Content uploaded by Stephanie Jat
Author content
All content in this area was uploaded by Stephanie Jat on Apr 23, 2024
Content may be subject to copyright.
The Internationa
l
Journal of
Speech,
Language
and the Law
ijsll (print) issn -
ijsll (online) issn -
Article
. –
© ,
https://doi.org/10.1558/ijsll.24331
Aliations
University of Cambridge, UK
email: scj45@cam.ac.uk kem37@cam.ac.uk aep58@cam.ac.uk
Received: 22 October 2022 Accepted after revision: 30 May 2023
Pausing and the ‘Othello Error’ :
Patternsofpausing in truthful and
deceptivespeech in the DyViS database
Stephanie C. Jat, Kirsty McDougall and Alice Paver
Abstract
e idea of detecting deception from speech is very attractive from a law enforce-
ment perspective, yet research considering the possibility has yielded conicting re-
sults, due to the practical diculties in investigating the topic. Scientic research
is yet to provide forensic linguistics with a reliable means of discerning lies from
truths. e present study explores the relationship between truthfulness and paus-
ing behaviour. Various aspects of the acoustics of pausing behaviour were inves-
tigated for Standard Southern British English in 30 mock police interviews from
the DyViS database (Nolan et al. 2009). A novel distinction was made between
prescribed and unprescribed lies, to delineate a potential source of dierences in the
unscripted content of speakers’ untruthful responses. Among pause duration meas-
ures, statistically signicant dierences were found across all three response types
(truth, prescribed lie, unprescribed lie) for response latency, between truths and
lies for initial lled pauses, and between unprescribed lies and the other response
types for silent pauses. For pause frequency measures, only internal lled pauses
showed a statistically signicant dierence: truths diered from both types of lies,
but prescribed lies did not dier from unprescribed lies. eories of cognitive eort
and attempted control are drawn on in accounting for these ndings.
: , , , ,
,
You told a lie, an odious damned lie.
William Shakespeare, Othello, Act V, Scene II, line 178
1. Introduction
Distinguishing falsity from truth has long occupied human interaction, be it
within families or in a court of law. Attempts have been made to develop objective
ways of detecting lies, e.g. the polygraph, which uses indicators such as physical
arousal to infer a high likelihood of deception, but such techniques have not
been demonstrated to achieve high levels of accuracy (American Psychological
Association 2004). erefore, despite the progress to date, ‘falsely accusing
innocents can be considered one of the major concerns about lie detection’
(Levine 2014: 740). Making a ‘disbelieving-the-truth mistake’, i.e. erroneously
judging that a speaker is lying despite being truthful, due to the misinterpretation
of ‘signs of nervousness’, has been coined by Ekman as the ‘Othello Error’. is
term follows from the title character in Shakespeare’s play who makes this mistake
and consequently murders his innocent wife (Ekman 1992; Mann 2018).
It has been suggested that focusing on cognitive aspects of lying when judging
speaker honesty may contribute to more accurate assessments (Levine 2014).
Lying has been recognised as requiring an increase in cognitive load since the
speaker has to fabricate and remember a believable lie, as well as self-monitor their
outward behaviour to appear truthful (Mann 2018). Researchers such as Beňuš,
Enos, Hirschberg and Shriberg (2006) argue that, since studies have demonstrated
that both silent and lled pauses are good indicators of speech planning and
cognitive eort, they may be useful for cueing deception. Interestingly, some
studies have found in fact that truthful answers tend to include more pausing than
deceitful ones (Arciuli, Mallard and Villar 2010; Beňuš et al. 2006). Comparison
of pausing behaviour in deceptive versus truthful speech warrants more detailed
investigation to establish whether, as Levine (2014) suggests, it may assist with
avoidance of the Othello Error.
1.1 Denition and classication of ‘lying’
e denition of lying is not clear cut. Most standard denitions refer to lying
as making an assertion with the deliberate intent to mislead others. A formal
notation of lying is given by Falkenberg (cited in Hardin 2018), in terms of
Speaker S, time t, and the statement p:
S lied at t, if and only if
(a) S asserted at t that p,
(b) S actively believed at t that not p.
‘ ’
e signicance of the speaker’s intention of misleading others to believe that
which is untrue rules out non-literal uses of language such as irony and teasing
(Hardin 2018). However, it would be more comprehensive to include other
means of misleading others, such as intonation and armative or negative head
movement, to avoid the oversimplication of assuming that lying only occurs
through outright lies (DePaulo, Kashy, Kirkendol, Wyer and Epstein 1996).
Hardin further species that lying uses spoken communication, distinct from
the more general notion of deception, which may be gestural or appearance-
based. Since the present investigation looks at only one type of deception, lying
via speech, the two terms will be used interchangeably here.
More than one theoretical model has been proposed to account for the
process of deception. McCornack’s (1992) Information Manipulation eory
(IMT) explains that deceit is the ‘transforming’ (Diana 2014) of information.
is is done against Grice’s Cooperative Principle (1975) and four maxims of
conversation (Quality, Quantity, Relevance and Manner), in such a way that
untruths are created either in its delivery or interpretation. In this view, deception
occurs via the covert violation of one or more of these maxims, meaning that false
assertions (violating Quality) are not the only way to lie, since the maxims may
account for the inference of implications (McCornack 1992). Buller and Burgoon
(1994), as well as other scholars, have postulated the Interpersonal Deception
eory (IDT), which takes a more goal-based approach regarding the speaker’s
purpose and strategies to attempt deception and its detection. Taking this a
step further, Anolli, Balconi and Ciceri (2002) have developed the Deceptive
Miscommunication eory (DeMiT), which places import on the intentionality
of the speaker. ese three frameworks are fundamental to the denition of
lying and the challenges in experimental methods presented below. Mann (2018)
emphasises that false speech such as acting, irony or teasing, primarily diers
from lying in the intention of the speaker. Vrij and Heaven (1999) explain the
distinction between objective and subjective cues for lying, and the importance
of separating these dierent types of lies. According to them, concealing
information should not be considered the same as fabricating information, with
the latter providing a larger cognitive challenge and resulting in more hesitations
and speech errors. Galasiński (2018) chooses to recognise three types of untruths:
falsication, distortion and taking words out of context, further nuancing the
linguistic denitions of lying, and emphasising the importance of experimental
methods not confounding between types of lies which might aect the results and
cause inaccuracies and unreliability.
Zuckerman, DePaulo and Rosenthal (1981) oer a multifactor approach to
modelling lying. Within this approach, deceptive behaviours may depend on four
factors: arousal, emotional reactions, cognitive eort and behaviour control. e
,
factor of arousal assumes that a liar, compared to a truthteller, will experience
heightened psychological eects from anxiety, which result in physical signs
of arousal such as faster speech, dgeting and nervousness. Lying is associated
with three key emotions: guilt, fear and ‘duping delight’ (Ekman and O’Sullivan
1989: 310), which may further inuence the speaker’s behaviour. Cognitive
eort refers to the additional mental load that telling a lie involves: the speaker
manufacturing a credible non-truth, ensuring the lie is consistent with previously
established information, remembering and maintaining this lie for the remainder
of the exchange, as well as paying attention to their demeanour and behaviour
to avoid coming across as insincere. is last aspect of cognitive burden is
realised as attempted behaviour control, a coping strategy under stress (Lazarus
and Folkman 1984). Liars may self-monitor their behaviour more closely in
order to appear more genuine. is may lead to an unnaturally smooth utterance
with little pausing, hesitation or other disuencies (Walczyk, Igou, Dixon and
Tcholakian 2013).
Other models such as Self-Presentation, Activation-Decision-Construction
and Working-Memory all point towards an increase in cognitive load for liars
compared to truthtellers (Walczyk et al., 2013). Motivation has also been suggested
as an important factor to consider when studying deception (Gustafson andOrne
1963; Burgoon and Buller 2015).
