Investigative interviews with online sexual offenders: A discursive analysis of verbal cues of deception

Abstract

The objective of this study is to investigate how individuals convicted of online sexual crimes use language to deceive police investigators during investigative interviews. Discursive and interactional cues that may indicate deception were identified based on the suspects’ responses to questions asked by the police investigator, that is, whether the questions had a high or low potential to elicit deceit. To refine the selection of deceptive answers to be analyzed, the information that was contradicted or questioned during the interview was then targeted for each suspect. The results show that seven elements – grammatical negations, argumentative markers, uncertainty markers, use of the conditional, ignorance markers, sincerity markers, and volubility – distinguish between truthful and deceptive responses. Findings suggest that these elements are used as strategies to convince the investigator of the truthfulness of the (deceptive) statements being made and to avoid providing information that could be refuted during the interview.

Keywords

Child luring child pornography deception investigative interviews online crimes suspects sexual offenders verbal cues

Introduction

Two of the main objectives of the investigative interview are to establish whether the individual being interrogated is the perpetrator or their accomplice for the crime being investigated and to understand the motivations behind the crime. Investigators are interested in obtaining valid, investigation-relevant information, preferably a confession and a detailed account of events (St-Yves and Landry, 2004; Vrij et al., 2014). Confessions are generally defined from a dichotomous perspective in which the suspect either does or does not confess (Bergeron and Deslauriers-Varin, 2020). Recent studies, however, suggest that confession falls on a continuum: an individual may reveal specific details while hiding others or may introduce false information along with true (Laforest, 2020). Investigators are often confronted not with strict deniers but rather with manipulative individuals who distort reality. Examining how suspects use language can thus be expected to reveal useful information that can help distinguish truthful from deceitful information in suspect discourse.

Therefore, this study, resulting from a collaboration between researchers in linguistics and criminology, aims to examine how suspects use language to conceal the truth during investigative interviews. The work falls within the long tradition of studies on lying but deviates from it in its approach: while most studies on the subject have been conducted in the field of psychology (DePaulo et al., 2003; Hartwig et al., 2011; Mann et al., 2002; Vrij et al., 2014), this study is based on forensic linguistics (Coulthard et al., 2020; Gibbons, 2003; Shuy, 1993). The field of forensic linguistics involves the application of linguistic knowledge and theory, in the present case discourse analysis, to forensic, legal, or criminal contexts (Coulthard, 2010; Lagorgette, 2010). According to Shuy (2015), discourse analysis is particularly well-suited for examining verbal interactions during investigative interviews, particularly in the context of detecting deception (see also Shuy, 1998).

Deception: Theoretical overview

One of the most critical aspects of investigative interviewing is obtaining the suspect’s account of the events (St-Yves and Meissner, 2014), which allows the suspect to clarify and explain the situation, an account that can be either partially or entirely truthful. An interviewee may employ a wide range of strategies to avoid self-incriminating, such as denial, blaming someone else, and lying (Feld, 2013). All these strategies contribute to the phenomenon of ‘deception’, which includes not only lying but also providing answers that are partially true or false, failing to reflect the full severity of the facts, or omitting crucial elements (Laforest, 2020). Consequently, ‘deception’ and ‘lying’ will be used interchangeably in this text, despite some terminological distinctions. Lying is commonly defined as making a statement that one believes to be false, with the intent of leading others to accept it as true (Carson, 2010; Mahon, 2008). Deception, on the other hand, encompasses all forms of communication – or lack thereof – aimed at preventing the listener from forming an accurate understanding of relevant events (Laforest, 2012).

The importance of being able to detect lies, particularly those that involve crimes as they have implications for public safety, has led to extensive research on the topic (e.g. DePaulo et al., 2003; Hartwig et al., 2011; Laforest et al., 2020; Mann et al., 2002; Vrij et al., 2014). Most studies of deception detection rely on experimental designs conducted with volunteers, typically university students in psychology, who are asked to produce a narrative of events they have experienced or events they have never experienced but are trying to present as true (i.e. mock interviews and mock suspects; see, e.g. Hartwig et al., 2011; Wachi et al., 2018). One issue with these studies is that the stakes in play in an experimental study are far lower than those for individuals being interrogated by the police. Study participants risk, at worst, being exposed or suffering minor consequences such as having to pay a certain amount of money, while a criminal suspect risks incarceration, or even the death penalty in some countries. The effort and strategies deployed by a liar will, therefore, differ from context to context.

In an attempt to overcome the limitations of laboratory studies, some researchers have focused on lies produced in the context of high-stakes, real-world cases where the consequences for the (alleged) deceiver can be severe: emergency calls (Harpster et al., 2009; Laforest, 2012; Laforest et al., 2020; Miller et al., 2021), investigative interviews with suspects (Mann et al., 2002; Shuy, 1998), criminal trials (Fornaciari and Poesio, 2013), or plea appeals in the media for the return of a missing loved one (Brinke and Porter, 2012; Wright Whelan et al., 2014). The methodologies used in these studies are similar: the discourse of individuals who have provided accurate information is compared to that of individuals who, post-investigation, were found guilty of the alleged crime or of being an accomplice to it. Among the key findings, it was observed that deceptive individuals tend to use more frequent and longer silences compared to truthful individuals (Mann et al., 2002). Their responses are also more likely to include generalizing expressions such as everyone, always, and all the time (Shuy, 1998), are typically shorter in terms of duration and quantity of words uttered, and provide less reliable information (Brinke and Porter, 2012; DePaulo et al., 2003; Shuy, 1998), are less coherent and include more contradictions (Hartwig et al., 2011; Laforest et al., 2020) and contain fewer references to themselves, often avoiding the use of first-person pronouns and determiners such as I, me, and mine (Hancock et al., 2008; Newman et al., 2003).

