisher2010) found that even simple concurrent tasks, such as maintaining eye-contact while lying, imposed additional processing demands and amplified behavioural indicators of deception.
Further, self-regulating behaviours, such as avoiding mannerisms associated with lying (e.g. fidgeting, moving ones’ legs) during spoken deception, are also expected to lead to altered processing demands (DePaulo, Reference DePaulo1992). Additionally, self-regulation during deception is linked to ideas about what a lie sounds or looks like since liars try to control behaviours that they believe liars exhibit rather than actual cues to deception. Hartwig and Granhag (Reference Hartwig and Granhag2015) reported that that common beliefs are that liars show behaviours such as gaze aversion, increased hesitations and increased fidgeting.
1.5. Aim and present study
The present study is conducted against the backdrop of previous research linking deception to altered processing demands across a range of tasks and modalities – from increased inhibitory control in simple lie paradigms (Bott & Williams, Reference Bott, Williams and Meibauer2019; Suchotzki et al., Reference Suchotzki, Verschuere, Van Bockstaele, Ben-Shakhar and Crombez2017; Walczyk et al., Reference Walczyk, Roper, Seemann and Humphrey2003) to broader effects such as reduced consistency (Leins et al., Reference Leins, Fisher and Vrij2012), and greater strain when performing concurrent tasks while lying, mainly in speaking (Vrij et al., Reference Vrij, Mann, Fisher, Leal, Milne and Bull2008, Reference Vrij, Mann, Leal and Fisher2010). While these studies provide important insights, they often focus on isolated moments of deception – brief instances where participants are instructed to lie about a specific detail in their response, often within stimulus–response paradigms.
In contrast, the present study investigates the effects of adopting a deceptive stance – a sustained communicative orientation towards deception – during the production of multi-sentence texts in speaking and writing, here narrative accounts. This approach assumes that producing a narrative with deceptive intent imposes altered processing demands that influence planning, monitoring and revision processes throughout the task. Here, lying is not treated as a momentary act, but as a mode of engagement that permeates the speakers’ or writers’ approach to the communicative goal. This deceptive stance is assumed to affect speaking and writing processes regardless of the severity of the lie or of the amount of information that is altered in relation to the actual events that form the base for the narratives.
This distinction between transient acts of lying and a sustained deceptive stance is especially important in the context of extended narrative production where speakers and writers must continuously coordinate linguistic content and form within the limits of working memory and attention. A deceptive stance is expected to modify these processing demands by engaging different forms of monitoring and control (DePaulo, Reference DePaulo1992), such as inhibition of truthful representations (Williams et al., Reference Williams, Bott, Patrick and Lewis2013), and shifts in planning effort (DePaulo et al., Reference DePaulo, Lindsay, Malone, Muhlenbruck, Charlton and Cooper2003; Mann et al., Reference Mann, Vrij and Bull2002; Vrij, Reference Vrij2015; Zuckerman et al., Reference Zuckerman, DePaulo and Rosenthal1981).
Furthermore, while prior research has linked deception to specific linguistic cues in speech (Leal et al., Reference Leal, Vrij, Deeb and Fisher2024; Vrij et al., Reference Vrij, Fisher, Mann and Leal2006), relatively little is known about how deceptive intent affects writing processes. By comparing the production of narrative texts in both modalities, the present study addresses a critical gap in the literature and broadens our understanding of how communicative stance shapes the processing dynamics of real-time language production. In doing so, this study contributes to theoretical models of language production by examining how truthfulness versus deception influences the coordination of planning, monitoring and revision processes during extended discourse. It also offers practical implications for applied settings, such as forensic linguistics where the ability to identify markers associated with deception is of significant value.
The aim of this study is to explore how adopting a deceptive stance influences the overarching speaking/writing processes involved in language production and to determine whether deceptive intent exerts an observable influence on these processes. The influence on speaking and writing processes is investigated through the examination of disfluencies in speaking, that is, pausing behaviour (non-verbal disfluencies) and repetitions and reformulations (verbal disfluencies), as well as pauses and revisions in writing, at a global level across modalities. Data consist of narrative accounts, collected in an experimental setting where participants retold the events of four short films. This resulted in two truthful and two deceitful narratives, in both speaking and writing, from each participant. Importantly, the study does not assess participants’ memory accuracy for the film events. Rather, it assumes that truthful recall of the events in the films provides a valid basis for adopting a deceptive stance, whereby participants intentionally modify otherwise accurate narratives to conceal or alter key details.
The first guiding research question is if adopting a deceptive stance (as opposed to a truthful stance) leads to observable changes in speaking and writing processes. The present study focuses on verbal and non-verbal disfluencies in speaking, and pauses and revision in writing, as these measures offer well-established means of capturing the temporal dynamics of language production. Disfluency measures have long been shown to be sensitive indicators of planning and monitoring demands in speech (Clark & Wasow, Reference Clark and Wasow1998; Goldman Eisler, Reference Goldman Eisler1968), and corresponding measures of pausing and revision behaviour provide comparable insights into the temporal and structural dynamics of written composition (Matsuhashi, Reference Matsuhashi1981; Spelman Miller, Reference Spelman Miller, Sullivan and Lindgren2006). In both speaking and writing, these processes vary systematically with factors such as expertise, genre and task demands (Cislaru et al., Reference Cislaru, Doquet and Olive2025; Clark, Reference Clark1996), supporting their role as plausible markers of how a deceptive versus a truthful stance may alter processing demands during language production across modalities. The expectation is that adopting a deceptive stance will alter the processing demands of language production, resulting in measurable adjustments in disfluency behaviour in speaking and in pausing and revision behaviour in writing.