In general, researchers have agreed upon the multifactor model to explain
their results when analysing linguistic cues to deception. Most studies show that
interlocutors nd high-stakes and suitably motivated lies more convincing and
harder to discern from truths (Frank and Ekman 1997; Gustafson andOrne 1963;
Wright Whelan, Wagsta and Wheatcro 2013). Complex lies dier from simple
lies in terms of cognitive load, and the disparity in mental eort manifests as
dierent cues (Vrij and Heaven 1999).
e present study adopts the multifactor approach in line with Zuckerman et
al. (1981) described above, focusing on cognitive eort and attempted control
as key factors inuencing deceptive pausing behaviour. e current study also
introduces a novel distinction between prescribed and unprescribed responses
(see section 2.2), based on the notion of sanctioned and unsanctioned lies rst
introduced by Miller and Sti (1993). While sanctioned lies are those which
interviewees are instructed to tell, unsanctioned lies are usually elicited by
allowing the interviewee to choose whether to lie or tell the truth (Feeley and
DeTurck 1998). Unsanctioned lies may be argued to be more motivated than
sanctioned ones, as they are voluntarily told by the speakers, which in turn may
lead to more realistic evidence for cues to deception. More recent investigation
has found that sanctioned liars display more speech errors than truthtellers while
unsanctioned liars made fewer errors (Sporer and Schwandt 2006). However,
‘ ’
other vocal cues such as speech rate, hesitation frequency and response latency
showed no disparity based on varying accountability of lying in this work. e
three-way distinction of response types adopted in this study aims to probe the
relationship between varying cognitive eort and resulting pausing behaviour.
1.2 Speech cues in lie detection
Many linguists, psychologists and sociologists have investigated the human
behaviour of lying, mostly in relation to crime suspects being interrogated, a
situation for which the ability to dierentiate lies from the truth would be practical
and signicant, contributing to just outcomes. While the use of polygraph tests as
stereotypical ‘lie detectors’ has been widespread historically, in recent years there
has been growing acknowledgment that they are vulnerable to the Othello Error.
is is because polygraph tests make use of cues signalling stress as opposed to
deception, thus their accuracy has been overestimated (Hollien 1990; Faigman,
Fienberg and Stern, 2003; Fienberg 2005; Vrij 2008). Walczyk et al. (2013) argue
that the Guilty Knowledge Test (GKT), an alternative to the classic polygraph test,
may be more reliable. e GKT homes in on suspects’ reactions to information
regarding a crime using multiple choice questions: those with guilty knowledge
should show arousal upon hearing details of the crime, while those who are
innocent should have no reaction to the same details. However, similar to the
polygraph test, interviewee reactions may not be reliably attributed solely to their
guilt.
As outlined in Kirchhübel (2013), perception tests have been carried out to
test the legitimacy of deception cues such as eye contact, head movement, speech
hesitations and speech errors. Results show that some individuals may have an
above-average ability to detect lies, either due to intuition or occupational training
(such as intelligence services). However, most people’s accuracy in detecting lies
is a matter of chance.
In the domain of speech, the last decade has not seen much research into phonetic
cues to deception since Kirchhübel (2013, but see also Zhang, McGettigan and
Belyk 2022, described further below). Kirchhübel (2013) gives a comprehensive
overview of studies investigating phonetic cues to deception undertaken prior
to her own study. Linguistic studies oen distinguish between verbal and non-
verbal cues to deception. Use of verbal cues in deceptive utterances usually diers
between contexts and languages, and includes more impersonal vocabulary, fewer
past-tense verbs and more indirect messages (Hardin 2018; Ebesu and Miller
1994). Non-verbal cues are phonetic and can be split into non-temporal cues,
such as pitch, formant frequencies, amplitude and voice quality, and temporal
cues, as are the focus of the present study.
e present study examines temporal features such as frequency of pausing,
duration of hesitations and response latencies (i.e. time taken to respond),
,
which could be argued to be more robust than spectral ones since they are less
susceptible to alterations by recording equipment (Kirchhübel, 2013). Beňuš et
al. (2006) found that temporal features yielded signicant dierences between
deceptive and truthful speech whereas non-temporal features did not. e
literature on temporal cues to deception gives incongruent results, as summarised
thoroughly by Kirchhübel (2013: 116–121). For example, for lled pauses, Anolli
and Ciceri (1997), Vrij and Mann (2001) and Bello (2006), among others, found a
greater frequency in deceptive versus truthful speech; Buller and Burgoon (1994),
Granhag and Strömwall (2002), and others, instead found the reverse; while
studies such as Heilveil and Muehleman (1981), Bond, Omar, Mahmoud and
Bonser (1990) and Davis, Markus, Walters, Vorus and Connors (2005) found no
dierence. Parliament and Yarmey (2002), among other studies, found liars show
decreased response latency; Harrison et al. (1978), Farrow et al. (2003) and others
reported the opposite results (increased response latency); while no change in
response latency was found by Kraut and Poe (1980) nor Verschuere et al. (2004).
is lack of consistency may, however, be the result of varied interaction between
cognitive load and attempted control. Ebesu and Miller (1994) suggest that the
conicting analyses are due to the lack of dierentiation between types of deceit,
and that by accounting for more specic lying ‘types’, results may become more
consistent. Zhang et al. (2022) simulated a realistic environment for generating
deception via a novel social deduction board game paradigm. While acoustic
cues like pitch can be controlled by the speaker, temporal cues such as pausing
behaviour are oen unconscious and reliant on cognitive eort. eir study
concluded that lies tended to have more frequent and longer pauses, while vocal
pitch cues did not show conclusive ndings.
Despite some contending results, research has begun to uncover that, in terms
of temporal measures, pausing behaviour may be a useful indicator of deception,
as can be supported by the attempted control theory (Beňuš et al. 2006). Since
pausing is considered to be a subconscious but deliberate speech phenomenon, it
can be argued as more likely to be unaected by truthfulness (Clark and Fox Tree
2002; Fox Tree 2002; Roberts, Meltzer and Wilding 2009). Pausing, both silent
and lled, tends to be used to signal retention of the conversational oor while
planning speech (Clark and Fox Tree, 2002). Sti, Corman, Krizek and Snider
(1994) found that pausing behaviour during deception varies between speakers
and suggest a ‘baseline’ comparison instead of absolute isolation of ‘cues’. Both
cognitive load and attempted control seem to be the more prominent factors
aecting nonverbal cues to deception, whose roles warrant further investigation.
e present study attempts to recognise the disparity in cognitive eort involved
in deception dependent on the level of ‘scriptedness’ of the response, and adopts
‘ ’
a three-way distinction (prescribed lie, unprescribed lie, truth; see section 2.2)
between response types to investigate this.
1.3 Some challenges
Obtaining suitable data for analysis of deception is not straightforward. e
availability of real-life speech material containing high-stakes lies is very limited,
so studies have oen been carried out using simulated deception. Since lying is
based on speaker intent, manufactured settings enable a researcher to identify
truths from lies condently, whereas in real-life high-stakes situations liars do
not reveal their falsity voluntarily, making it dicult to be sure of the ground
truth (Mann 2018; Galasiński 2018). However, using simulated lying situations
is arguably ‘unrealistic’ and may lead to incorrect conclusions. Kirchhübel (2013:
28) describes a further theoretical issue that the ‘task of acting itself could be
seen as deceptive’, and so the participants are being ‘doubly deceptive’ by playing
the role required for the scenario: by telling the ‘truth’ they are lying, while ‘their
denial is true outside of the experiment’.
Furthermore, lie complexity may be subject to individual cognitive capacity,
as well as whether the lie had been prepared. Interviewer style has been shown
to cause disparity in interviewee response behaviour, as well as whether the
interviewer seems suspicious, and the type of question asked (Kirchhübel 2013;
Walczyk et al. 2013). Moreover, Sni et al. (1994) discovered that speakers tend
to adapt to their interlocutional situation over the course of a conversation, which
results in a decrease in response latencies over time. is means that the response
latency for truthful and untruthful responses at the beginning of an interview
may not be comparable with those at the end of the same exchange.
e challenges above highlight the need for strictly controlling variables as far
as possible, and for taking all of these factors into consideration when selecting
data and drawing conclusions.
1.4 Research question
e present study explores the relationship between lying and pausing behaviour
in a set of mock police interviews from the DyViS database (Nolan, McDougall,
de Jong and Hudson 2009). e elicitation procedure used for DyViS required
participants to lie at various stages in the interview. e same set of prompts
was used for each interview such that comparable data were elicited from
all participants. By comparing the pausing patterns produced in response to
dierent interview prompt types, this investigation aims to establish whether
there is a relationship between deceptiveness of responses (truth, prescribed
lie, unprescribed lie; see section 2.2) and the type, duration and frequency of
pausing. If such relationships are found, it may be that pausing has a role to play in
determining the likelihood of a sample of speech containing deceptive utterances.