Other researchers, however, have shown that those who are lying tend to produce longer statements than those who are sincere (Hancock et al., 2008), while others found no difference in the use of singular first-person pronouns between sincere and insincere individuals (Brinke and Porter, 2012; Wright Whelan et al., 2014), or even that insincere individuals use such pronouns more frequently (Fornaciari and Poesio, 2013). These inconsistencies can be explained in several ways, including the language used by the deceivers, the interactional contexts in which the deceit was produced (e.g. monolog vs dialogs), the reaction of the interlocutor toward the lie (suspicious or not), and the operational definition of the studied markers (e.g. what constitutes an ‘informative’ answer?). Furthermore, these studies do not consider the pragmatic and sociolinguistic dimensions of discourse: ‘variation is inherent in individuals’ linguistic practice, and both the communication situation and the age, level of education, social position, and geographical origin of speakers who share the same language can lead to changes in their way of expressing themselves’ [authors’ translation] (Laforest et al., 2020: 195–196).

The present study

Following these findings, the aim of this study is to identify the discursive and interactional markers of deception used by suspects in the context of investigative interviews. To achieve our objective, a discourse analysis was conducted on transcripts of investigative interviews with individuals convicted of online sexual crimes. Online sexual offenders differ from other sexual offenders (Babchishin et al., 2015). They are, on average, younger at the time of the offenses, have a higher level of education, and earn higher salaries – factors that influence the way speakers express themselves (Labov, 2001).

Unlike other professional settings, investigative interviews are focused on information gathering and the pursuit of truth (St-Yves and Landry, 2004). This focus is maintained by the structure of the exchanges – a question-and-answer interaction (the investigator asks the questions, the suspect responds, but not vice versa) – and by the choice and consistency of topics throughout the interview (the committed crime, not the most recent hockey game; Allard-Gaudreau, 2022). The interviews can last for hours and be punctuated by numerous pauses and silences, which is not the case for emergency calls and televised interviews. Furthermore, investigators often show suspicion toward the statements of suspects (Antaki and Stokoe, 2017).

Considering these interactional particularities and the fact that perpetrators of online sexual crimes are generally confronted with numerous pieces of material evidence linking them to the crime (Fortin and Paquette, 2018), it is possible to hypothesize that they will need to employ various discursive strategies to (try to) persuade the investigator to accept their version of events and/or to justify why evidence links them to the crime. The objective of this article is thus to identify the discursive changes that occur when suspects attempt to deceive the investigator about the crime under investigation, and the strategies they use to make their testimony credible and avoid self-incrimination.

Methods

This study is derived from a larger project, the Projet de Recherche sur l’Exploitation Sexuelle des Enfants en Ligne (PRESEL). This project takes a psychological and criminological approach to studying online sexual crimes using data from Québec, a French-speaking province of Canada. One of the objectives of this larger research was to identify the factors that impact suspects’ decisions to confess or not to their crime during the investigative interview, as well as to establish the socio-demographic and offending profile of ‘deniers’ – those who have committed a crime but deny it throughout the interview (Deslauriers-Varin and Bergeron, 2021). A second key objective of this project was to gain a deeper understanding of the characteristics of men involved in the online sexual exploitation of children, with the aim of identifying risk factors for offending behavior and developing effective prevention and intervention strategies. This project has made it possible to collect several information about adult males convicted of online sexual crimes and their offenses, including information derived from 74 video-taped police investigative interviews. Based on this larger sample, 21 transcriptions of interviews of individuals convicted of at least one online sexual crime – either child pornography, child luring, or both – were available for the present study.

Since our objective was to investigate how these individuals use language when they try to deceive the interrogator, transcripts in which suspects remain silent or confess immediately, acknowledging responsibility and the reprehensible nature of their actions, were excluded (n = 10). The final data set consisted of transcriptions of 11 video-taped investigative interviews. As mentioned in the introduction, this article examines not only suspects who deny committing the alleged crime but also those who minimize its severity or fail to disclose all relevant information. Consequently, in many instances, suspects do not outright deny committing the crime but instead may, for example, acknowledge having committed it only once with a single victim despite evidence from the investigation indicating multiple victims. In this regard, all the interviews in our corpus include the disclosure of evidence linking the suspects to the crimes committed, such as sexual conversations with children (via their personal email address, personal ID or cell phone), images showing them masturbating in front of children, or pedo-pornographic images stored on their personal computers. It should be noted that after investigators disclose this material evidence, all suspects eventually acknowledge, at least in part, having committed the acts they are accused of (although they may, e.g. continue to lie about their motivations or intentions).

All interviews were conducted in French by investigators who specialize in online sexual crimes and analysis was done using the original interviews. Selected excerpts were translated into English by the authors for this article, with the translations validated by an accredited translator (see the original transcripts in French in the Appendix). It should be noted that analyzing non-verbal behaviors was impossible since they were not consigned within the text.

Participants

Suspects had an average age of 34.36 years (SD = 14.47) at the time of the interview and ranged in age from 20 to 63 years old. One out of the 11 participants had a university degree, three had had college or vocational training (e.g. plumbing), and six had high-school level education or lower. All suspects were male. One had been convicted of child pornography, six of child luring (i.e. online sexual solicitation of minors), and four had been convicted of both child pornography and child luring. Five out of the 11 suspects had also been convicted of other sexual crimes, including offline sexual offending involving children or sexual incitement. Five of the 11 suspects had prior criminal records.

