Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Susceptibility to auditory hallucinations is associated with spontaneous but not directed modulation of top-down expectations for speech

Susceptibility to auditory hallucinations is associated with spontaneous but not directed... Auditoryverbalhallucinations(AVHs)—orhearingvoices—occurinclinicalandnon-clinicalpopulations,buttheirmechanismsremain unclear.Predictiveprocessingmodelsofpsychosishaveproposedthathallucinationsarisefromanover-weightingofpriorexpectations inperception.Itisunknown,however,whetherthisreflects(i)asensitivitytoexplicitmodulationofpriorknowledgeor(ii)apre-existing tendency to spontaneously use such knowledge in ambiguous contexts. Four experiments were conducted to examine this question in healthy participants listening to ambiguous speech stimuli. In experiments 1a (n=60) and 1b (n=60), participants discriminated intelligible and unintelligible sine-wave speech before and after exposure to the original language templates (i.e. a modulation of expectation). No relationship was observed between top-down modulation and two common measures of hallucination-proneness. Experiment 2 (n=99) confirmed this pattern with a different stimulus—sine-vocoded speech (SVS)—that was designed to minimize ceiling effects in discrimination and more closely model previous top-down effects reported in psychosis. In Experiment 3 (n=134), participants were exposed to SVS without prior knowledge that it contained speech (i.e. naïve listening). AVH-proneness significantly predicted both pre-exposure identification of speech and successful recall for words hidden in SVS, indicating that participants could actually decode the hidden signal spontaneously. Altogether, these findings support a pre-existing tendency to spontaneously draw upon prior knowledge in healthy people prone to AVH, rather than a sensitivity to temporary modulations of expectation. We propose amodelofclinicalandnon-clinicalhallucinations, acrossauditoryandvisualmodalities, withtestablepredictionsforfutureresearch. Keywords: consciousness; ketamine anesthesia; EEG markers of consciousness; perturbational complexity index Introduction 1992). Although not always framed as a ‘top-down’ model of hallucinatory experience, this grounded much research in the Hallucinations have long been considered a product of top-down metacognitive domain, consistent with cognitive approaches to processes: what the mind brings to our perception of the world, psychosis in clinical practice (Morrison et al. 1995). not the other way round (Esquirol 1832). Auditory verbal hal- lucinations (AVHs) in particular have been studied extensively Recent interest in predictive processing approaches has because of their association with schizophrenia, occurring in reframedtheputativeroleoftop-downprocessesinhallucination. 60–90% of cases (Bauer et al. 2011) and at rates that are often Underthepredictiveprocessingframework(PPF),allofperception double those seen for other modalities (Waters et al. 2014). AVHs and cognition is the result of a trade-off between generative mod- have been proposed to result from various internal sources such elsoftheworld,shapedbypriorexpectationsandpredictionerror, as memories, imagery, and self-talk or inner speech (Mintz and i.e. the gap between expectation and sensory input (Clark 2013; Alpert 1972; Waters et al. 2003; Seal et al. 2004). Difficulties in dis- Hohwy 2014). Hallucinations have been posited as an imbalance tinguishing the internal from external were interpreted as a prob- betweenpriorexpectationandpredictionerror(FletcherandFrith lemwith‘realitymonitoring’,inwhichdisruptionstosourcemon- 2009; Jardri and Denève 2013; Powers et al. 2016). Such accounts itoring could explain how self-generated cognitive states could have been argued to be consistent with source-monitoring theo- becomeperceptualexperiences(Feinberg1978;Bentall1990;Frith ries (Wilkinson 2014; Griffin and Fletcher 2017 ; Corlett et al. 2019) Received: 1 October 2021; Accepted: 13 January 2022 © The Author(s) 2022. Published by Oxford University Press. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. 2 Alderson-Day et al. and may even reflect a generalization of prediction mechanisms we followed a similar test-train-test procedure to Teufel and inherent in earlier theories (Pickering and Clark 2014). Neverthe- colleagues, using the CAPS (Bell et al. 2006) and PDI (Peters et al. less,theyinvolveashiftinemphasisawayfromthemetacognitive 2004) to measure unusual perceptual experiences and delusional monitoring of self, focusing instead on expectation and learning beliefs.Experiment1b(n=60),runinparallel,usedanalternative as being central to hallucination. measure specific to AVH-proneness: a version of the Launay– Supporting evidence for a PPF approach to hallucinations was Slade Hallucination Scale-Revised (Bentall and Slade 1985b; Mor- providedbyTeufeletal.