,
2. Methodology
2.1 Speech data
e DyViS database provides recordings of 100 male speakers of Southern
Standard British English, aged 18–25, from the University of Cambridge,
undertaking four spoken tasks.
e current investigation analyses recordings from Task 1, a mock police
interview in which the participant assumes the role of suspect in a drug-tracking
crime. e participant views PowerPoint slides containing information about the
crime scenario such as names of people and information about them, street names
and venues, timing of events, etc. Most of the information is coloured black on the
slides: these are details which the suspect is free to talk about. Some information
is given in red: these details the suspect is required to lie about or conceal. Since
interviewer style may aect speaker behaviour unpredictably (Dunbar, Jensen,
Burgoon, Kelley, Harrison, Adame and Bernard 2015; Burgoon and Buller 2015),
in the present study, the interviews undertaken by only one of the two DyViS
interviewers (Gea de Jong) were used. irty interviews (DyViS Speakers 46–54,
56, 58–60, 62–65, 68, 69, 71, 73, 75–79, 84, 85, 87, 93) were analysed.
2.2 Prescribed and unprescribed lies
e present investigation draws a distinction between prescribed and unprescribed
lies (‘PL’ vs ‘UL’), based on the content of the participants’ responses. Although the
DyViS database does not give the option for the participants to tell sanctioned or
unsanctioned lies, there are prompts (e.g. Do you know Robert Freeman?) which
have ‘prescribed’ untruthful answers, given by the information on the PowerPoint
slides (Robert Freeman is the participant’s accomplice whom the participant is
required to deny knowing). Responding to these kinds of prompts should require
less cognitive eort since the speaker does not have to create a narrative. On the
other hand, there are also prompts for which the interviewees are required to
conceal information, but the answers are not provided in the slides, such as What
time did you get home? e answers to these types of prompts could be labelled
as ‘unprescribed’, requiring the interviewee to come up with their own believable
answers. Responses of this kind are denoted ‘UL’ in the present study.
e same prescribed versus unprescribed distinction applies to truths in the
DyViS interviews. Prompts such as Which lm did you catch the end of? do not
have a denite answer oered in the slides but are a follow-up prompt consistently
asked by the interviewer selected for this study. It would be empirically inadequate
to ignore such a dierence between the types of prescribed and unprescribed
answers given by the speakers, since this may aect cognitive load and, in turn,
the deceptive behaviour observed in the individual (Sti et al. 1994).
‘ ’
e present analysis, therefore, adopts a categorisation between truthful (T)
and false answers, and whether the false responses were prescribed (PL) or
not (UL). e T responses in the present study are all prescribed truths as the
elicitation technique did not prompt a sucient number of unprescribed truths
for analysis. e distinction between prescribed and unprescribed lies may mirror
the real-life circumstance of whether the suspect has prepared an alibi before
police interrogation, with prescribed lies corresponding to responses based on
their preprepared alibi, and unprescribed lies reecting responses to unforeseen
questions which have not been preconceived as part of their alibi. is provides
a potential transferability between this novel experimental variable and real-life
situations.
2.3 Prompt selection
e prompts and PowerPoint slides used by the interviewer to guide each
interview are provided in the documentation accompanying the DyViS database.
A subset of these prompts was selected as the prompts to the responses studied in
the present investigation. e study denoted ‘critical’ prompts as those which hold
incriminating information, to which the speakers must give a ‘dishonest’ answer.
e interrogation can be plausibly split in accordance with the information slides:
• [5]–[13] Basic information and ‘setting the scene’
• [14]–[17] Suspect’s relations
• [18]–[20] Suspect’s car
• [21]–[26] Suspect’s workplace
• [28]–[32] Neighbourhood where the crime took place
• [33]–[38] Suspect’s sister and scene of the crime
• [39]–[42] Suspect’s activities the day aer the crime
e prompts in the rst two sections were not included in the dataset, in order
to ensure that the participant had sucient time to immerse himself in the char-
acter and give more convincing responses. Moreover, the prompts selected were
chronologically very close to each other, regarding the same or similar topics, so
that the participants were in the same ‘mindset’ when giving the responses.
Table 1 shows the interview prompts selected sorted into three types: those
eliciting truths, prescribed lies and unprescribed lies. e right-hand column lists
the prompts according to type followed by the number of the corresponding
DyViS PowerPoint slide in square brackets.
e analysis of the 30 speakers’ responses to the prompts listed in Table 1
yielded the dataset described in Table 2.
e complete raw dataset is available in the data repository: https://doi.
org/10.17863/CAM.95142.
,
Prompts which are given an answer in the slides but were improvised upon
by a number of interviewees were not counted as unprescribed. e interviewer
asked a couple of follow-up prompts consistently across the interviews which
are classied as unprescribed truths. As mentioned earlier, due to the insucient
number of unprescribed truths, only prescribed truths were used in the present
comparison.
In developing the methodology for the present study, it is important to take
into account the possibility of speakers acclimatising to the task. Sti et al.
Table 1: List of prompts selected from the Task 1 interview of the DyViS database used in this study
Type of response
elicited Prompt [slide number used in the interview] (Nolan et al., 2009)
Truth (T) What sort of car do you drive? [18]
Imagine you’re travelling north on the bypass. What’s the big set of buildings
on your right? Do you ever go there? [28]
Finally, we’d like to ask about your sister. Could you tell us rst of all where
she lives and which road she lives on? [33]
And what does your sister do? [37]
Thank you – we understand that she works in a school some four miles away.
How does she get to work? [38]
Prescribed Lie (PL) Is it fully functioning at the moment? [18]
On the Thursday in question, did you give someone a lift after work? [19]
Your sister’s husband works in a local hotel. Could that be the Reef? [29]
(regarding public phone booth) Did you make a call here last Thursday? [31]
She always cycles? Does she have a car? [38]
Unprescribed Lie
(UL)
We think you might have a travelling companion. A camera shows you with
someone else in the car last Thursday evening. [19]
Don’t you know someone who works there? [29]
(regarding public phone booth) Have you ever used this phone? [31]
The big lake on the right – do you know it well? We think you might have
spent the evening there last Wednesday. What were you up to? [35]
Does anyone in the household keep a car? [38]
Table 2: Summary statistics describing the dataset
Min. total speaking
time (s)
Max. total speaking
time (s)
Mean total speaking
time per speaker (s) Std. dev. (s)
27.80 122.09 65.96 26.09
Min. time per turn (s) Max. time per turn (s) Mean time per turn (s) Std. dev. (s)
0.16 23.21 3.80 3.55
Mean number of T
replies per speaker
Mean number of PL
replies per speaker
Mean number of UL
replies per speaker
5.63 5.80 5.90
‘ ’
(1994) proposed a decay-impulse intervention model in analysing the ‘adaptive
pattern’ of response latency over time. ey acknowledge that speakers are
adept at acclimatising to their conversational circumstances and so, when faced
with consecutive ‘critical’ questions, the rst response will take the longest, the
next slightly less time, and so on in a decreasing conguration. is implies
the signicance of comparing speaker responses to questions asked either
consecutively or within the same line of conversation. It should be noted that the
decay-impulse model may dier between individuals based on their cognitive
abilities, which should be accounted for.
e selection of prompts attempted to balance open-endedness and directness
across the set, as less cognitive eort is required to formulate responses to close-
ended prompts, and speakers may feel more threatened by prompts which directly
target their honesty, aecting their motivation to lie (Kirchhübel 2013).
Since impromptu follow-up prompts (those not in the interviewer’s instructions)
were asked inconsistently across interviews and can be considered to be of a
dierent nature to primary prompts, only responses to the listed ‘instructed’
prompts were considered in the analysis to ensure that the prompt type was
consistent. For instance, in the following sequence of prompts about information
on a single slide, all requiring prescribed lies, only the response to the rst prompt
was included in the data analysed:
We were wondering whether you knew anyone called Robert?
e name ‘Freeman’ doesn’t ring any bells?
Weren’t you at school with a Robert Freeman?
Follow-up prompts requiring response types distinct from that of the main
prompt were included. (Appendix A provides an example of prompts eliciting all
three response types within one PowerPoint slide (slide [38]), unlike the example
above which all require the same response type). Responses to prompts which
were requests for clarication (e.g. I’m sorry?) were discounted, and the next
response was taken for measurement. When a speaker’s response required back-
channeling (marked as bc in the transcript), the turn was recognised as having
ended. Erroneous responses (i.e. truthful answers when a lie is prescribed) were
not included in the dataset.
2.4 Variables
e present study investigates ve pausing-based variables relating to both lled
pauses and periods of silence occurring during speakers’ interactions. e term
lled pause (FP) was used to describe a spoken section such as um or uh, not
including any surrounding silence. Here, FPs may serve as hesitations and/or
discourse markers during an interaction; all FPs were included in the present
,
study regardless of their possible discourse functions. A period of silence within
a speaker’s turn was dened as a silent pause (SP), while the period of silence
before a speaker responded to a prompt was dened as response latency (RL).
e ve variables under examination are: pause type, pause duration, frequency of
initial lled pauses, hesitancy rate (internal lled pauses) and hesitancy rate (silent
pauses). ese are dened and further specied as follows.