Selection of deceptive statements

Because not all parts of a false testimony are necessarily deceitful, we chose to compare, for each suspect, the moments when they are likely to lie to the moments when they are likely to tell the truth. This approach is based on the research conducted by Vincent et al. (2007) in which the authors compared answers in which interviewees tell the truth with answers in which they conceal reality, rather than comparing the discourse of sincere individuals with that of insincere individuals as previously done (see e.g. Brinke and Porter, 2012; Hartwig et al., 2011; Laforest et al., 2020; Mann et al., 2002). In their study, Vincent et al. (2007) categorized the analyzed discourses into responses to two types of questions: those that potentially promote deceit due to their interpretation as preludes to criticism or negative judgment, thereby threatening the interviewee’s face (i.e. the social image an individual maintains), and those that do not encourage deception because truth-telling does not jeopardize the speaker’s face. Vincent et al. (2007) have shown that potentially threatening questions for interviewees are those in which an honest answer would imply the transgression of social norms, such as when a police investigator asks a suspect if he is sexually attracted to children as opposed to asking him how many children he has.

Building on the research by Vincent et al. (2007, the transcripts were first categorized based on the nature of the questions asked by police investigators: (1) questions that strongly incite deception, as they pertain to taboo subjects where truthful answers may lead to negative judgment, potential rebuke, or, in this study’s context, risk incriminating the suspect or compromising their account of events; and (2) questions that weakly incite deception, which are those that, if answered truthfully, do not breach social norms and are neither incriminating nor compromising.

Looking at online crimes means that numerous pieces of physical evidence were collected prior to the interview (e.g. images, conversations, IP addresses). Consequently, the second phase of this project was to compare the suspect responses with the available evidence. To strengthen the classification of answers based on their potential to be deceptive, all transcripts were read thoroughly to identify contradictions or inconsistencies in each suspect’s discourse. For instance, several suspects asserted that they were unaware the downloaded images depicted children, despite the file name indicating that the subjects were minors (e.g. 13yo.lolita.pthc.jpeg; see Fortin and Paquette, 2018). These contradictions or inconsistencies were then used as indicators of deception, as suggested by several authors (Hartwig et al., 2011; Laforest, 2012; Shuy, 1998). Moreover, using a sample of convicted offenders – all participants were found guilty at the end of the investigation – made it possible to assert with greater certainty that, for example, a suspect’s denial of having committed the acts in question was a lie.

While a lie can be analyzed in terms of the question asked, it can also be identified based on the reaction it provokes in the interlocutor (Laforest, 2020). The third and final phase of the analysis involved identifying responses whose veracity was questioned or doubted at some point during the interview by the police investigator – considered a reliable indicator that the information is likely deceptive (Antaki and Stokoe, 2017). The investigator’s reaction is particularly relevant when the information provided cannot be corroborated by physical evidence, such as the suspects’ intentions or motivations.

This approach allowed us to specifically target (potentially) deceptive information in the discourse of each suspect in the corpus.

Data extraction and analysis

Because interviews vary in length and since the majority of transcripts comprise dozens, sometimes hundreds, of pages, it was decided to analyze 25 answers ‘with a weak potential to be deceptive’ and 25 answers ‘with a strong potential to be deceptive’ for each. As mentioned, answers ‘with a strong potential to be deceptive’ were identified from the transcripts based on the topic addressed in the question – whether it was incriminating or not –, the presence of contradictions or inconsistencies in the suspect’s testimony, and/or the investigator’s suspicion regarding the response given. Excerpt 1 provides an example of an answer classified as having a strong potential to be deceptive based on these indicators (I for investigator; S for suspect; all names are fictional):

Excerpt 1_Louis [suspect arranged meetings with underage girls but says he never went]

I Are there any other [chats] that I’m going to find, [about] the kind of meetings that you didn’t go on? There could be, couldn’t there?

S There might be. I don’t know.

I You don’t seem to be convinced, you’re not convincing me either.

Conversely, answers were identified as having a weak potential to be deceptive based on the banality of the topics covered in the question–for example, questions about the suspect’s job, hobbies, and family. These types of questions are generally asked at the beginning of the interview and are intended to create a climate of trust and help the suspect relax (St-Yves and Landry, 2004). These answers were also selected because the content was not contradicted or questioned by the investigator.

This procedure allowed for the analysis of 50 responses – or speaking turns – per suspect, which for many suspects encompassed most of the responses with high or low potential to be deceptive that met our selection criteria. In total, 550 responses were analyzed, with 275 falling into the category of ‘weak potential to be deceptive’ and 275 into the category of ‘strong potential to be deceptive’. This sample size is sufficient to provide a comprehensive understanding of how suspects use language when they are (potentially) deceptive or not. The responses average 40 words, with the shortest consisting of a single word (e.g. yes or no) and the longest consisting of 852 words.

Following data extraction (i.e. 550 speaking turns), responses classified as having a low potential to be deceptive were compared with responses classified as having a high potential to be deceptive in order to identify any differentiating elements, that is, any elements likely to be present much more often in one group than in the other. The approach adopted is the classic ‘bottom-up’ approach to data, which allows discursive phenomena to emerge during data analysis. The analysis carried out was therefore inductive, with the objective of investigating how suspects in the data set used language to deceive the police investigator about some aspect of their narration, rather than, as with ‘top-down’ approaches, starting with pre-existing categories and searching for examples of them in the corpus (e.g. silences; Clift and Haugh, 2021).

Discourse analysis as performed in the present study investigates the meaning of discourse as it unfolds in interaction, one utterance after another, within a specific social context (Bhatia et al., 2008; Handford and Gee, 2023). Unlike content analysis, discourse analysis does not focus solely on what is said in a specific context but also on how it is said and the effect that these words produce on the interaction (Allard-Gaudreau, 2022). The analyses performed therefore consider both the discursive and interactional dimensions of the data examined (Laforest, 2012).

For each analyzed response, the presence or absence of each differentiating element that emerged from our analysis was recorded as 0 = absent, 1 = present. To determine if the occurrence of these elements was significantly different for the two types of answers, paired t-tests were conducted using SPSS 28 software. To measure the extent of the association between the studied variables and sincerity and non-sincerity, partial eta-squared (η2) was calculated. Values of 0.01 indicate a small effect size, 0.06 a medium effect, and 0.14 a large effect (Cohen, 1988).