The second guiding research question is whether these observable changes will be more salient in one modality than the other. The expectation is that such alterations will be more pronounced in writing than in speaking given that writers can distribute processing over time, whereas speakers are constrained by the immediacy of the message formulation.
2. Method
Data in the present study were obtained in an online experimental setup where participants re-told events they witnessed in four elicitation films in their first language Swedish (henceforth, L1). The data are part of The Lund Corpus on True and Invented Narratives in Speech and Writing that is planned to be made publicly available in 2026.
2.1. Design
The experimental setup was a 2 × 2 study with modality (speaking/writing) and veracity (truthful/deceitful) as the conditions. All participants completed four different tasks where they re-told the events from one of the four films: one truthfully in speaking, one truthfully in writing, one deceitfully in speaking and one deceitfully in writing. The tasks were counterbalanced through a Latin squares design across veracity, modality and film so that all possible orders for truthful/deceitful and spoken/written narratives occurred for each film.
2.2. Participants
Inclusion criteria for participants were to be 18–40 years old, to have Swedish as an L1 and to have completed at least 1 year of university studies (= 60 ECTS). These criteria were used to avoid declining working memory due to age, to avoid effects of L2 due to, for example, delayed lexical retrieval or grammar (Barkaoui, Reference Barkaoui2019; Chenoweth & Hayes, Reference Chenoweth and Hayes2001), and to ensure a comparable baseline concerning writing skills across participants. Exclusion criteria were any self-reported speaking, reading or writing difficulties to avoid possible impact on language processes. Participants were recruited through social media and posters at different universities in Sweden.
A total of 47 participants between 18 and 40 were recruited. Three participants were removed from the data set due to technological issues or data corruption, five for failing to follow the instructions to alter the narrative in the deceitful condition and eight for lying by omission (i.e. opting to not address the event that we prompted them to change). This left 31 participants (F = 15) between 20 and 37 years old (mean = 26.935, SD = 5.013) who produced 124 narratives in their L1.
2.2.1. Ethical considerations
As established by Swedish authorities and specified in the Swedish Act concerning the Ethical Review of Research involving Humans (2003:460), the present study does not require specific ethical review due to the following reasons: (1) it does not deal with sensitive personal data, (2) it does not use methods that involve a physical intervention, (3) it does not use methods that pose a risk of mental or physical harm, and (4) it does not study biological material taken from a living or dead human that can be traced back to that person. The data collection was conducted in accordance with the ethical act, and informed consent was obtained prior to participation in the study through a written form containing information about the data that would be collected and how it would be stored. All participants were informed that they had the right to withdraw from the study without consequences at any time. Participants were given compensation for their time.
2.3. Materials
Data were collected remotely through Zoom on participants’ own computers due to the Covid-19 pandemic. Out of the 31 participants in the sample, 16 used a Windows computer and 15 a Mac computer. For the spoken tasks, 22 participants used a laptop-internal microphone and 9 participants an external microphone. For the written tasks, 27 participants used a laptop keyboard and trackpad and 4 participants an external keyboard and mouse. Spoken accounts were recorded using Audacity(R) (version 3.0.2, Audacity, 2021), and written accounts were recorded using the keystroke logging program ScriptLog, recording all keypresses during writing (Wengelin & Johansson, Reference Wengelin, Johansson, Kruse, Rapp, Anson, Benetos, Cotos, Devitt and Shibani2023).
The experimental stimuli consisted of four wordless elicitation films depicting minor misdemeanours. Each film had three characters with distinct roles: a witness, a perpetrator and a victim. English summaries of the films can be found in Appendix A.
2.4. Copytask
In the preparation phase, participants performed a copytask in ScriptLog with the purpose of establishing a baseline for typing proficiency (Van Waes et al., Reference Van Waes, Leijten, Roeser, Olive and Grabowski2021), and this revealed an average transition time of 0.127 ms between keypresses within words (SD = 0.022, min 0.085, max = 0.171), indicating a typing proficiency typical for L1 adults as demonstrated in other studies (Wengelin et al., Reference Wengelin, Johansson, Frid and Johansson2024).
2.5. Procedure
The experimental procedure had three main parts: preparation, experimental tasks and debriefing. In the preparation, the software Audacity and ScriptLog were tested to ensure that they were correctly installed on participants’ computers. Then, the copytask and a short test sample of audio recording took place, and the resulting files were transferred through a safe repository server.
In all four tasks, participants first watched a film and were then asked to re-tell what happened, from start to finish with as many details as possible. In the truthful condition, they were told to re-tell things just as it happened, and in the deceitful condition, they were told to imagine that you want to protect person X (perpetrator) and lay blame on person Y (witness) instead. They were not given a time or word limit for their narratives and were told that they were allowed to speak/write until they were done. These instructions were designed to elicit a sustained deceptive stance while allowing participants to produce naturalistic narratives at their own pace.
The experimental setup differs from real-life settings where people can be assumed to be more invested lying, for example, to profit or to help someone else – something which may influence the linguistic processes. To compensate for this and to motivate participants to be convincing, the study raised the stakes of the situation (Vrij, Reference Vrij2015) by informing participants that their accounts would be listened to/read by someone who would guess if they were true or not. During the debriefing, participants were informed that there would not be anybody who would guess if their narratives were true or not. See Appendix B for a translation of the complete instructions.