1. Pause type: e kind of pause used, either response latency (RL), silent
pause (SP), initial FP or internal FP, as explained in Table 3. Some examples
are illustrated with waveforms and spectrograms in Figure 1.
2. Pause duration: e duration of a silent or lled pause determined by
inspection of the waveform and spectrogram. For SP and RL, the pause
was measured from the end of the waveform of the last segment (including
any unvoiced or glottal segments), to the start of waveform of the next seg-
ment. For FP, the pause was measured from where the waveform showed
evidence of the beginning of the /ə/ vowel of the pause, to i) the end of the
waveform of the /m/ segment for a nasalised FP; or ii) the start of the wave-
form for the rst segment of the next word. Mean values of pause durations
were calculated to examine the possible disparity between response catego-
ries (T/PL/UL).
Figure 1: Example of acoustic data used in analysis in Praat transcribed orthographically with pauses (RL, initial FP,
internal FP, SP) annotated (Speaker 52). ‘hes’ stands for ‘hesitation’ as used in the transcription provided with the
DyViS database.
‘ ’
Table 3: Denition and exemplication of variables investigated
Variable Denition Example(s)
Response latency (RL) The period of silence between the end of the interview-
er’s prompt and the beginning of the response.
A response which overlapped with the end of the
interviewer’s prompt, i.e. an anticipatory response, was
not included as a RL datapoint. This was due to the very
small number of anticipatory responses (two instances).
Interviewer: You say that you drive to work. What sort of car do you drive?
[RL: duration 0.88s]
Interviewee (Speaker 52): um it’s a v w beetle it’s um new type
Initial Filled Pause
(initial FP)
A turn-initial lled pause, excluding any following
silence. For lled pauses, the presence or absence of a
nasal was also noted, expressed as +/-N (see Section 2.5
for methodological treatment).
Interviewer: You say that you drive to work. What sort of car do you drive?
Interviewee (Speaker 52): um [initial FP: duration 0.30s; +N] it’s a v w
beetle it’s um new type
Interviewer: You say that you drive to work. What sort of car do you drive?
Interviewee (Speaker 48): uh [initial FP: duration 0.33s; -N] i’ve got a v w
beetle
Internal Filled Pause
(internal FP)
A turn-internal lled pause, excluding any surrounding
silence.
Interviewer: You say that you drive to work. What sort of car do you drive?
Interviewee (Speaker 52): um it’s a v w beetle it’s um [internal FP: duration
0.28s; +N] new type
Interviewer: You say that you drive to work. What sort of car do you drive?
Interviewee (Speaker 51): um i just drive a volkswagen beetle one of the new
type it’s uh [internal FP: duration 0.23; -N] sky-blue
Silent Pause (SP) A pause either due to a grammatical or intonational
boundary or speech planning and hesitation, with no
vocal presence. SPs are always turn-internal. Intentional,
loud exhales (BLOW), as well as any other respiratory
noises, are counted as silent pauses.
Interviewer: You say that you drive to work. What sort of car do you drive?
Interviewee (Speaker 52): um it’s a v w beetle it’s [SP: duration 0.17s] um
new type
Example of a (BLOW):
Interviewee: i may hes i i (BLOW) [SP: duration 0.30s] well hes
Note that the interviewee’s responses are given in the transcription notation provided by the DyViS database.
,
3. Frequency of initial lled pauses: is was calculated using:
FP initial Frequency =
number of times the speaker
begins a turn with a FP
number of turns taken
4a. Hesitancy rate (internal lled pauses): is was calculated in occurrences/
second using:
internal FP HR =
number of times the speaker uses a turn internal FP
total duration of target speaker’s speech turns per
response category (in seconds)
4b. Hesitancy rate (silent pauses): is was calculated in occurrences/second
using:
SP HR =
number of times the speaker uses a turn internal SP
total duration of target speaker’s speech turns per
response category (in seconds)
e total duration of each target speaker’s speech turns per response category was
measured as the sum of speaking time (excluding RL) that the interviewee uses
to answer all the prompts associated with each type of answer (T/PL/UL). is is
measured in seconds.
Appendix A provides excerpts of transcripts from two dierent interviews
demonstrating the exchange resulting from the same prompts, to illustrate the
relatively unscripted nature of the responses. e variables under investigation
have been highlighted in the transcripts. It should be noted that the transcripts
of the speakers’ utterances are those which are provided with the original DyViS
database, and that the interviewer’s utterances were transcribed for the present
study, using the same approach.
2.5 Data collation
e data were collected manually using recordings imported into Praat (Boersma
and Weenink 2020). An Excel spreadsheet was used to collate the data and
organise results according to response category (T, PL, UL).
Both initial and internal FP were initially subdivided into +N and -N categories
as noted in section 2.4, but a preliminary inspection of the data showed little
obvious eect of deception on the presence of a nasal. Use of er versus erm has
been observed to exhibit notable speaker-specicity in SSBE (McDougall and
Duckworth 2018), and one might speculate that this speaker-specicity has a
greater inuence than any eect of deception; thus, further evaluation of +/- N
patterning was not pursued here.
‘ ’
2.6 Hypotheses
is study investigates whether dierent types of pausing display phonetic
dierences when comparing truths to prescribed and unprescribed lies, and
whether these dierences are signicant. If response type proves signicant,
although the exact cognitive processes cannot be tested here, it is inferred that
the signicant distinctions will be due to cognitive eort and attempted control.
Based on the ndings in previous studies, the following hypotheses are made:
1. Speakers will have fewer silent and lled pauses when answering with lies
than truths (owing to attempted control), and if the answer is prescribed
and not unprescribed (less cognitive eort), resulting in the pattern
PL<UL<T for pause frequency.
2. Speakers will have longer RL for truths than prescribed lies and for unpre-
scribed lies than truths (since more speech planning will be involved),
resulting in PL<T<UL for RL duration.
3. Speakers will have longer silent and lled pauses when deceptive responses
are not prescribed (UL) and shorter ones for prescribed answers (where
they control their responses so as not to appear deceitful). is will lead to
a PL<T<UL order of pause duration.
2.7 Statistical analysis
Statistical analysis was conducted in R using linear mixed eects models. Linear
mixed eects models were calculated for each of (1) the duration of the four
pause types (RL, initial FP, internal FP, SP), (2) the frequency of occurrence
of an initial FP, (3) the hesitancy rates for the occurrences of internal FP and
SP, in R Studio (R Core Team 2020) using the lme4 package (Bates, Mächler,
Bolker and Walker 2015). Pause duration and pause frequency were both treated
as continuous independent variables, and response type (T/PL/UL) was the
predictor for both. Speaker number (for (1)–(3)) and prompt number (for (1))
were included as random intercepts. P-values were estimated via t-values using
lmerTest (Kuznetsova, Brockho and Christensen 2017), which approximated
these values using the Satterthwaite method. A model of the same structure was
calculated for each pause type aer reordering the levels for response type to
allow for comparison between all three categories of T/PL/UL.
3. Results
3.1 Pause duration
3.1.1 Response latency (RL)
Boxplots showing the duration of the response latency for T, PL and UL response
types are given in Figure 2. is gure shows that unprescribed and prescribed lies
,
appear to have longer RL than truths, with unprescribed lies in turn signicantly
longer in duration than prescribed lies ([T] mean=0.743s, std dev=0.477s;
[PL] mean=0.822s, std dev=0.550s; [UL] mean=1.03s, std dev=0.540s). is
observation was supported by the statistical analysis: a multiple regression model
found a statistically signicant dierence in RL for prescribed lie compared to
truth (β=0.12, t(506)=2.31, p=0.021), unprescribed lie compared to truth (β=0.31,
t(506)=5.96, p<0.001) and unprescribed lie compared to prescribed lie (β=0.19,
t(506)=3.67, p=0.0003).
3.1.2 Initial lled pauses (initial FP)
Boxplots showing the duration of the initial lled pauses for T, PL and UL
response types are given in Figure 3. is gure shows that unprescribed
and prescribed lies have longer initial FP than truths as shown in the gure
([T] mean=0.360s, std dev=0.557s; [PL] mean=0.481s, std dev=0.234s; [UL]
mean=0.519s, std dev=0.189s). is was conrmed by the statistical analysis: a
multiple regression model found a statistically signicant dierence in initial FP
duration for prescribed lies compared to truth (β=0.14, t(221)=4.35, p<0.001) and
unprescribed lie compared to truth (β=0.17, t(221)=6.02, p<0.001). No signicant
dierences were found in the duration of initial FP comparing prescribed and
unprescribed lies (β=0.03, t(221)=0.93, p=0.354).