Results

Our findings indicate that there are noticeable differences in how suspects use language when they are truthful or deceitful. The differentiating elements were grouped into seven categories: (1) grammatical negations, (2) argumentative markers, (3) uncertainty markers, (4) use of conditional/hypothetical, (5) ignorance markers, (6) ‘sincerity’ markers, and (7) volubility. A response may simultaneously contain multiple categories of markers, for example, negations and conditional verbs, as well as multiple markers from the same category (e.g. several negations within the same response). However, a single occurrence per category was recorded, based on its presence (1) or absence (0) (e.g. as soon as there was a ‘grammatical negation’, the response was considered ‘grammatically negative’). Analysis suggests that the frequency with which markers are used is indicative of an intention to deceive. The presence of only one of these elements should not, therefore, be considered a necessary sign of insincerity, as a truthful response can contain one (or more) of these elements.

Since the analyses were inductive, some of these categories were created for the purposes of this study. Indeed, elements contributing to the same discursive or interactional phenomenon, which, to our knowledge, had not previously been the subject of empirical study, were grouped into novel categories. This is the case, for example, with expressions indicating the speaker’s lack of knowledge about the requested information, which was grouped into a category we have termed ‘ignorance markers’. Other elements were categorized based on existing concepts, such as ‘uncertainty markers’ [or ‘hedges’]. This approach enabled the identification of novel categories of discursive markers not previously associated with deception. Consequently, each category, except for the volubility, that is, the number of words per response, will be qualitatively illustrated in the following sections.

Categories of markers

1. Grammatical negations are sentence structures that in French include ne and pas or ne and another negative element, such as jamais or rien (‘never’ or ‘nothing’). Unlike other studies, words with negative prefixes such as UNappetizing, and UNreasonable, or words with negative connotations like ugly, and bad, were excluded from the analysis. Responses consisting of a single ‘no’ were not considered ‘grammatical negations’; instead, the focus of the analysis was on the discursive elaboration that followed the utterance of this minimal response element (e.g. ‘Do you have brothers?’ – ‘No, I only have two sisters, but I don’t see them often ’).

Excerpt 2 in which the suspect denies having accessed websites containing child sexual exploitation material and being sexually attracted to minors, is an example of a potentially deceptive response that contains grammatical negations (indicated in bold in the text):

Excerpt 2_Jacques [answer with a strong potential to be deceptive]

I Have you ever visited child pornography sites?

S No, [. . .] I’m not interested, not at all come on, I was saying earlier I’m not a pedophile and I’m not interested in children

As observed in Excerpt 2, negations enable suspects to deny facts, events, and even states, likely anticipating subsequent questions – such as those concerning potential sexual preferences in this case. By specifying what they have not done or said (or describing the kind of person they are not), suspects avoid detailing the events that occurred.

2. Argumentative markers, in this research, are lexical items that express either causality (because, due to), opposition (but, however), or consequence (therefore, so (used to indicate consequently)), between (counter-)arguments and conclusions.

Excerpt 3 was produced by a highly reluctant suspect to provide the sought information and who systematically minimizes the actions committed, while the evidence has shown that the acts were much more severe than what the suspect implies.

Excerpt 3_Stefan [answer with a strong potential to be deceptive]

I And has it happened that, have there been times when you have masturbated [in front of his webcam]?

S Yes, but not for long, not for long there, the webcam, but not for long, <correct> it happened, but you know, not even, not even a minute [. . .].

The use of argumentative markers can sometimes indicate, as in this excerpt, a concession or even opposition to the investigator’s line of questioning (‘yes, but. . .’). These markers thus allow suspects to justify their actions (or inaction) and, more broadly, to influence the investigator toward accepting their version of events.

3. Uncertainty Markers are, in this study, expressions that present an element of information as not necessarily true or false, such as I’m not certain/sure, I believe, I imagine, I suppose, and maybe. They correspond to the notion of hedges in English-language literature (in French, they are referred to as ‘marqueurs d’incertitude’).

Excerpt 4, in which the suspect claims to be unsure about the age of the little girl seen in the photo he sent to an acquaintance, even though he had stated she was a minor just minutes earlier, is an example of this.

Excerpt 4_Eric [answer with a strong potential to be deceptive]

I A girl [who appears in the photo the suspect sent to an acquaintance] about how old?

S Well, I don’t know, I think you can’t even see her face, <ok> but you know, she must be over 18, otherwise um could there be anything else? No, <correct> I don’t think so.

This kind of marker is useful when the suspect is providing information that he does not fully endorse or whose truthfulness he cannot guarantee (‘I think we can’t even see her face’ vs ‘we can’t even see her face’). These markers also contribute to making the suspects’ speech imprecise (‘she must be an adult’ vs ‘she is an adult’).

4. Use of the conditional verbs in French ending in rais (e.g. j’aurais, the equivalent of would in English), and the hypothetical, usually expressed in English by the adverb if or si in French.

In excerpt 5, the suspect justifies his possession of numerous files of juvenile pornography by stating that he did this in order to catch ‘real pedophiles’ and report them to the authorities (his mens rea); a version of events that is strongly questioned by the investigator.

Excerpt 5_Mark [answer with a strong potential to be deceptive]

I For the past two years, um, have you ever provided any information to the police or something like that regarding um, certain places or um?

S I would have, if I had managed to catch to get one like that, it would have already been done <ok> but, they [pedophiles] are too hard they are too hard to catch <ok> even you guys [police officers] have a hard time, you know [. . .].

In this excerpt, as in several other excerpts that illustrate deceptive statements, the use of this category of markers seems to indicate (1) the hypothetical nature of a situation and (2) the suspect’s lack of commitment to their statements (this is especially the case in French, e.g. ‘je ne pense rais pas’ [I wouldn’t think so]), much like the uncertainty markers discussed previously.