2.6. Transcription of spoken data
Transcriptions followed the CHAT-format (MacWhinney, Reference MacWhinney2000), and an example of this can be seen in Example 1. The transcriptions of the spoken data followed Modified Swedish Orthography inspired by Nivre et al. (Reference Nivre, Allwood, Grönqvist, Ahlsen, Gunnarsson, Hagman, Larsson and Sofkova Hashemi1999) to ensure comparability across veracity conditions as well as to streamline measures across modalities (Gullberg et al., Reference Gullberg, Johansson and Johansson2024). Silences of 200 ms or longer were coded as pauses, and filled pauses (e.g. saying ‘eh’) were also identified. Repetitions, repetitions with small changes and reformulations were also annotated in the transcripts.
Example 1. Example from one of the transcripts. Parentheses with a punctuation mark indicate silent pauses and ‘eh’ indicates filled pauses. Pronunciation variations are followed by a lexeme correction in square brackets
‘Eh, yes (.) eh so it is a garden café then (.) eh and eh (.) eh a girl sits on (.) a chair and reads a magazine, and she has her eh bag next to her’.
2.7. Data sample and data curation
The data used in this study are openly available through the OSF repository at https://doi.org/10.17605/OSF.IO/7RQ9G. The final data sample consisted of 124 narrative accounts: 31 spoken truthful narratives, 31 spoken deceitful narratives, 31 written truthful narratives and 31 written deceitful narratives. These data differ from previous studies where, in speaking, data have often been in the form of answers in interview settings (Vrij et al., Reference Vrij, Fisher, Mann and Leal2006; Walczyk et al., Reference Walczyk, Roper, Seemann and Humphrey2003), as well as in writing, where elicited texts have been shorter (Banerjee et al., Reference Banerjee, Feng, Kang, Choi, Moschotti, Pang and Daelemans2014). As a contrast, our data sample consists of longer, un-interrupted narratives in both speaking and writing. Table 1 offers an overview of the measures used in this study. Important to note is that the measures are not compared across modalities due to methodological challenges related to differences between speaking and writing (Gullberg et al., Reference Gullberg, Johansson and Johansson2024).
Table 1. Measures used for examining cognitive load during speaking and writing
Note: The different rows show the different measures that are used in this study, and how this was operationalised.
Text length is measured as the number of characters in the text, and this is divided into two categories: final text length and linear text length, as can be seen in Table 2. This is a useful distinction when examining written language production, since in writing, the final text (i.e. the text that the writer hands over to the reader) will differ from the linear text (i.e. the text that is produced during the writing process). The linear text is usually longer and includes all characters that were produced during writing, that is, all characters in the final text plus the deleted characters. For speaking, the final text length was measured as the number of characters in the transcripts minus the characters that were involved in repetitions and reformulations. Characters reflecting repetitions and reformulations were then instead included in the measure for linear text length.
Table 2. Final and linear text length examples
Note: The right-most column Length shows the calculated number of characters for a final text sample and a linear text sample in both writing and speaking. Translations are made as close as possible to the original Swedish, hence the non-idiomatic expressions.
Typically, text length has been measured as the number of words in both speaking and writing, but when examining real-time language production, measuring the number of words poses challenges. For instance, it is not clear how to deal with word fragments in both speaking and writing and how to deal with word repetitions in speaking and deleted words in writing. Instead, number of characters allows for a dynamic measurement of on the one hand how much that was produced in total (linear text) as well the length of the message (final text) without having to set up criteria for what constitutes a word (Gullberg et al., Reference Gullberg, Johansson and Johansson2024).
Different pause criteria for speaking and writing were adopted due to modality differences where speaking is faster than writing. In speaking, non-verbal-disfluencies in the form of pauses were categorised into silent pauses and filled pauses. Silent pauses were operationalised as silences of 200 ms or longer (Goldman Eisler, Reference Goldman Eisler1968), and filled pauses were operationalised as hesitations with vocal markers such as ‘eh’ or ‘mm’. The total number of pauses in speaking include the sum of the number of silent and filled pauses. The operationalisation of pauses in speaking differs from some of the previous studies that have examined pausing in relation to deception, commonly, mainly filled pauses (or speech hesitations) have been considered (see DePaulo et al., Reference DePaulo, Lindsay, Malone, Muhlenbruck, Charlton and Cooper2003 for an overview). However, some have also used thresholds for silent pause identification, like Mann et al. (Reference Mann, Vrij and Bull2002) who used a pause threshold of 500 ms.
In writing, a pause was operationalised as an inactivity between keypresses of 2000 ms or longer. Importantly, pauses always occur between each keypress during keyboard writing. To manage the vast amount of pause data that is produced during a data collection with keystroke logging, the common procedure in writing studies is to apply ad hoc criteria to explore pausing behaviour. A commonly used ad hoc criterion is 2000 ms (Wengelin, Reference Wengelin, Sullivan and Lindgren2006). There are writing studies that have adopted more dynamic pause thresholds that take into consideration individual variations between writers or across the writing session (Chenu et al., Reference Chenu, Pellegrino, Jisa and Fayol2014; Olive, Reference Olive2014), but such approaches are more appropriate in between-subject designs, where they serve the purpose to eliminate variations between writers (van Waes et al., Reference Van Waes, Leijten, Roeser, Olive and Grabowski2021). Here, a fixed pause threshold is assumed to be sufficient for comparing the writing processes in the within-subject design that is used, especially since the participants are adults with high typing proficiency, meaning that the fluctuation in pausing due to low-level processes related to transcription and orthography can be expected to be very low.