(2015),inastudyofindividualswithanat- rison et al. 2000). It also included participants intentionally riskmentalstateforpsychosis.Patientsandhealthycontrolswere recruited to expand the potential range of individual differences asked to discriminate monochrome Mooney (1957) images, before in hallucination-proneness (specifically, people with a history of andafterexposuretotheiroriginaltemplates(picturesofhumans imaginary companions; Fernyhough et al. 2019), and an added and animals). While both groups improved their discrimination condition that sought to further prime potential templates for after viewing the templates, clinical participants showed signifi- speech. Based on the modulation hypothesis, we hypothesized cantly enhanced discrimination compared to controls, consistent thatimprovementsindiscriminationfollowingtemplateexposure with top-down information being given greater weight in their should be associated with higher hallucination-proneness. perceptual processing. Teufel and colleagues then replicated this One problem with SWS is that some participants attempting finding in a sample of 40 healthy participants rated for psychosis- to understand it may go from not understanding it all before pronenessonmeasuresofhallucination-likeexperiences(r=0.42, training to suddenly understanding it all, while others may spon- theCardiffAnomalousPerceptionsScale;Belletal.2006)anddelu- taneously learn to decode it. This learning profile can make it sional traits (r=0.33, the Peters Delusion Inventory; Peters et al. hard to compare with perceptual learning for Mooney images. So, 2004), with higher scores on these scales being associated with a in Experiment 2 (n=99), we tested the same modulation effect greater improvement in discrimination following exposure to the but with a new stimulus, sine-vocoded speech (SVS). We devel- templates (Teufel et al. 2015). opedthisparticularstimulussetwiththeaimofofferingatighter These findings speak to visual processes—but what of voices, control on some of the potential learning effects inherent to the most common kind of hallucination in psychosis? Various SWS comprehension—making it more comparable to Teufel et al. source-monitoring studies have demonstrated biases in auditory (2015). Asinexperiments1aand1b, weexpectedthatmodulation signal detection in people with hallucinations—often on white of discrimination would be related to hallucination-proneness. noise tasks (Bentall and Slade 1985a)—but facilitatory effects like Finally, having tested modulation effects using SVS, we those described by Teufel and colleagues have not typically been returned to the behavioural design from Alderson-Day et al. studied. (2017),examiningSVSperceptionundernaïvelisteningconditions Analogous to Mooney images, sine-wave speech (SWS; Remez (Experiment 3, n=134). According to the naïve listening hypothe- et al. 1981; Rosen et al. 2011) is a perceptually ambiguous stimu- sis, we predicted that hallucination-proneness would be higher in lus derived from speech that allows for exploration of top-down those who were quicker to recognize that SVS contained hidden effects on perception. SWS is not usually identified as intelligible speech. After the naïve listening procedure, we also tested them speech by naïve listeners; instead, it requires prior training to be ontheirmemoryforthehiddenwords,thereforeprovidingamore recognized and understood. In a recent study, a sample of non- objectivetestofspontaneousdecodingofthehiddenspeech.Data clinicalvoice-hearers(NCVH)—individualswithfrequentAVHbut and analysis code for each of the experiments are available via noneedforclinicalcare(Johnsetal. 2014; Petersetal. 2016)—were OSF. scanned in fMRI while naïvely listening to SWS (Alderson-Day and Lima et al. 2017). Instead of being told to listen for speech, Experiment 1a participants were instructed to listen for an unintelligible target Modulating prior knowledge of sine-wave speech. sound amidst a range of SWS stimuli. Despite this, a majority Theaimofourfirstexperimentwastodevelopamodulationof of the NCVH group identified speech in the SWS spontaneously expectationintheauditorymodalityandtoseehowthisrelatedto and without any training. When asked to estimate the point at hallucination-proneness scores. In contrast to Teufel et al. (2015), which they recognized the hidden speech (visual markers had who used 12 blocks of before/after trials, we chose to play all been displayed indicating numbered ‘rounds’ during the scan), 90 trials, train on the whole set, and then retest for all trials NCVH participants reported doing so significantly earlier than a (a ‘one-shot’ procedure). This was chosen to minimize any poten- matched