5. Ignorance Markers, as elaborated in this study, refer to an inability to provide the requested information, often attributed to a lack of knowledge or a memory defect. The analysis helped identify relevant expressions such as not knowing, not remembering, having no knowledge, not being aware, or having no idea.

Excerpt 6, the suspect claims not to remember if he discussed sexuality with one of his underage victims, a statement the investigator refutes later in the interview by reading excerpts of conversations the suspect had had with the victim, prompting the suspect’s memory to improve.

Excerpt 6_Eric [answer with a strong potential to be deceptive]

I Do you remember talking about sex with her [victim 1]?

S I honestly don’t remember that

...

I Did you talk about sex together? Take your time before answering.

S Look, I’m not going to lie to you, I don’t remember.

While the claim of ignorance may be accurate in some cases as suspects may not have the answers to all the questions asked, in many cases, it appears to be used as a strategy to avoid having to provide compromising information.

6. ‘Sincerity’ Markers, in the present analysis, correspond to expressions such as I swear, frankly, honestly, that’s the truth, I’m not lying to you, or I’m being sincere with you that are used to indicate the (alleged) sincerity of the speaker and attest that what he is saying is ‘true’ and ‘authentic’.

In excerpt 7, it is noticeable that the suspect relies heavily on ‘‘sincerity’ markers’ to try to convince the investigator that he asked for the age and prior consent of the people he talked about sex on the internet, a claim that is later invalidated during the interview.

Excerpt 7_Sebastien [answer with a strong potential to be deceptive ]

I And did you um, did you masturbate, did you ever do stuff um [on his webcam]?

S I’ll tell you frankly yes, <ok> but me as I told you I always asked people if they were of age, <ok> when they were underage, I tried to I tried to go further, to be really sure that she was that age [. . .] I’m being honest with you, I’m telling you what it’s like.

Analysis showed that sincerity markers were often found in answers that contained both hypothetical formulas or verbs in the conditional tense and ignorance markers, as seen in excerpt 8.

Excerpt 8_Maxime [answer with a strong potential to be deceptive]

I Do you remember when you bought that computer? A couple of years or?

S Ah maybe not so much, I don’t remember, I’d say maybe a year or a year and a half, I honestly don’t know.

The combination of ‘Ignorance markers’ and ‘sincerity markers’ is one of the most frequently found combinations in this category (see also excerpt 6, ‘I honestly don’t remember’ and ‘I’m not going to lie to you, I don’t remember’). This combination appears to be used to reinforce the fact that suspects are unable to provide the requested information on the pretext that they ‘sincerely’ do not know the answer or not remember it.

Comparative analyses

As can be seen in Table 1, paired t-test analysis indicates that the seven categories of markers differentiate significantly between potentially sincere and potentially deceptive responses, with markers used more frequently in answers with a strong potential to be deceptive than in answers with a low potential to be deceptive: grammatical negations (75% vs 26%), argumentative markers (49% vs 25%), uncertainty markers (24% vs 4%), use of conditional/hypothetical (25% vs 3%), ignorance markers (24% vs 4%), ‘sincerity’ markers (15% vs 1%), and volubility (56 vs 24 words per turn). In other words, when suspects were likely to lie, they tended to produce a greater number of answers containing one or more of the elements analyzed, sometimes 15 times more often, as in the case of ‘‘sincerity’ markers’.

Table 1.

Percentage of responses containing the categories of markers analyzed according to their potential to be deceptive or not (M ± SD).

Categories	Types of answers		T test
Categories	Weak potential to be deceptive (n = 275)	Strong potential to be deceptive (n = 275)	t	p	η ²
Grammatical negations	26.18 (11.64)	75.27 (14.62)	13.263	<0.001	0.95
Argumentative markers	25.09 (15.60)	49.09 (18.60)	4.914	<0.001	0.71
Uncertainty markers	7.64 (9.37)	25.09 (13.75)	6.314	<0.001	0.80
Conditional/hypothetical	2.91 (3.14)	24.73 (14.40)	5.700	<0.001	0.76
Ignorance markers	3.64 (4.54)	23.64 (13.20)	4.599	<0.001	0.68
‘Sincerity’ markers	1.45 (3.70)	14.55 (12.17)	4.097	0.002	0.63
Volubility^a	24.33 (15.33)	55.91 (34.92)	4.143	0.002	0.63

Results for volubility are presented as a function of the mean number of words uttered per response.

Discussion

This study focused on investigative interviews with individuals who, when interviewed about online sexual crimes, attempted to deceive the police investigator about the commission of the crime, specific actions, or their intentions. Drawing on the methodology outlined by Vincent et al. (2007), a comparative analysis was conducted for each suspect regarding the way they used the language when responding to questions with either weak or strong potential to incite lying regarding the subject discussed (i.e. incriminating or not). To enhance the classification, responses categorized as ‘potentially deceptive’ were determined by information that either caused doubt in the investigator about its truthfulness or was contradicted at some point during the interview.

Our results indicate that responses deemed potentially sincere could be differentiated from those considered potentially insincere using seven categories of markers: grammatical negations, argumentative markers, uncertainty markers, use of conditional/hypothetical, ignorance markers, ‘sincerity’ markers, and volubility. Our analyses showed that these markers are used much more frequently when suspects are trying to conceal reality than when they are telling the truth. Findings also showed that the number of these markers was much greater in deceptive responses than in truthful ones, suggesting that deceptive responses often contain multiple markers from both the same and different categories (see, among others, excerpt 4, which contains ‘negations’, ‘argumentative markers’, ‘ignorance markers’ and ‘uncertainty markers’). This explains in part why deceptive responses in the current study tended to contain a greater number of words than sincere responses – a result that differs from other studies on the subject (DePaulo et al., 2003; Mann et al., 2002; Vrij et al., 2011, 2021).