Following from these different pause criteria across speaking and writing is an expectation that pausing behaviour will differ across the modalities. Such differences are also expected based on the inherent constraints that characterise speaking and writing, respectively, especially the fact that speaking is produced much faster than writing.
The verbal disfluency length/deleted characters is the difference between the final text length and the linear text length, that is, calculated by subtracting the number of characters in the final text from the number of characters in the linear text. In speaking, this measure is extracted from the transcripts and measures the number of characters that were repeated, changed or reformulated; in writing, this measures the number of characters that were deleted.
2.8. Analyses
All narratives were manually checked to ensure that the truthful condition only contained narratives that were truthfully accounting for the events in each film, respectively, and that the narratives in the deceitful condition had been altered according to the instructions. In the analytical approach, increased planning was operationalised as the total number of pauses for both speaking and writing, as well as the verbal disfluency length in speaking and number of deleted characters in writing. Previous studies on deception, as well as studies of language production processes in both speaking and writing, often use proportional measures (Barkaoui, Reference Barkaoui2019) – such as number of disfluencies in speaking and number of pauses and deletions in writing divided by text length – to account for variability in output. In contrast, the present study adopts a different analytical approach.
Generalised mixed effect models (GLMMs) were fitted using the lme4 package (version 1.1.35.3, Bates et al., Reference Bates, Maechler, Bolker and Walker2015) and lmerTest (version 1.1-3, Kuznetsova et al., Reference Kuznetsova, Brockhoff and Christensen2017) in R (version 2024.12.1.563, RStudio Team, 2020). Because the dependent variables represented count data (e.g. number of pauses, disfluences or deletions), models with a Poisson distribution were used as the default. When the Poisson models exhibited overdispersion or failed to converge, a Negative Binomial distribution was employed instead, as it provides more flexible variance structure suitable for over-dispersed count data. All models were checked for overdispersion using performance (version 0.15.2, Lüdecke et al., Reference Lüdecke, Ben-Shachar, Patil, Waggoner and Makowski2021). Model assumptions were evaluated through visual inspection of residual plots (normality, homoscedasticity, linearity and independence). All models, except one, revealed no apparent violations. In the model for filled pauses, the inspection of residuals revealed a deviation from linearity for the predictor final text length. To address this, a log transformation was applied to this predictor. Post-transformation, the residual diagnostics indicated that model assumptions were adequately met. Participants and film were added as random effects, and random slopes were added following the recommendations of Matuschek et al. (Reference Matuschek, Kliegl, Vasishth, Baayen and Bates2017) and retained if they improved model fit and did not cause convergence issues. The full statistical analyses can be found in a markdown document available in the Open Science Framework at https://doi.org/10.17605/OSF.IO/7RQ9G.
Data summaries were made with vtable (version 1.4.7, Huntington-Klein, Reference Huntington-Klein2024) and data wrangling performed with tidyverse (version 2.0.0, Wickham et al., Reference Wickham, Averick, Bryan, Chang, McGowan, François, Grolemund, Hayes, Henry, Hester, Kuhn, Pedersen, Miller, Bache, Müller, Ooms, Robinson, Seidel, Spinu and Yutani2019), dplyr (version 1.1.4, Wickham et al., Reference Wickham, François, Müller and Vaughan2023), datawizard (version 0.13.0, Patil et al., Reference Patil, Makowski, Ben-Shachar, Wiernik, Bacher and Lüdecke2022) and broom (version 1.0.6, Robinson et al., Reference Robinson, Hayes and Couch2024). Further sjPlot (version 2.8.16, Lüdecke, Reference Lüdecke2024) was used to print models, ggeffects (version 2.0.0, Lüdecke, Reference Lüdecke2018) to make initial regression graphs and ggplot2 (version 3.5.1, Wickham, Reference Wickham2016) to make graphs.
3. Results
The results are presented in two sections: first follows the results for speaking and then the results for writing. In both sections, first we the report the results for the pause analyses, that is, in speaking, the non-verbal disfluencies and then the results for the verbal disfluencies (speaking) and revisions (writing). Note that no comparisons across the modalities are conducted, only between the veracity conditions within the modalities. Table 3 shows the descriptive statistics for the different outcome variables in the truthful and deceitful condition for both speaking and writing.
Table 3. Data sample in speaking and writing
Note: The top half of the table shows the descriptive statistics for the spoken data sample, and the bottom half of the table shows the descriptive statistics for the written data sample.
3.1. Spoken modality
3.1.1. Non-verbal disfluencies
The non-verbal disfluencies are divided into the total number of pauses, the number of silent pauses and the number of filled pauses. The best-fit model for total pauses included an interaction between veracity and text length, with participant as a random intercept. For silent pauses, the best-fit model included text length as a covariate with participant as a random intercept. For filled pauses, the best-fit model included an interaction between veracity and text length, with participant intercepts. Full model results are presented in Table 4.
Table 4. Result table of the generalised linear mixed effects model for non-verbal disfluencies, divided in total pauses, silent pauses and filled pauses in speaking
Note: The columns show the incidence rate ratios, the standard error, the confidence interval, the z-value and the p-value. The different outcome variables are given in the rows under each heading. Significant p-values are in bold.
For total and filled pauses, veracity was a significant predictor showing a decrease in the number of pauses in the deceitful condition (see Figure 1). However, veracity was not a significant predictor of the number of silent pauses. Final text length was a significant predictor in all three models.
Figure 1. Model predictions for the total number of pauses in truthful vs deceitful narratives in speaking. The y-axis shows the predicted number of filled pauses, and the x-axis shows the two veracity conditions. The error bars show the 95% confidence interval.