Figure 2: Boxplots of the duration of response latency for the three response types in seconds (‘×’ represents the
mean for each response type).
‘ ’
3.1.3 Internal lled pauses (internal FP)
Boxplots showing the duration of the internal lled pauses for T, PL and UL
response types are given in Figure 4. is gure shows that no statistically
signicant dierences were found in the duration of internal FP among the
three response types ([T] mean=0.341s, std dev=0.157s; [PL] mean=0.326s, std
Figure 3: Boxplots of the duration of initial lled pauses for the three response types in seconds (‘×’ represents the
mean for each response type).
Figure 4. Average duration of internal lled pauses across response types in seconds (‘×’ represents the mean for
each response type).
,
dev=0.138s; [UL] mean=0.324s, std dev=0.160s) (T-intercept: [PL] β=-0.03,
t(271)=-1.00, p=0.318; [UL] β=-0.02, t(271)=-0.80, p=0.428; PL-intercept: [UL]
β=0.00803, t(271)=0.33, p=0.745).
3.1.4 Silent pauses (SP)
Boxplots showing the duration of the silent pauses for T, PL and UL response
types are given in Figure 5. is gure shows that unprescribed lies yielded longer
SP than both truths and prescribed lies ([T] mean=0.444s, std dev=0.468s; [PL]
mean=0.488s, std dev=0.498s; [UL] mean=0.563s, std dev=0.567s). is was
shown in the statistical analysis: a multiple regression model found a statistically
signicant dierence in SP duration for unprescribed lies compared to truths
(β=0.12, t(911)=2.95, p=0.004) and unprescribed lies compared to prescribed
lies (β=0.09, t(911)=2.14, p=0.033). No signicant dierences were found in SP
duration for truths versus prescribed lies (β=0.04, t(911)=0.81, p=0.420).
3.2 Pause frequency
e mean frequency of initial lled pauses across the 30 speakers for T, PL and
UL response types is given in Table 4. No statistically signicant dierences
were observed in the frequency of initial FP among the three response types
(T-intercept: [PL] β=0.00889, t(85)=0.22, p=0.825; [UL] β=-0.00556, t(85)=-0.14,
p=0.889; PL-intercept: [UL] β=-0.01, t(85)=-0.36, p=0.719).
Figure 5: Boxplots of the duration of silent pauses for the three response types in seconds (‘×’ represents the mean
for each response type).
‘ ’
3.3 Hesitancy rate
e mean hesitancy rate of internal lled pauses for T, PL and UL response types
is given in Table 5. Truths produced the highest hesitancy rate of internal FP. is
is conrmed by a multiple regression model which found a statistically signicant
dierence in hesitancy rate for prescribed lies compared to truths (β=-0.09,
t(85)=-4.86, p<0.001) and unprescribed lies compared to truths (β=-0.08, t(85)=-
4.00, p<0.001). No signicant eects were found in the hesitancy rate of internal
FP comparing prescribed and unprescribed lies (β=0.02, t(85)=0.86, p=0.395).
e mean hesitancy rate of silent pauses for T, PL and UL response types is
given in Table 6. No signicant dierences were found for the hesitancy rate of
SP comparing response types, (T-intercept: [PL] β=-0.0006, t(85)=-0.02, p=0.987;
[UL] β=0.02, t(85)=0.58, p=0.561; PL-intercept: [UL] β=0.02, t(85)=-0.60,
p=0.550).
4. Discussion
e current study explored the potential for pausing behaviour to indicate
whether a response is a truth, prescribed lie or unprescribed lie, and whether
comparing the duration and frequency of pauses in interviewee responses might
be an eective avenue for avoiding the Othello Error. A number of statistically
signicant dierences were found among the features of pausing measured for T,
PL and UL responses (see Table 7 for a summary of the main statistical ndings),
conrming some aspects of the hypotheses formulated in section 2.7, although
the overall picture was more complex than predicted.
Table 4: Mean frequency of initial lled pauses for the three response types across the 30 speakers, measured in
number of times per number of turns
T PL UL
mean (s) 0.183 0.192 0.178
std dev (s) 0.232 0.173 0.219
Table 5: Mean hesitancy rate of internal lled pauses across response types, measured in occurrences per second
T PL UL
mean (s) 0.177 0.085 0.101
std dev (s) 0.119 0.078 0.086
Table 6: Mean hesitancy rate of silent pauses for the three response types, measured in occurrences per second
T PL UL
mean (s) 0.430 0.429 0.451
std dev (s) 0.181 0.209 0.120
,
Hypothesis 1 predicted that speakers would use pauses more frequently when
telling the truth than lying and that pauses would occur more frequently for
unprescribed lies than prescribed ones, i.e. PL<UL<T, for both lled and silent
pauses. Although neither initial FP frequency nor SP rate were signicantly
aected by response type, truths showed a higher occurrence of internal FP per
second of speech than either unprescribed or prescribed lies. Kirchhübel, Stedmon
and Howard (2013) explain a greater prevalence of pauses in truthful speech using
the attempted control theory (Lazarus andFolkman 1984; see also Vrij and Heaven
1999, and Arciuli et al. 2010), such that speakers subconsciously produce more
lled pauses within a response when telling a truth. Comparing lies to a baseline
‘truth’, the disparity between the present results for frequency and duration was
not predicted by the hypotheses. is shows some similarity to ndings in existing
research. For example, Davis et al. (2005) found a negative correlation between
deception and hesitancy rate (corresponding to FP in this study), while pauses (i.e.
SP) showed no signicant eects. Beňuš et al. (2006) also found higher frequency
of FP for truths. Kirchhübel’s (2013: 116–120) table compiling the results of many
previous studies split pauses into hesitations (FP) and pauses (SP) as well and
noted a number of other studies with signicant ndings of lower frequency of
hesitations (FP) when speakers are lying. Vrij and Heaven (1999) draw on cognitive
theory once more to make a connection between a higher frequency of FPs when
lies were hard, but a decrease if the lies were easy. Kirchhübel et al. (2013) reported
participants using fewer FPs in each of two increasingly provocative interviews in
which they were required to lie. e fact that the current study shows signicant
results only for an increased internal FP rate in truths could provide evidence that
the speakers made use of attempted control more overall, mostly overriding the
varying levels of cognitive eort involved in dierent categories of responses, even
in a manufactured deceptive scenario.
Table 7: Summary of statistical signicances yielded by the data
Duration Frequency/Hesitancy Rate
RL Signicant dierences between all
three response types: T<PL<UL
N/A
Initial FP Signicant dierence between truths
and lies, but not between types of lies:
T<PL, T<UL
No signicant dierences
Internal FP No signicant dierences Signicant dierence between truths
and lies, but not between types of lies:
UL<T, PL<T
SP Signicant dierence between UL and
each of PL and T only: T<UL, PL<UL
No signicant dierences
‘ ’
Hypothesis 2 predicted that RL would be longer for truths than prescribed
lies, and longer again for unprescribed lies, i.e. PL<T<UL. e results showed
signicant dierences between all three response types for RL but ordered
T<PL<UL. e current study provides evidence for longer RL in UL responses,
which consist mostly of fabricating information on the interviewee’s part.
Its results support the idea that dierent types of lies may aect the cognitive
eort required in producing them. For example, concealment-type lies may be
answers to yes/no prompts, while fabricated lies might be more open-ended,
answering prompts which provide the speaker with opportunities to invent an
alternate narrative. Walzcyk et al. (2003) used a false-information paradigm to
extract deceptive responses according to open-ended versus Yes/No questions,
and reported that open-ended responses had longer RL, consistent with the
present study’s ndings. It should be noted that the distinction between Yes/No
and open-ended responses falls in line with Vrij and Heaven’s (1999) theoretical
discussion earlier in section 1.2. e higher cognitive load needed to recall a
lie and to control one’s appearance in order to come across as calm and honest
may also be behind the longer RL for PL than T found here. e results replicate
some of those from Beňuš et al. (2006), who found longer response latencies for
lies, and a higher frequency of pauses for truths. is is also a result shown by
Rockwell et al. 1997 (cited in Kirchhübel 2013). More recently, Zhang et al. (2022)
found consistent patterns as well and attributed the higher frequency and longer
duration of pauses in lies to the eects of cognitive eort. From a psycholinguistic
angle, Loy, Rohde and Corley (2018) discuss the importance of keeping in mind
the interactive nature of deceit, as it is in fact signs of cognitive load which
listeners take to be cues to deception.