Indeed, producing longer statements in terms of the number of words spoken allows speakers to produce more complex and richer discourse, which helps account for some of the differences observed in this study. However, the analyses indicate that, even when the volume of speech is similar, the two types of responses manifest differently. For example, excerpt 9, categorized as a potentially non-deceptive response, contains 46 words – comparable to many of the excerpts presented earlier – but does not include any of the markers analyzed in the study (see also excerpt 5 produced by the same participant):

Excerpt 9_Mark [answer with a weak potential to be deceptive]

I And, uh, were you raised on a farm, on a piece of land?S Oh no, no <no> no, my father worked in the sawmills <ok> my mother too, and then, uh, when the mill closed, that’s when they decided to move to the city, to find work. After that, when they found something, they moved the family, that’s it.

Given that online crimes leave substantial material evidence linking offenders to the crime (Fortin and Paquette, 2018), our hypothesis was that they would use a range of discursive strategies to attempt to persuade the investigator to accept their version of events and/or to explain why evidence connects them to the crime. As shown by Deslauriers-Varin et al. (2011), the suspects’ perception of the quality of the evidence against them is a key factor that heavily influences the decision of suspects to either confess to or deny the crime. With the abundance of evidence gathered, deceptive suspects of online sexual offences can hardly deny committing the alleged crime – which is not the case for other types of crime such as offline sexual abuses whose evidence often relies solely on the testimony of the child victim (Cyr, 2023). Therefore, they must employ various strategies to avoid providing incriminating information or to prevent presenting the facts as they actually occurred. This is what our results tend to indicate, thus confirming our hypothesis. Indeed, our analysis suggests that the use of these seven categories of markers serves two main functions: to try to convince the investigator that the (false) information provided is genuine, and to distance themselves from it from an enunciative perspective, that is, to disclaim responsibility for the propositional content stated.

The analyses show that suspects who were lying about the crime being investigated tended to resort to different strategies to convince the investigator that what they were saying was true. The usage of ‘‘sincerity’ markers’, which include expressions meant to convey good faith and honesty (e.g. frankly, I’m not lying to you, I swear to you), supports this idea. To our knowledge, such expressions have never been associated with deception (see Table 1). Tseronis (2010) argues that such expressions can play a strategic role when they are used to qualify a thesis. For the author, these expressions can be seen as a way for the speaker to present their thesis in a manner that manages the burden of proof to their advantage – in a manner to appear more convincing. Argumentative markers such as because, but and so also appear to be an explicit sign of a suspect’s intention to persuade the investigator to accept a thesis or a conclusion, as suggested by Tseronis (2011). These two categories seem therefore to have an inherently argumentative function, reinforcing the idea that suspects of online crimes must make greater efforts to be believed than participants in laboratory studies on lying–and probably than other types of sexual offenders.

Our findings also suggest that suspects trying to deceive the investigator employed diverse strategies to distance themselves from parts of the information they provided, presenting it as potentially either true or false or claiming ignorance about it. These results are consistent with those of Hartwig et al. (2011), who showed that liars are prone to verbal evasiveness. For instance, deceptive responses are characterized by the more frequent use of ‘ignorance markers’ such as I don’t know and I don’t remember and ‘uncertainty markers’ such as I think and I’m not sure, which allow suspects to provide partial, uncertain, or vague information (or even no information at all). This is also the case with certain uses of the conditional tense. Indeed, the conditional tense can be used to express hesitation, doubt, and indecision (Dendale, 1993). Like uncertainty markers, in French, the conditional tense is part of the epistemic modality used to indicate the speaker’s level of commitment to the reliability of what is being stated. The findings suggest that employing these strategies makes it possible for suspects to avoid providing potentially incriminating information, anticipating that the investigator will contradict or refute the information provided (e.g. I: ‘You told me earlier that. . .’–S: ‘No, I said that I thought that. . .’) or anticipating more broadly negative consequences (such as imprisonment). These markers thus appear to have an avoidance function. The same is true of several grammatical negations and hypothetical structures. Rather than stating what happened, suspects often reported what did not happen or what could have happened if a certain thing had occurred or not.

Previous studies have suggested other clues that indicate that the suspect is lying (e.g. the use of generalizing expressions, silences), but these were not found in the current research, for several reasons. First, the transcripts lacked indications of the suspects’ nonverbal responses or of (micro)pauses and their duration. Consequently, counting silences, which has previously proven to be a reliable method of identifying deception (Laforest, 2012; Mann et al., 2002), was not possible. Second, as our analysis was done on a limited number of excerpts, it was impossible to calculate contradictions and inconsistencies (Hartwig et al., 2011; Laforest et al., 2020) as these are usually apparent in the testimony as a whole rather than within the same response. The same problem arose in considering the amount of information provided (Shuy, 1998), as this is difficult to measure based on a small number of extracts versus the entire testimony. Other previously reported clues, such as the use of first-person pronouns (Hancock et al., 2008; Newman et al., 2003) and generalizing expressions (Shuy, 1998), were not noted in any particular way.

Despite the encouraging results discussed here, further studies are needed to better understand and identify the discursive and interactional changes that occur when suspects lie during investigative interviews. One way to validate our results could be to compare the responses produced by suspects who, at the end of the investigation, were cleared of the suspicions against them, to those of the suspects in the present corpus. It is possible, for instance, that the use of expressions such as I swear and honestly is not necessarily a sign of lying but rather a genuine intent to convince the police investigators.

Conclusion

This study is one of the few to examine actual investigative interviews, employing tools and methodologies unique to forensic linguistics, in this case, applied discourse analysis. It highlights the importance of closely monitoring how suspects use language during the interview process, particularly if the investigator suspects that the interviewee is lying. Overall, our results show that suspects who try to persuade the investigator of the authenticity of their narrative while simultaneously distancing themselves from the information provided are likely being deceptive.