A significant interaction between veracity and text length was also observed for filled pauses (see Figure 2), where longer final text length in the deceitful condition showed an increase in the number of filled pauses, whereas shorter deceitful texts had fewer filled pauses than truthful ones.
Figure 2. Predicted number of filled pauses in the deceitful vs truthful condition on final text length in speaking. The y-axis shows the total number of filled pauses, and the x-axis shows the final text length. The ribbons show the 95% confidence interval.
3.1.2. Verbal disfluencies
The best-fit model included text length as a co-variate to veracity, with no interactions and with participant as a random intercept. The full model results are presented in Table 5. Veracity did not predict the total length of verbal disfluencies, but final text length did predict the verbal disfluency length.
Table 5. Result table of the generalised linear mixed-effects model for the total length of verbal disfluencies in speaking
Note: The columns show the incidence rate ratio, the standard error, the confidence interval, the z-value and the p-value. The different outcome variables are given in the rows under each heading. Significant p-values are in bold.
3.2. Written modality
3.2.1. Pauses
The best-fit model included a three-way interaction between veracity, final text length and time on task, with participant as random intercepts. The full model results can be seen in Table 6. While neither veracity nor text length alone significantly predicted the number of pauses, time on task did. Further, there was significant interactions between all predictors, showing that (1) writers who spend more time on task tend to have more pauses in the deceptive condition, (2) writers who spend relatively short time on task typically do not show any pausing differences between the truthful and the deceitful condition, and (3) writers who produce longer final texts tend to have differences between truthful and deceitful production moderated by time on task; see Figure 3 for an illustration of this.
Table 6. Result table of the generalised linear mixed-effects model for the number of pauses in writing
Note: The columns show the incidence rate ratio, the standard error, the confidence interval (95%), the z-value and the p-value. The different outcome variables are given in the rows. Significant p-values are in bold face.
Figure 3. Predicted number of pauses in the deceitful vs truthful condition in writing for shorter final text length (left), mean final text length (centre) and longer final text length (right) on time on task. The y-axis shows the predicted number of pauses, and the x-axis of each plot shows the time on task, divided by different final text length (the three plots). The ribbons show the 95% confidence interval.
3.2.2. Revisions
The best-fit model included veracity, text length and time on task as predictors, as well as interactions between veracity and text length and veracity and time on task and random intercepts for participants. The full result of the model can be seen in Table 7.
Table 7. Result table of the generalised linear mixed-effects model for the number of deletions in writing
Note: The columns show the incidence rate ratio, the standard error, the confidence interval (95%), the z-value and the p-value. The different predictors are given in the rows. Significant p-values are in bold.
The results revealed no significant main effect of veracity on the number of deleted characters. However, both final text length and time on task significantly predicted the number of deleted characters. There was also a significant interaction between veracity and final text length, where longer final text lengths in the truthful condition were associated with a higher number of deleted characters, illustrated in Figure 4.
Figure 4. The predicted number of deleted characters as modulated by final text length in writing. The y-axis shows the predicted number of deleted characters, and the x-axis shows the log transformed text length. The ribbons show the 95% confidence interval.
4. Discussion and conclusions
The aim of this study was to examine the effect of adopting a deceptive stance, that is, a sustained communicative orientation towards deception during extended language production, in real-time speaking and writing. To summarise, the first research question asked whether adopting such a stance would lead to observable alterations in the language production processes. Based on previous research, we expected deception to affect markers of processing such as verbal and non-verbal disfluencies in speaking, pausing and revisions in writing (Mann et al., Reference Mann, Vrij and Bull2002; Vrij, Reference Vrij2015). The results revealed that these processes were indeed influenced by deception, though not in a uniform pattern: in some cases, there were no differences between the veracity conditions, and in some cases, pausing and revision increased during deception, while other findings revealed that those behaviours occurred less during deception. The second research question addressed whether observable traces of deception during language production would be more salient in one modality than the other. It was expected that alterations in processing demands would be more evident in writing, given that writers can distribute writing processes over time, whereas speakers must manage planning and monitoring under more immediate temporal constraints. Moreover, spoken production involves consideration of a possible listener’s perspective, a listener who in a real-life setting can interrupt, but also be helpful through interactions, feedback and questions. While there was no listener present during the data collection, all participants were aware of that their spoken accounts would be listened to afterwards and that the experimental leader was nearby. From this, it is reasonable to expect that similar aspects of communicative monitoring persisted although no actual listener was present. The results were consistent with this expectation, indicating more processing adjustments during deceptive writing than during deceptive speaking.
4.1. Deceptive stance and language production
Regarding the first question, the results showed that in speaking, there was a main effect of deception on the non-verbal disfluencies, operationalised as the total number of pauses. This effect was also present when limiting the analysis to only filled pauses, but no main effect was found when limiting the analysis to the number of silent pauses, in line with the findings in the meta-analysis by DePaulo et al. (Reference DePaulo, Lindsay, Malone, Muhlenbruck, Charlton and Cooper2003). The total number of pauses and the number of filled pauses, analysed separately, revealed that deceptive speech was found to contain fewer total and fewer filled pauses. The importance of distinguishing filled pauses from silent pauses is supported by previous studies, which has identified increases in speech hesitations (e.g. Feeley & deTurck, Reference Feeley and deTurck1998; Knapp et al., Reference Knapp, Hart and Dennis1974; Mann et al., Reference Mann, Vrij and Bull2002) and speech disturbances (Bond et al., Reference Bond, Kahler and Paolicelli1985; DePaulo et al., Reference DePaulo, Rosenthal, Rosenkrantz and Green1982; Kasl & Mahl, Reference Kasl and Mahl1965) during deception. These constructs overlap with filled pauses in their operationalisation but may also include features of verbal disfluencies, such as false starts and word repetition.