Hypothesis 3 predicted that durations of both silent and lled pauses would
assume a PL<T<UL ordering. While there were no signicant dierences between
the three response types in the duration of internal FPs, initial FPs yielded shorter
durations for truths compared with both lie types. Silent pauses in truthful
responses and in prescribed lies were shorter than those in unprescribed lies but
showed no signicant dierence in duration between truths and prescribed lies.
Unprescribed lies furnished longer initial FPs than the other two response types,
which may be due to cognitive eort involved in formulating a lie as opposed to
telling the truth, because of the need to suppress or alter one’s knowledge and to
remember the new information, as opposed to the more straightforward recall of
truthful answers (Sti et al. 1994; Krakovsky 2009; Walczyk et al. 2013; Sporer and
Schwandt 2006; although see Adams-Quackenbush 2016 for counterevidence).
e statistical signicance of the present results supports the cognitive approach
to deception, suggesting the inuence of lying and its cognitive demands may be
manifested in increased pausing duration.
,
RL and initial FP were arguably the most eective of the variables examined
at highlighting dierences between truths and lies, with these pause types in
both categories of lies having longer duration than those in truths in the present
dataset. is is in contrast to SP, which only showed signicant dierences for
unprescribed lies. ese ndings were not predicted by the hypotheses.
e boxplots in section 3 show that the distribution of values for each
response type (T/PL/UL) overlap to large extents and involve small numbers of
milliseconds’ dierence. It is therefore necessary to emphasise that the signicant
results by no means prescribe absolute benchmarks for deception judgements.
Although statistically signicant eects were found, it would be presumptuous
to claim that the ndings from this experiment can conclusively determine the
truthfulness of an utterance from a speaker in any real-life criminal investigations
or interviews.
Arciuli et al. (2010) make an apt observation that the ‘normal’ conditions (here,
Truths) used as a baseline for comparison with the deceptive utterances are only
‘interview normal’, and may not necessarily represent a conversational, or real-
life norm for any given speaker. is is similar to the Observer’s Paradox, the
diculty of capturing natural data in a real-life setting when subjects are being
observed (Labov 1972: 209).
Further, SP and FP can be expected to show considerable intraspeaker variability
(McDougall and Duckworth 2018), and this might be clouding deception-related
patterns here. Moreover, silent pauses may have a grammatical purpose or be
used as a tool for speech planning and repair regardless of truthfulness (Levelt
1989).
Since each of the four pause types showed at least one aspect of statistical
signicance, future research should explore whether an approach which uses
the pause types in combination produces greater levels of discrimination among
deceptive and truthful utterances (e.g. an isolated FP, or one followed by a SP).
is is important because the present results show large overlaps between the
ranges (in seconds) of the durations of pauses for each response type, and a
similar picture of overlap for the frequency results. Indeed, where signicant
results have been found, the dierence between lies and truths may be smaller
than the range of overlap between response types. A larger dataset would be
needed since the diverse possible combinations of pausing behaviour (e.g. FP+SP;
SP+FP; SP+FP+SP; etc.) will lead to a small number of each combination for the
chosen prompts.
e methods in the present study controlled many potentially relevant
variables, with interviewer, prompt selection, and speaker sex, age and accent
background all held uniform. e elicitation methods used in Task 1 of the DyViS
database allowed for a tangible way of measuring dierences between truths and
‘ ’
lies, as well as distinguishing lies which were prescribed by the information given
to participants of the experiment. It is important to bear in mind that real-life
situations do not oer such unambiguous responses, nor promise such clarity of the
degree of deception in response types. Moreover, Bortfeld, Leon, Bloom, Schober
and Brennan (2001) and Galasiński (2018) discuss the notion of ‘interpersonal
coordination’, pointing out that discourse is a socially constitutive and ideological
structure, and is dependent on many contextual factors, dened by the speaker’s
individual and cultural beliefs. ey also reference Bok’s (1999) ‘truth bias’, the
expectation that one’s interlocutor would tell the truth. Personality traits (e.g.
Machiavellianism; Knapp, Hart and Dennis 1974) and cultural speaking norms
regarding hesitation use further aect how speakers may use pausing behaviour in
truthful responses (Bond et al. 1990), but these factors are less straightforwardly
controlled for in real-life situations. Real-life deception is more nebulous, and
sometimes is a matter of degrees in actual discourse, and conversation may allow
for more than one level of deception. e cognitive and emotional eort involved
in lying is also dependent on the intellect and personality of the speaker, which
aect their linguistic response to cognitively complex activities (Vrij and Heaven
1999).
5. Conclusion
is study explored the signicance of pausing behaviour in deceptive speech
when compared to truthful responses by analysing 30 simulated police interviews
from Task 1 of the DyViS database. e results showed statistically signicant
eects of lying on both durational and frequency-of-occurrence dimensions of
pausing behaviour. Although the measures considered do not oer completely
conclusive, absolute landmarks when deciding between liars and truthtellers,
they provide evidence in support of potential dierences in pausing behaviour
according to utterance truthfulness. Of course, the ‘truth’ in this study is an
experimental truth as opposed to the objective truth, and some of the detected
cognitive load may be due to the participants reading and interpreting slides
and instructions. e present study paves the way for further research into
pausing behaviour as a function of utterance truthfulness, possibly by combining
dimensions of pauses such as duration and frequency or two or more pause types,
which may illuminate more distinctive eects of deception. It is important to
emphasise that any apparent indicator of deception must be treated with caution,
given the large overlap between temporal measures relating to truths and lies.
Further, other potential sources of inuence such as speaker-specicity and
pragmatic factors must not be overlooked. Considering too narrow a range of
factors when judging an utterance’s truthfulness will increase one’s susceptibility
to the Othello Error.
,
e categorisation of lies into prescribed and unprescribed was an original
feature in the analysis of the data; this can be connected to the real-life dierence
between responses which are part of the suspect’s alibi and those which are not.
is begs for further research into ways in which an observer might be able to
distinguish between prescribed and unprescribed lies in real life situations. e
possibility of unprescribed lies bearing more elaborate responses may aect the
overall behaviour of pausing, as well as mixing certain half-truths into such
responses. Further in-depth study of the semantic and cognitive boundaries
between the two types of lies is also needed.
Of course, many of the ndings in this study showed numerical variables that
displayed a large area of overlap for both duration and frequency of pausing types
according to response types, and another direction for future research might
be to combine the measures in various ways to nd a set of temporal variables
which would better be able to delineate response truthfulness accurately and
conclusively, paralleling work in the eld of speaker comparison (e.g. ‘disuency
proles’, McDougall and Duckworth 2018). is could take steps closer to
developing a full application of pausing behaviour for dierentiating between
truths and lies, for which the current study has found preliminary signicant
results.
Acknowledgments
Many thanks to Marju Kaps for her help in the University of Cambridge Phonetics
Laboratory and with soware. anks are also due to three anonymous reviewers
for their feedback.
Data accessibility statement
e data and statistical summary that support the ndings of this study are openly
available at https://doi.org/10.17863/CAM.95142.
About the authors
Stephanie Jat is a current PhD candidate at the University of Cambridge. Her
research interests include the syntax-phonology interface, and forensic phonetics,
particularly the phonetic behaviours of deception.
Kirsty McDougall is an Assistant Professor of Phonetics at the University of
Cambridge, and a Fellow of Selwyn College, Cambridge. Her research interests
range across speaker characteristics, theories of speech production, phonetic real-
isation of varieties of English, and forensic phonetics. Among other things, her
forensic phonetic research has focussed on speaker-distinguishing properties of
dynamic features of speech, perceived voice similarity and its implications for the
selection of foils for voice parades, and the development of techniques for ana-
lysing individual dierences in disuency behaviour. She is a member of IAFPA.
‘ ’
Alice Paver is a research assistant working on the ‘Improving Voice Identi-
cation Procedures’ (IVIP) project in the Phonetics Laboratory at the University
of Cambridge. She previously completed her MSc in Forensic Speech Science at
the University of York and her MA in English Language and Literature at the
University of Edinburgh. Her research interests include forensic phonetics, socio-
phonetics, speaker similarity and accent judgements. She is a member of IAFPA.
References
Adams-Quackenbush, N. (2015) e Eects of Cognitive Load and Lying Type on
Deception Cues. esis for Master of Science in Applied Science (Psychology and Law).
Saint Mary’s University, Halifax, Nova Scotia.
American Psychological Association. (2004) e truth about lie detectors (aka polygraph
tests). Retrieved 10 August 2020, from https://www.apa.org/research/action/polygraph
Anolli, L., Balconi, M. and Ciceri, R. (2002) Deceptive miscommunication theory
(DeMiT): a new model for the analysis of deceptive communication. In L. Anolli, R.