Footnotes

Appendix

Acknowledgements

The authors thank the professional reviewer who translated the excerpts presented in this study and conducted a linguistic review of the article.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study is part of a larger research project that received funding from the Social Sciences and Humanities Research Council of Canada (430-0268-2019).

Ethical considerations

This research has received the approval of CÉRUL 2019-319 R-4 / 04-11-2024.

ORCID iD

Noémie Allard-Gaudreau

Author biographies

Noémie Allard-Gaudreau is a postdoctoral fellow at the School of Social Work and Criminology at Université Laval. Holding a Ph.D. in Letters and Social Communication, she has been working for nearly ten years on investigative interviews conducted with victims and suspects of sexual crimes against children. Her research focuses on various aspects of language use by interviewees, including the credibility of the information provided, reluctance to answer questions, and deception. Since 2022, she has also been teaching at the International Summer School in Forensic Linguistics, which takes place annually, alternating between Quebec (Canada) and France.

Nadine Deslauriers-Varin is Professor of Criminology at Université Laval, QC, Canada. She is the Associate Director of the Centre International de Criminologie Comparée and Co-Chair of the Scientific Committee for the International Investigative Interviewing Research Group. She also serves on the editorial board of several scientific journals.Most of her research projects are conducted in collaboration with police agencies and focus on improving investigative practices, particularly suspect interviewing, as well as enhancing the investigation and prevention of sexual, violent, and serial crimes. She has recently trained nearly 200 police investigators on best practices in investigative interviewing with suspects and the factors that promote the elicitation of investigative-relevant information.

Sarah Paquette is a Senior Lecturer in Forensic Psychology at the University of Portsmouth (United Kingdom). Previously, she served as a Scientific Researcher specializing in sexual violence within the Internet Child Exploitation (ICE) unit of the Sûreté du Québec, the provincial police force in Quebec, Canada. Her research focuses on identifying risk factors associated with online and contact sexual offenses against children, as well as enhancing police investigation and interrogation techniques. Her projects aim to develop best practices for preventing child exploitation, create effective intervention strategies, and devise systematic case prioritization tools for law enforcement.

Francis Fortin has been working in the field of criminal investigations and intelligence since 1999. He is an associate professor at the School of Criminology at the University of Montreal. His research focuses on cybercrime, criminal intelligence, data mining, and forensic analysis. He has taught specialized courses on emerging criminal topics at several universities in Quebec, including classes on conducting cyber investigations and criminal intelligence.

References

Allard-Gaudreau

(2022) Entrevues d’enquête après d’enfants victimes d’agression sexuelle: une analyse multdimensionnelle du discours des victimes [Investigative interviews with child victims of sexual assault: A multidimensional analysis of victims’ discourse]. (Doctorat thesis, Université du Québec à Trois-Rivière).

Antaki

Stokoe

(2017) When police treat straightforward answers as uncooperative. Journal of Pragmatics 117: 1–15.

Babchishin

Hanson

VanZuylen

(2015) Online child pornography offenders are different: A meta-analysis of the characteristics of online and offline sex offenders against children. Archives of Sexual Behavior 44(1): 45–66.

Bergeron

Deslauriers-Varin

(2020) Obstacles à la confession : exploration des profils explicatifs de la non-confession en contexte d’interrogatoire policier [Obstacles to Confession: Exploring Explanatory Profiles of Non-Confession in Police Interrogation Contexts]. Revue internationale de criminologie et de police technique et scientifique 72: 387–408.

Bhatia

Flowerdew

Jones

(2008) Approaches to discourses analysis. In: Bhatia

Flowerdew

Jones

(eds) Advances in Discourse Studies. London: Routledge, pp.1–18.

Brinke

Porter

(2012) Cry me a river: Identifying the behavioral consequences of extremely high-stakes interpersonal deception. Law and Human Behavior 36(6): 469–477.

Carson

(2010) Lying and Deception: Theory and Practice. Oxford: Oxford University Press.

Clift

Haugh

(2021) Conversation analysis and sociopragmatics. In: Haugh

Kádár

Terkourafi

(eds) Cambridge Handbook of Sociopragmatics. Cambridge: Cambridge University Press, pp.617–638.

Cohen

(ed.) (1988) Statistical Power Analysis for the Behavioral Sciences, 2nd edn. New York: Erlbaum.

10.

Coulthard

(2010) Forensic Linguistics: The application of language description in legal contexts. Langage et société 132(2): 15–33.

11.

Coulthard

May

Sousa-Silva

(eds) (2020) The Routledge Handbook of Forensic Linguistics, 2nd edn. London: Routledge.

12.

Cyr

(ed.) (2023) Recueillir la parole de l’enfant témoin ou victime: de la théorie à la pratique [Gathering the speech of the child witness or victim: From theory to practice], 2nd edn. Malakoff: Dunod.

13.

Dendale

(1993) Le conditionnel de l’information incertaine: marqueur modal ou marqueur évidentiel? [The conditional of uncertain information: Modal marker or evidential marker?] In: XXe Congrès International de Linguistique et Philologie Romanes, Zurich, 6–11 April 1992.

14.

DePaulo

Lindsay

Malone

, et al (2003) Cues to deception. Psychological Bulletin 129(1): 74–118.

15.

Deslauriers-Varin

Bergeron

(2021) Suspect interviews and confessions in sex crime cases. Criminal Investigations of Sexual Offenses: Techniques and Challenges, 99–113.

16.

Deslauriers-Varin

Lussier

St-Yves

(2011) Confessing their crime: Factors influencing the offender’s decision to confess to the police. Justice Quarterly 28(1): 113–145.

17.