Notably, an interaction effect showed that an increase in filled pauses occurred primarily in longer deceptive texts, whereas shorter texts revealed the opposite pattern. This suggests that the sustained effort required to maintain a deceptive stance may dynamically shape processing demands and influence how speakers regulate monitoring and control during speech planning. Specifically, speakers may suppress filled pauses in shorter narratives to regulate overt cues to deception, whereas in longer narratives, the extended maintenance of deceptive stance may tax these control processes, resulting in an increased number of filled pauses (cf., Hartwig & Granhag, Reference Hartwig and Granhag2015).
In writing, the results showed no main effect of veracity on the number of pauses. However, a significant interaction emerged: when writers spent more time on task in the deceitful condition, the number of pauses increased more than in truthful texts. This finding partly supports our expectation of an increase in the number of pauses and aligns with previous findings (Banerjee et al., Reference Banerjee, Feng, Kang, Choi, Moschotti, Pang and Daelemans2014). Conversely, when writers spent less time on the task, deceitful texts were associated with fewer pauses compared to truthful ones. This could be explained by the deception being short enough to not incur increased processing demands. In contrast, during longer writing sessions, the sustained effort required to maintain a deceptive stance may heighten processing demands and challenge writers’ ability to regulate the production process effectively. This interpretation is consistent with the self-presentation theory of deception (DePaulo, Reference DePaulo1992), which posits that successful deception relies on continuous monitoring and control to suppress cues that might reveal the lie.
There was also an interaction between text length and veracity, where longer deceitful texts contained fewer pauses than truthful ones. One explanation is that during deception, the extra effort of fabricating content alters how writers allocate processing resources for planning and editing. As a result, they may focus on producing a coherent narrative flow rather than engaging in extensive revision, such as evaluation and re-readings to refine the text. In contrast, when writing truthfully, they may pause more frequently to evaluate, re-read and ensure that their descriptions are as clear and detailed as possible.
Collectively, the pause-related results in the written modality suggest that deceptive intent alters how writers regulate processing demands over time. Rather than reflecting a uniform increase in difficulty, pausing appears to be strategically adjusted depending on the duration and length of the writing task.
In terms of verbal disfluencies in speaking, there was no effect of veracity on the length of verbal disfluencies. Important to note is that the operationalisation of verbal disfluencies used in this study differs from previous studies such as Bond et al. (1985) and DePaulo et al. (Reference DePaulo, Rosenthal, Rosenkrantz and Green1982), where the number of disfluencies were measured, instead of here where the characters in the transcript involved in repetitions and reformulations were calculated.
In writing, there was no main effect of veracity on the number of deletions in writing, but there was an interaction effect between final text length and veracity showing that there were fewer deleted characters in longer final texts. These findings suggest that writing deceitfully does not result in more revisions; rather, it may reflect tighter control over the production process and increased pre-planning, whereby writers delete less because they generate less content overall. Although the instructions emphasised that the writers should be as detailed as possible, it is possible that, to manage the altered processing demands during deception, writers adopt a strategy focused on conveying only the central and necessary elements of the events. In contrast, truthful writing may allow for greater elaboration, reflection and revision of less central details. This interpretation aligns with previous findings that deceptive accounts tend to be less detailed (DePaulo et al., Reference DePaulo, Lindsay, Malone, Muhlenbruck, Charlton and Cooper2003) and that deceptive tasks are associated with reduced cognitive flexibility (Leins et al., Reference Leins, Fisher and Vrij2012; Vrij et al., Reference Vrij, Mann, Fisher, Leal, Milne and Bull2008).
4.2. Deceptive stance and modality differences
Regarding the second research question, the results show that there are cross-modality similarities, but also important differences. In speaking, a deceptive stance was associated with fewer total and filled pauses. In writing, a deceptive stance led to an increase in pausing with longer writing times and a decrease in pausing in longer texts. Additionally, deceivers showed a reduction in the number of deleted characters in longer final texts.
While deception influenced both speaking and writing through changes in pause dynamics and revision behaviour – suggesting that adjustments in control and monitoring processes occur across both modalities to regulate planning and editing – these effects were more salient and appeared across a greater number of markers in the written modality. One possible explanation is the more immediate time pressure in speaking: speakers must monitor and produce output in real time, limiting opportunities for reflection and adjustment. In contrast, writers can distribute processing demands more flexibly throughout the writing session. However, this does not necessarily mean that deception is easier to detect in writing, given the complex interplay between writing time, text length and deceptive intent.
Taken together, these results suggest that writing was more readily influenced by deceptive intent than speaking, with effects appearing across a greater number of process markers. Yet, the results suggests that the direction of these effects is difficult to predict, as they likely reflect a dynamic interaction between adjustments in control and monitoring to regulate deceptive output and the sustained processing demands that vary with task duration and text length.