Ciceri and G. Riva (eds) Studies in New Technologies and Practices in Communication.
Say Not to Say: New Perspectives on Miscommunication 73–100. Amsterdam: IOS Press.
Arciuli, J., Mallard, D. and Villar, G. (2010) ‘Um, I can tell you’re lying’: linguistic markers
of deception versus truth-telling in speech. Applied Psycholinguistics 31: 397–411.
https://doi.org/10.1017/S0142716410000044.
Bates, D., Mächler, M., Bolker, B. and Walker, S. (2015) Fitting linear mixed-eects
models using lme4.Journal of Statistical Soware67(1): 1–48. https://doi.org/10.18637/
jss.v067.i01.
Beňuš, S., Enos, F., Hirschberg, J. and Shriberg, E. (2006) Pauses in deceptive speech.
Proceedings ISCA 3rd International Conference on Speech Prosody, Dresden,
Ge r man y.
Boersma, P. and Weenink, D. (2020) Praat: doing phonetics by computer [Computer
program]. Version 6.1.38. http://www.praat.org/
Bok, S. (1999) Lying: Moral Choice in Public and Private Life. New York: Vintage Books.
Bond, C. F., Omar, A., Mahmoud, A.and Bonser, R. N.(1990) Lie detection across
cultures.Journal of Nonverbal Behavior14:189–204. https://doi.org/10.1007/
bf00996226
Bortfeld, H., Leon, S. D., Bloom, J. E., Schober, M. F. and Brennan, S. E. (2001)
Disuency rates in conversation: eects of age, relationship, topic, role, and gender.
Language and Speech 44(2): 123–147. https://doi.org/10.1177/00238309010440020101
Buller, D. B. and Burgoon, J. K. (1994) Communication eory. New York: McGraw-Hill.
Burgoon, J. K. and Buller, D. B. (2015) Interpersonal deception theory. In C. R. Berger,
M. E. Rolo, S. R. Wilson, J. P. Dillard, J. Caughlin and D. Solomon (eds) e
International Encyclopedia of Interpersonal Communication 1–6. Hoboken, NJ: John
Wiley & Sons. https://doi.org/10.1002/9781118540190.wbeic170
Clark, H. H. and Fox Tree, J. E. (2002) Using ah and um in spontaneous speaking.
Cognition 84: 73–111.
,
Davis, M., Markus, K. A., Walters, S. B., Vorus, N. and Connors, B. (2005) Behavioral
cues to deception vs. topic incriminating potential in criminal confessions. Law and
Human Behavior 29: 683–704. https://doi.org/10.1007/s10979-005-7370-z
DePaulo, B. M., Kashy, D. A., Kirkendol, S. E., Wyer, M. M. and Epstein, J. A. (1996)
Lying in everyday life. Journal of Personality and Social Psychology 70: 979–995. https://
doi.org/10.1037/0022-3514.70.5.979
Diana, B. (2014) A cognitive approach to deception detection: multimodal recognition of
prepared lies. Unpublished PhD thesis, University of Milano Bicocca.
Dunbar, N. E., Jensen, M. L., Burgoon, J. K., Kelley, K. M., Harrison, K. J., Adame, B. J.
and Bernard, D. R. (2015) Eects of veracity, modality, and sanctioning on credibility
assessment during mediated and unmediated interviews. Communication Research
42(5): 649–674. https://doi.org/10.1177/0093650213480175
Ebesu, A. S. and Miller, M. D. (1994) Verbal and nonverbal behaviors as a function of
deception type.Journal of Language and Social Psychology13(4): 418–442. https://doi.
org/10.1177/0261927X94134004
Ekman, P. (1992)Telling Lies: Clues to the Deceit in the Marketplace, Politics, and
Marriage. New York: W. W. Norton & Co.
Ekman, P. and O’Sullivan, M. (1989) Hazards in Detecting Deceit. In D. C. Raskin (ed.)
Psychological Methods in Criminal Investigation and Evidence 297–332. New York:
Springer Publishing.
Faigman, D. L., Fienberg, S. E. and Stern, P. C. (2003)e limits of the polygraph. Issues
in Science and Technology20(1): 40–46. Available at: https://repository.uchastings.edu/
faculty_scholarship/134
Feeley, T. H. and DeTurck, M. A. (1998) e behavioural correlates of sanctioned and
unsanctioned deceptive communication. Journal of Nonverbal Behavior 22(3): 189–
204. https://doi.org/10.1023/a:1022966505471
Fienberg, S. (2005) To tell the truth: on the probative value of polygraph search
evidence.Jurimetrics46(1): 107–116. Retrieved 20 August 2020, from www.jstor.org/
stable/29762921
Fox Tree, J. E. (2002) Interpreting pauses and ums at turn exchanges. Discourse Processes
34(1): 37–55. https://doi.org/10.1207/s15326950dp3401_2
Frank, M. G. and Ekman, P. (1997) e ability to detect deceit generalizes across dierent
types of high-stake lies. Journal of Personality and Social Psychology 72(6): 1429–1439.
https://doi.org/10.1037/0022-3514.72.6.1429
Galasiński, D.(2018)Lying and discourse analysis. In J. Meibauer (ed.)e Oxford
Handbook of Lying517–528. Oxford: Oxford University Press.
Grice, H. P. (1975) Logic and conversation. In P. Cole, and J. Morgan (eds)Studies in
Syntax and Semantics III: Speech Acts 183–198. New York: Academic Press.
Gustafson,L. A.andOrne,M. T.(1963)Eects of heightened motivation on the
detection of deception.Journal of Personality and Social Psychology 47:408–411.
https://doi.org/10.1037/h0041899
Hardin, K.(2018)Linguistic approaches to lying and deception.In J. Meibauer (ed.)e
Oxford Handbook of Lying56–70. Oxford: Oxford University Press.
‘ ’
Hollien, H. (1990) e Acoustics of Crime: e New Science of Forensic Phonetics (Applied
psycholinguistics and communications disorders). New York: Plenum Press.
Kirchhübel, C. (2013) e acoustic and temporal characteristics of deceptive speech.
Unpublished PhD thesis, University of York.
Kirchhübel, C., Stedmon A. W. and Howard D. M. (2013) Analyzing deceptive speech.
In D. Harris (ed.) Engineering Psychology and Cognitive Ergonomics: Understanding
Human Cognition 134–141. EPCE 2013. Lecture Notes in Computer Science 8019.
Berlin, Heidelberg: Springer. https://doi.org/10.1007/978-3-642-39360-0_15
Knapp, M. L., Hart, R. P. and Dennis, H. S. (1974) An exploration of deception as a
communication construct. Human Communication Research 1: 15–29. https://doi.
org/10.1111/j.1468-2958.1974.tb00250.x
Krakovsky, M. (2009) e load of lying: testing for truth. Scientic American Mind 20(4):
10. https://doi.org/10.1038/scienticamericanmind0709-10a
Kuznetsova, A., Brockho, P. B. and Christensen, R. H. B. (2017) lmerTest package: tests
in linear mixed eects models.Journal of Statistical Soware82(13): 1–26. https://doi.
org/10.18637/jss.v082.i13
Labov, W. (1972)Sociolinguistic Patterns.Philadelphia: University of Pennsylvania Press.
Lazarus,R. S.andFolkman,S.(1984)Stress, Appraisal and Coping.New York:Springer.
Levine, T. (2014)Encyclopedia of Deception. Los Angeles: Sage.
Loy, J. E., Rohde, H. and Corley, M. (2018) Cues to lying may be deceptive: speaker and
listener behaviour in an interactive game of deception.Journal of Cognition1(1): 42.
http://doi.org/10.5334/joc.46
Mann, S.(2018)Lying and lie detection.In J. Meibauer (ed.)e Oxford Handbook of
Lying408–419. Oxford: Oxford University Press.
McCornack, S. (1992) Information manipulation theory. Communication Monographs 59:
1–16. https://doi.org/10.1080/03637759209376245.
McDougall, K. and Duckworth, M. (2018) Individual patterns of disuency across
speaking styles: a forensic phonetic investigation of Standard Southern British English.
International Journal of Speech Language and the Law 25(2): 205–230. https://doi.
org/10.1558/ijsll.37241
Miller, G. R. and Sti, J. B. (1993) Deceptive Communication. Newbury Park, CA: Sage.
Nolan, F., McDougall, K., de Jong, G. and Hudson, T. (2009) e DyViS database: style-
controlled recordings of 100 homogeneous speakers for forensic phonetic research.