Feld

(2013) Real interrogation: What actually happens when cops question kids. Law & Society Review 47(1): 1–36.

18.

Fornaciari

Poesio

(2013) Automatic deception detection in Italian court cases. Artificial Intelligence and Law 21(3): 303–340.

19.

Fortin

Paquette

(2018) Online sexual exploitation of children. In: Lussier

Beauregard

(eds) Sexual offending: A criminological perspective. London: Routledge, pp. 237–256.

20.

Gibbons

(2003) Forensic Linguistics: An Introduction to Language in the Justice System. Oxford: Blackwell.

21.

Hancock

Curry

Goorha

, et al (2008) On lying and being lied to: A linguistic analysis of deception in computer-mediated communication. Discourse Processes 45(1): 1–23.

22.

Handford

Gee

(2023) The Routledge Handbook of Discourse Analysis. New York: Routledge.

23.

Harpster

Adams

Jarvis

(2009) Analyzing 911 homicide calls for indicators of guilt or innocence: An exploratory analysis. Homicide Studies 13(1): 69–93.

24.

Hartwig

Granhag

Stromwall

, et al (2011) Detecting deception in suspects: Verbal cues as a function of interview strategy. Psychology Crime and Law 17(7): 643–656.

25.

Labov

(2001) Principles of Linguistic Change. Social Factors. Oxford: Blackwell.

26.

Laforest

(2012) The false report during an emergency call: Using discourse analysis to detect deceit. In: Tomblin

MacLeod

Sousa-Silva

Coulthard

(eds) Proceedings of The International Association of Forensic Linguists’ Tenth Biennial Conference, pp.139–152. Birmingham: Center for Forensic Linguistics.

27.

Laforest

(2020) Peut-on détecter le mensonge? L’exemple des appels d’urgence [Can deception be detected? The example of emergency calls]. In: Denault

(ed.) La science au service de la pratique [Science in the Service of Practice]. Montréal: Éditions Yvons Blais, pp.35–54.

28.

Laforest

Rioux-Turcotte

St-Yves

(2020) Détecter l’appelant dissimulateur au service d’urgence 9-1-1: Une analyse discursive et interactionnelle de la tromperie [Detecting the deceptive caller in 9-1-1 emergency services: A discursive and interactional analysis of deception]. Criminologie 53(2): 193–218.

29.

Lagorgette

(2010) Présentation. Linguistique légale et demande sociale: les linguistes au tribunal [Presentation. Forensic linguistics and social demand: Linguists in the courtroom]. Langage et société 132(2): 5–14.

30.

Mahon

(2008) The Definition of Lying and Deception. Stanford Encyclopedia of Philosophy. Available at: https://plato.stanford.edu/entries/lying-definition/?ref=ameasuredapproach.blog (accessed September 2024).

31.

Mann

Vrij

Bull

(2002) Suspects, lies, and videotape: An analysis of authentic high-stake liars. Law and Human Behavior 26(3): 365–376.

32.

Miller

Merola

Opanashuk

, et al (2021) “911 What’s your emergency?”: Deception in 911 homicide and homicide staged as suicide calls. Homicide Studies 25(3): 256–272.

33.

Newman

Pennebaker

Berry

, et al (2003) Lying words: Predicting deception from linguistic style. Personality and Social Psychology Bulletin 29(5): 665–675.

34.

Shuy

(1993) Language Crimes: The Use and Abuse of Language Evidence in the Courtroom. Cambridge: Blackwell.

35.

Shuy

(1998) The Language of Confession, Interrogation and Deception. London: Sage.

36.

Shuy

(2015) Discourse analysis in the legal context. In: Tannen

Hamilton

Schiffrin

(eds) The Handbook of Discourse Analysis, vol. 2. Oxford: John Wiley & Sons, pp.822–840.

37.

St-Yves

Landry

(2004) Psychologie des entrevues d’enquête : De la recherche à la pratique [Psychology of investigative interviews: from research to practice]. Montréal: Éditions Yvon Blais.

38.

St-Yves

Meissner

(2014) Interviewing suspects. In: St-Yves

(ed.) Investigative Interviewing: The Essentials. Montréal: Éditions Yvon Blais, pp.153–198.

39.

Tseronis

(2010) Les marqueurs d’une stratégie de gestion de la charge de la preuve : “franchement” et “en fait” qualifiant une thèse [The markers of a strategy for managing the burden of proof: ‘frankly’ and ‘in fact’ qualifying a thesis]. Revue Verbum 32(1): 73–91.

40.

Tseronis

(2011) From connectives to argumentative markers: A quest for markers of argumentative moves and of related aspects of argumentative discourse. Argumentation 25(4): 427–447.

41.

Vincent

Laforest

Bergeron

(2007) Lies, rebukes and social norms: on the unspeakable in interactions with health-care professionals. Discourse Studies 9(2): 226–245.

42.

Vrij

Deeb

Leal

, et al (2021) Plausibility: A verbal cue to veracity worth examining? European Journal of Psychology Applied to Legal Context 13(2): 47–53.

43.

Vrij

Hope

Fisher

(2014) Eliciting reliable information in investigative interviews. Policy Insights from the Behavioral and Brain Sciences 1(1): 129–136.

44.

Vrij

Leal

Mann

, et al (2011) A comparison between lying about intentions and past activities: Verbal cues and detection accuracy. Applied Cognitive Psychology 25(2): 212–218.

45.

Wachi

Kuraishi

Watanabe

, et al (2018) Effects of rapport building on confessions in an experimental paradigm. Psychology Public Policy and Law 24(1): 36–47.

46.

Wright Whelan

Wagstaff

Wheatcroft

(2014) High-stakes lies: Verbal and nonverbal cues to deception in public appeals for help with missing or murdered relatives. Psychiatry Psychology and Law 21(4): 523–537.