4.3. Theoretical implications
The study set out to explore how speakers and writers solved the problem of deciding what to say and how to say it while operating under the additional constraint of adopting a deceptive stance during language production. As the results show, deception influenced both speaking and writing, leaving traces of altered processing demands across both modalities. In speaking, this was reflected in a decreased number of total pauses, specifically driven by a decrease in the number of filled pauses. In writing, the frequency of pausing interacted with both time on task and text length, indicating that the influence of deception varied dynamically with these factors. One observation is that when faced with altered processing demands, speakers and writers adapt their production processes differently to maintain communicative coherence. These adjustments depend on inherent characteristics of the modalities. Speaking leaves little room for reflection, typically involves real-time monitoring of their own language output under strict temporal constraints, which may limit opportunities for overt disfluencies during deception. In contrast, the written modality offers the opportunity to pause, evaluate and plan, whenever this is needed, resulting in different patterns of processing adjustments that unfold over longer time scales. Theoretically, these findings contribute to our understanding of how control and monitoring processes regulate language production under competing communicative goals. They suggest that while the same underlying systems for planning and execution are engaged across modalities, their temporal organisation and flexibility differ, offering insight into how speakers and writers manage processing demands when communicative intent is constrained by deception.
4.4. Limitations and future directions
Several methodological limitations should be considered when interpreting the findings of this study. First, participants’ motivation to deceive was relatively low: their only incentive was to make an unknown recipient believe their account. In real-world contexts, deception often occurs under high pressure with strong personal or social consequences. Moreover, in speaking, participants did not engage with a live listener, which may have reduced the social and cognitive demands typically involved in face-to-face communication. Future research could introduce higher stakes deception scenarios or interactive settings to better approximate real-world communicative contexts.
Second, this study focused exclusively on the production side of deception and did not evaluate the quality, credibility or perceived believability of the deceptive output. As such, it remains unclear whether the observed linguistic patterns correspond to successful or unsuccessful deception. Future studies could address this by incorporating rating tasks or lie-detection accuracy measures to examine the relationship between production markers and deception outcomes.
Third, this study is based on elicited data, and as such, there might be differences in how language production processes are affected in real-world cases when people lie. However, by examining elicited data collected in a controlled environment, a baseline for possible alterations in language production processes can be established and then built upon in future research, for example, in ecologically valid contexts – such as forensic interviews or written witness statements – could also help bridge experimental and applied perspectives on deception.
Finally, while this study examined how adopting a deceptive stance influences language production at a global level, future research could move towards a more fine-grained analysis by comparing deceptive and truthful segments within the same narrative. Such an approach would allow for a better understanding of how linguistic and cognitive strategies fluctuate within deceptive accounts and how local shifts in veracity affect production processes. For example, individual or dynamic pause thresholds could be used in a future study.
5. Conclusion
This study demonstrates that adopting a deceptive stance in language production leads to measurable alterations in linguistic processes in both speaking and writing. These findings indicate that deception influences processing demands in complex and modality-specific ways, rather than through a uniform pattern across conditions. The results point to a dynamic interplay between control, monitoring and evolving processing requirements shaped by modality and task characteristics. In speaking, deceptive intent was associated with fewer total pauses – particularly in longer texts – suggesting that speakers adjust their production processes to regulate markers of uncertainty. In writing, deception manifested more consistently across multiple markers, including interactions between pausing, writing time and text length, alongside reduced revision activity. This pattern suggests that writers manage the demands of deception by modulating how planning, monitoring and revision unfold over time.
Together, these findings contribute to our over-arching understanding of language production processes and how this is manifested during speaking and writing. It further underscores the role of modality in shaping deceptive language production and highlights how processing demands are dynamically regulated under competing communicative goals. Rather than being simple indicators of strain, markers of linguistic processing, such as pauses and revisions, appear to reflect ongoing adjustments that align with the communicative intent.
These insights carry important implications for forensic linguistics. Recognising how deceptive behaviour manifests differently in speech and writing can inform the evaluation of suspect statements, witness accounts and testimonies in legal settings. Notably, while more observable markers emerged in written language, these patterns were complex rather than consistent, suggesting that written accounts offer a rich but nuanced window into deception-related processing dynamics and hold potential for advancing forensic analysis.
Data availability statement
The data that support the findings of this study are openly available in the Open Science Framework at https://doi.org/10.17605/OSF.IO/7RQ9G.
Acknowledgements
The authors gratefully acknowledge the support of the Lund University Humanities Lab in preparing the elicitation material, especially Peter Roslund for helping to produce the elicitation films and Johan Frid for discussions about data collection, transcriptions and analyses. They also acknowledge the important contribution of Sigrid Svensson who helped in data collection and transcription and Ingrid Eliasson who helped in validating transcriptions. They further acknowledge the participants who generously helped them by partaking. They also acknowledge that idea for the project originated from discussions in the think tank Intelligent Intelligence at Lund University and are grateful for this connection. They also thank the Marcus and Amalia Wallenberg Foundation who through a generous grant made the whole project possible and the Crafoord foundation who supported the transcriptions and compilations of the texts with an additional grant.
Author contribution
Kajsa Gullberg: conceptualisation, data curation, formal analysis, investigation, methodology, validation, visualisation, writing – original draft, writing – review and editing.
Victoria Johansson: conceptualisation, funding acquisition, methodology, project administration, supervision, validation, writing – original draft, writing – review and editing.
Roger Johansson: conceptualisation, formal analysis, funding acquisition, methodology, supervision, validation, writing – review and editing.
Funding statement
The funders had no role in the design of the study.
Competing interests
The authors declare none.
A. Appendix A: Summaries of the elicitation films
A.1. The beach
02:49 minutes long.
Three persons are on a beach: a man in black, a woman in white and a woman in a bikini and sunhat. The woman in the sunhat sits in a sun lounger reading a magazine and eating an ice cream, and the woman in white sits on a picnic blanket reading a book. The man in black sits on a picnic blanket just looking around, and after a while, he gets up and grabs his tote bag. As he does this, he drops his phone and puts it back in the bag again. But, when he gets up, it falls out again! This time he does not notice and heads off without the phone.