International Journal of Speech Language and the Law 16(1): 31–57. https://doi.
org/10.1558/ijsll.v16i1.31
R Core Team. (2020) R: A Language and Environment for Statistical Computing. Vienna:
R Foundation for Statistical Computing. https://www.R-project.org/
Roberts, P. M., Meltzer, A. and Wilding, J. (2009) Disuencies in non-stuttering adults
across sample lengths and topics. Journal of Communication Disorders 42(6): 414–427.
https://doi.org/10.1016/j.jcomdis.2009.06.001
Sporer, S. L. and Schwandt, B. (2006) Paraverbal indicators of deception: a meta-analytic
synthesis. Applied Cognitive Psychology 20: 421–446. https://doi.org/10.1002/acp.1190
,
Sti,J. B.,Corman,S. R.,Krizek,R.andSnider,E.(1994)Individual
dierences and changes in nonverbal behavior: unmasking the changing
faces of deception.Communication Research 21:555–581. https://doi.
org/10.1177/009365094021005001
Vrij, A.(2008)Detecting Lies and Deceit: Pitfalls and Opportunities. Chichester: Wiley.
Vrij, A. and Heaven, S. (1999) Vocal and verbal indicators of deception as a
function of lie complexity. Psychology, Crime and Law 5(3): 203–215. https://doi.
org/10.1080/10683169908401767
Vrij, A. and Mann, S. (2001) Telling and detecting lies in a high-stake situation: the case
of a convicted murderer.Applied Cognitive Psychology 15(2): 187–203.https://doi.
org/10.1002/1099-0720(200103/04)15:2<187::AID-ACP696>3.0.CO;2-A
Walczyk, J., Igou, F., Dixon, A. and Tcholakian, T. (2013) Advancing lie detection by
inducing cognitive load on liars: a review of relevant theories and techniques guided by
lessons from polygraph-based approaches. Frontiers in Psychology 4(14): 1–13. https://
doi.org/10.3389/fpsyg.2013.00014
Wright Whelan, C., Wagsta, G. and Wheatcro, J. (2013) High-stakes lies: verbal and
nonverbal cues to deception in public appeals for help with missing or murdered
relatives.Psychiatry, Psychology and Law21(4): 523–537. https://doi.org/10.1080/1321
8719.2013.839931
Zhang,Z., McGettigan,C. and Belyk,M. (2022)Speech timing cues reveal deceptive
speech in social deduction board games. PLoS ONE 17(2): e0263852.https://doi.
org/10.1371/journal.pone.0263852
Zuckerman, M., DePaulo, B. M. and Rosenthal, R. (1981) Verbal and nonverbal
communication of deception. In L. Berkowitz (ed.) Advances in Experimental Social
Psychology 14: 1–59. New York: Academic Press.
Anolli, L. and Ciceri, R. (1997) e voice of deception: Vocal strategies of naive
and able liars. Journal of Nonverbal Behavior 21(4): 259–284. https://doi.
org/10.1023/a:1024916214403
Bello, R. (2006) Causes and paralinguistic correlates of interpersonal equivocation.
Journal of Pragmatics 38: 1430–1441. https://doi.org/10.1016/j.pragma.2005.09.001
Granhag, P. A. and Strömwall, L. A. (2002) Repeated interrogations: Verbal and non-
verbal cues to deception. Applied Cognitive Psychology 16(3): 243–257. https://doi.
org/10.1002/acp.784
Heilveil, I. and Muehleman, J. T. (1981) Nonverbal cues to deception in a psychotherapy
analogue. Psychotherapy: eory, Research and Practice 18: 329–335. https://doi.
org/10.1037/h0088382
Parliament, L. and Yarmey, A. D. (2002). Deception in eyewitness identication. Criminal
Justice and Behavior 29. 734–746. https://doi.org/10.1177/009385402237925
Harrison, A. A., Hwalek, M., Raney, D. F. and Fritz, J. G. (1978) Cues to deception in an
interview situation. Social Psychology 41: 156–161. https://doi.org/10.2307/3033575
Farrow, T. F. D., Reilly, R., Rahman, T. A., Herford, A. E., Woodru, P. W. R. and Spence,
S. A. (2003) Sex and personality traits inuence the dierence between time taken to
tell the truth or lie. Perceptual and Motor Skills 97: 451–460. https://doi.org/10.2466/
pms.2003.97.2.451
‘ ’
Kraut, R. E. and Poe, D. (1980) On the line: the deception judgements of customs
inspectors and laymen. Journal of Personality and Social Psychology 36: 380–391.
https://doi.org/10.1037/0022-3514.39.5.784
Verschuere, B., Crombez, G., De Clecq, A. and Koster, E. H. W. (2004) Autonomic
and behavioural responding to concealed information: dierentiating orienting and
defensive responses. Psychophysiology 41: 461–466. https://doi.org/10.1111/j.1469-
8986.00167.x
,
Appendix A: Two excerpts from the interviews used in this study
e following are two examples of the exchange between the interviewer and
Speakers 50 and 85 for the prompts accompanying slide 38, which elicited
responses for all three categories of T, PL and UL. e interviewer’s utterances
have been transcribed by the present author, done in the same way as the
transcripts provided for the Speakers in the form of Praat text grids (Nolan et al.
2009). Gestures such as breathing, swallowing and indeterminate noises used for
backchanneling are labelled in block capitals. Instances where the interviewer’s
utterance overlaps with the speaker are indicated.
Only utterances marked with the relevant variables in bold (i.e. [RL], hes, [SP])
were measured for this study, indicated here with a frame, e.g.
Speaker 50: [RL] yeah
All other utterances in response to the interviewer were counted as follow-up
prompts and excluded. FP were transcribed as hes in the database, which has
been retained here. e textboxes with T, PL and UL mark the response type of
the relevant speech.
Interviewer: hes we understand that she works in a place
f – some four miles away how does she get
to work
Speaker 50: how does she get to work
Interviewer: Yeah
Speaker 50: [RL] hes well i think she she goes on she
cycles [SP]
hes because
T
Interviewer: hes she always cycles
Speaker 50: [RL] hes yeah she cycles always [SP]
like i she she’s a bit of an environmental-
ist she she likes the bike [SP]
it’s good exercise as well [SP] always
cycles
PL
Interviewer: she likes the bike
Speaker 50: yeah
Interviewer: hes she does not have a car
Speaker 50: [RL] no [SP] doesn't own a car PL
Interviewer: somebody in her household have a car
Speaker 50: [RL] oh hes i mean her hes [SP] a fr- she
lives with a friend and she has a car yes
UL
‘ ’
In the interview with Speaker 85, the utterance immediately aer the BACK-
CHANNEL from the interviewer (in italics) was not counted as part of the turn
and excluded from the data for this interview, since the speaker appeared to be
responding to the backchanneling, which was counted as a follow-up prompt.
Where the interviewer’s utterance overlaps with the speaker’s, individual judge-
ments were made on whether the speaker ignored it or changed their response
according to the new input. Only in the former case were the speaker’s utterance
before and aer the interviewer’s input counted as being within the same turn.
Where [NO RL FOR THIS TURN] is indicated, the speaker responds imme-
diately to the prompt, leaving no perceptible latency between the interviewer’s
utterance and the speaker’s. No data for RL were provided for this particular
response.
Interviewer: hes I understand that she works in a place
about four miles away hes how does she get
to work?
Speaker 85: [NO RL FOR THIS TURN] oh she cycles T
Interviewer: she cycles
Speaker 85: yeah she's prolic cyclist
Interviewer: always cycles
Speaker 85:
[RL] oh well i mean it be rain [SP] wind
[SP] anything [SP]
she’s on the bike [SP]
the boys love her for it
PL
Interviewer: i’m sure they do
hes so she doesn’t own a car
Speaker 85: [RL] hes n- well [SP]
i think she owned a car once [SP]
but it was hes [SP]
(SWALLOW) it wasn't great for her because
it mean it costs so much to run cars these
days doesn't it
Interviewer: BACKCHANNEL
Speaker 85: and bikes are brilliant you break the
chain put a new chain on it costs you four
quid at halfords
it's brilliant
,
Interviewer: hes so she doesn’t own a car you think now
at the moment
Speaker 85: own a car right now i don't think she does
no
doesn't own one
Interviewer: (OVERLAP) anybody in her
anybody in her household own a car
Speaker 85: [RL] 's always possible her husband does
this enigma who we don't know anything
about
UL
Interviewer: (OVERLAP) this you that you don’t know
anything
Speaker 85: [SP] i i haven't seen a car there no UL
Interviewer: right
Speaker 85: [SP] but they’ve got a garage [SP]
but then that may be from the old car [SP]
they've lived there a long time
UL