The woman in white notices that he has dropped his phone, she looks around and then gets up and walks to the man’s blanket to grab his phone quickly. She walks back and collects her book, thermos and other belongings and puts them in her bag before grabbing her picnic blanket and shoes to leave. The woman in the sunhat has been watching her this whole time. When the thief has left, the sun lounging woman continues to read.
The man returns with an ice cream in one hand and his tote bag in his other hand. He sits down and continues to look out towards the sea while eating his ice cream. Suddenly, he starts looking for his phone but can find it in his bag, he looks around, feels his pockets and checks the bag again but cannot seem to find it. Finally, he gets up again and heads back in the same direction where he got the ice cream – presumably to see if he has dropped it somewhere on the way.
A.2. The driveway
02:02 minutes long.
A man in a black suit and sunglasses is standing outside a brick house holding a present and some flowers. He is looking around as if he is searching for someone and looks at his wristwatch and makes a rushed phone call; it is not possible to hear what he is saying but he does not look happy. While he is standing there, a young girl comes out from the side of the house with a red bike. She leans the bike against the door to the garage and heads back in before quickly re-emerging with a helmet in hand. She puts her helmet in the basket attached to the bike and then heads back in on the side of the house once again.
While she is gone, the man is still waiting, and he looks more and more annoyed. A hiker wearing a big backpack comes walking along the road. He stops for a moment to look at a map, then looks around to try and figure out where he should go next. He sees the bike and suddenly takes a few quick strides up to the bike and pulls it out to the road, gets on the bike and rides of. The man in a suit sees this and looks in the direct the man went off; he seems perplexed but soon after he waves at someone in the opposite direction and heads off.
When the girl comes out to the driveway again, both her bike and the man in a suit are gone and, in her hand, she has a bike lock that she was going to bring on her ride. She runs up to the street and looks both ways but does see anyone. She steps back again, looking confused and upset.
A.3. The exam
03:09 minutes long.
Two girls are in a classroom, sat across each other writing an exam. At the front of the classroom, a teacher sits watching over them. He has brown hair and is wearing a flannel shirt. The girl to the right is wearing a black shirt and has brown hair, and the other girl is wearing a yellow t-shirt and has blonde hair put up in a ponytail.
The girls are occupied with their exams, and the teacher gets up from his desk to walk around. He brings his green thermos with him and walks in front of the students’ desk looking down at their papers, making sure they are on track and not cheating. When he’s satisfied with the inspection, he heads back to his desk to sit down. He puts his thermos on the desk, and when he corrects his posture after sitting down, he accidentally knocks it over. Coffee spills everywhere! Luckily, he has some napkins in his bag, so he quickly grabs some napkins and kneels to wipe up the spilled coffee.
While he is busy, the girl with the black shirt looks around and then pulls out a note from her pencil case. She looks stressed as she unfolds the note and starts to copy text to her exam paper. She then quickly folds it up again and hides it in her sleeve. The other girl has been looking over at her, shocked, but when the teacher comes back up to the desk, both girls are once again busy writing.
After a short while, the teacher looks up from his desk and exclaims ‘C’est fini’ in French, the exam time is over. The girl in yellow heads up to hand in her exam first before collecting her things to leave. The girl in black packs up and brings all her stuff when she hands in her exam. When she turns around, the note that she had folded up in her sleeve falls out without her noticing. With both girls are gone, the teacher packs up all his things and starts walking towards the door. On his way, he notices a note on the floor and picks it up, he unfolds it and starts reading…
A.4. The garden café
01:58 minutes long.
A young girl reading a magazine at a table while enjoying a cup of coffee and some cake at a garden café. Another young girl, wearing a blue hoodie, comes walking into the café. She walks up to the serving tables where cakes, biscuits and all sorts of drinks are presented. Here, she stops and looks at the different options before heading over to a payment information sign. While she is paying, a man enters the café. He seems stressed and walks quickly, knocking over a chair next to the reading girl, accidentally tipping over her bag.
The man continues to the table, and just as the girl in the hoodie is going to pour her coffee, he cuts in front of her and pours himself a cup and puts on the table. He then proceeds to grab all sorts of cakes and biscuits, overloading his plate, and does not pay for anything. Meanwhile, the reading girl has packed up everything that fell out of her bag and walks up to the table, she sneaks in behind the girl in the hoodie and grabs a pepper mill and quickly grinds some pepper into his cup before quickly leaving the café.
The girl in the hoodie notices this but just grabs her own coffee and sits down at a table. Once the man has collected all his cake and collected his coffee, he sits down at a table. The young girl quickly hurries away as he sits down, and when he takes his first sip of coffee, he instantly spits it out.
B. Appendix B: Instructions
Before the film was shown, participants were informed of the task and which modality it would be in:
‘In this task you will see an events with three persons and then you will re-tell what happened in speech with as many details as possible, from start to finish’
After the film, the instructions were repeated and information about the veracity of the narrative was added.
Both conditions
You have seen an event with three persons on a beach. Two of these persons were active in the events and one saw the events. I would now like you to re-tell what happened with as many details as possible, from start to finish…
Truthful condition
… just as it happened.
Deceitful condition
… but imagine that you for some reason want to protect the person in white and instead lay blame on the person in the sun chair.
Both conditions
Remember that there is a person who has not seen this who will guess if your story is invented or not. There is no right or wrong in how you tell the story, and it is okay to think before you start if you want to.
Open access