control group. Subsequent tests of discrimination fol- tialtrainingeffectsoccurringacrossmultipleblocksoftestingand lowing the ‘reveal’ that speech was present, failed to identify training. We predicted that higher CAPS scores would be associ- any group differences. This suggested that voice-hearers may ated with greater increases in discrimination following template automatically draw upon top-down resources—such as speech exposure. We also explored this effect for delusion-proneness templates—whenfacedwithambiguoussensoryinput(Alderson- scores on the PDI. Day et al. 2017). Both experiments are consistent with top-down processing Method being linked to hallucinations, but they highlight contrasting Participants effects: a modulatory effect (Teufel et al. 2015) and a naïve listen- A convenience sample of 60 participants was recruited from a ingeffectonperception(Alderson-Dayetal. 2017). Theyalsodiffer university cohort (age M(SD)=21.22 (3.11), range 18–32years, 18 in design and stimuli, making it challenging to directly compare male). Individuals were invited to take part if they were native them. In this paper, we aimed to draw together these effects, English speakers with no hearing impairments or any previous adapting the SWS procedure across a series of experiments with psychiatricorneurologicaldiagnoses.Participantsreceivedcourse healthy participants to explore top-down effects on audition. We began with the original SWS stimuli used in Alderson-Day et al. While a convenience sample, this number was nevertheless sufficiently (2017) deployed in two parallel experiments examining modula- powered to identify an effect in a similar range to Teufel and colleagues’ tion effects (Experiments 1a and 1b). In Experiment 1a (n=60), observed effect size for the CAPS (r=0.42, 90% power, min. sample=50). Susceptibility to auditory hallucinations 3 credit or a gift voucher in recognition of their time. For this The Peters Delusion Inventory—21 item version (PDI; Peters et al. and the remaining experiments, written informed consent was 2004) is a shortened adaptation of the original 40-item PDI (Peters obtained for all participants and all procedures were approved by et al. 1999). Both measures have been used extensively as a mea- a university ethics committee. sure of proneness to unusual beliefs in the general population, have good convergent validity with other measures of schizo- typy, and have strong internal reliability (e.g. alpha>0.8). The Materials and procedure PDI has an identical structure to the CAPS (the latter being mod- SWS discrimination task elledontheformer).Frequencyofbeliefwasincludedasthemain SWS is created by tracking and modeling the formant tracks of outcome. spoken sentences using a sine-wave tone. This procedure can Our analytic approach sought to first assess changes in dis- be used to create potentially intelligible stimuli (in which the crimination and bias variables following the exposure phase, frequency and amplitude tracks of the same original sentence using paired t-tests. We then (i) followed Teufel and colleagues’ arecombined)—orunintelligiblestimuli(combiningthefrequency analysis by testing the relationship between CAPS scores and and amplitude information of two different sentences). Both are d-prime improvement using correlational analysis, and (ii) used typically perceived as unintelligible, but potentially intelligible partial correlation to test this association while controlling for SWS typically becomes comprehensible following training and confounds such as baseline performance. Correlations with exposure to 2–3 template sentences (Rosen et al. 2011). Experi- other change scores (i.e. beta and C), relations to pre-exposure ment1usedthesameSWSstimuliasinAlderson-Dayetal.(2017), 2 performance and associations with the PDI were included for which were first developed by Rosen et al. (2011). The original exploratory purposes. sentences were taken from the Bamford-Kowal-Bench (BKB) sen- tence set (Bench et al. 1979). Participants completed the task in a quiet university room. The task was presented using Psych- Results and discussion toolbox in MATLAB 2016 on a Windows PC with a 17'' monitor, Table 1 shows signal detection outcomes for the SWS discrim- using Sennheiser headphones for stimulus delivery. See Fig. 1 ination task. As would be expected, performance significantly for a summary of the design of this and the other reported improved following exposure to the original (i.e. non-masked) experiments. sentences, as indicated by an increase in d . However, bias also The SWS discrimination task was divided into two runs of 90 significantly increased, with participants being more likely to trials (45 intelligible SWS, 45 unintelligible SWS) occurring before say that speech was present after template exposure [before hits and after participants heard each of the original sentences that M(SD)%=67.9% (21.1%), false alarms M(SD)=20.6% (12.3%); after theintelligibleSWStrialswerebasedon(‘templateexposure’).On hits M(SD)%=88.6% (14.2%), false alarms=26% (14.9%)]. eachrun,participantslistenedto2.5sclipsofSWSandwereasked to decide whether the speech was present or not for each trial, Testing a modulation effect via bivariate allowing for signal detection measures to be calculated based on correlation hitrates(intelligibletrialsmarkedascontainingspeech)andfalse Following Teufel et al. (2015), we first tested for bivariate correla- alarm rates (unintelligible trials marked as containing speech). tionsbetweenchangeind ,CAPS,andPDIscores.Despitetheclear Signal detection theory (Stanislaw and Todorov 1999) was used to change in discrimination following exposure, no correlation was calculate discrimination (d ), plus two measures of bias: criterion observed between CAPS scores and change in d (Pearson’s prod- (C) and beta (β), the measure most typically used in source mon- uct, r (58)=0.02, P=0.864, 95% CI=−0.23:0.28), contrary to the itoring research on hallucinations (Brookwell et al. 2013). Where modulation hypothesis (see Fig. 2). A one-sided Bayesian analysis hitratesandfalsealarmswere0and1,theMacmillanandKaplan (usingJASPv.0.8.6withdefaultpriors)indicatedaBFof0.19forthe (1985) method was used (i.e. zero scores replaced with 0.5/n and 1 experimental hypothesis and 5.39 for the null (i.e. good evidence replaced with (n−0.5)/n). foralackofanyeffectofinterest).Asimilarresultwasevidentfor PDI scores (r (58)=0.11, P=0.400). Questionnaires In each experiment, questionnaires were collected after task- Controlling for baseline performance and other based measures were taken. confounds The Cardiff Anomalous Perceptions Scale (CAPS; Bell et al. 2006) is Analysisofdifferencescorescanbeaffectedbythevaluesatbase- a commonly used scale of hallucination-proneness that assesses line (Senn 2006) and are often thought to be less reliable than a range of unusual perceptual experiences—including auditory, the measures they derive from. To account for this, we also ran visual and gustatory phenomena—across 32 items. It correlates partial correlation tests controlling for baseline discrimination with other measures of schizotypy and hallucinations, such scores, using the ppcor package in R (Kim 2015). This adjustment as the Oxford Liverpool Inventory of Feelings and Experiences made no difference to the results (r=−0.04, P=0.758) suggesting (OLIFE; Mason et al. 1995), and has strong internal reliability that the overall null result was unlikely to be driven by baseline (alpha=0.87). Participants are asked to indicate whether they performance differences. haveeverhadaspecificexperience,andifso,howdistressing,how intrusive, and how frequent the experience was (on a 1–5 scale). Relations to bias To assess the general tendency to experience hallucinations, here Finally, we also tested for any relations between bias (C and we used the total frequency as the main CAPS outcome. β) and proneness to psychotic experiences. No significant cor- relations were observed between task and questionnaire scores (all P>0.10; see Supplementary Materials). Rosen and colleagues’ SWS stimuli were also noise-vocoded. This step wasomittedinouruseofthestimuliforthepresentpaperandinAlderson-Day The results of Experiment 1a, therefore, did not support etal. (2017), asnoise-vocodingcaninduceaneffectakintowhispering, andcan the idea of a modulatory effect of expectation being related to maketheunderlyingsinewavescohere—bothofwhichcouldpotentiallyreveal the underlying speech signal. hallucination-proneness. When exposed to new information via 4 Alderson-Day et al. Figure 1. Overview of experiments. (A) Experiment 1a: Participants heard 90 trials comprised of potentially intelligible and unintelligible sounds and judged whether each sound contained speech (pre-exposure). They were then exposed to the target clear speech exemplars from which the intelligible trials were made (exposure) and then asked again to judge which trials contained speech (post-exposure). (B) Experiment 1b: Participants took part in the same paradigm as Experiment 1a but half the participants were primed by listening to a busy auditory scene and the other half were not. (C) Experiment 2: Participants heard blocks of 10 trials using the same pre-exposure, exposure, post-exposure cycles in Study 1. (D) Experiment 3: Participants took part in a naïve listening experiment in which they were tasked with identifying sounds with a specific acoustic quality (noise-vocoded sounds). They were not informed that some sounds contained speech. They were then asked whether they had heard any speech in the naïve listening task and took part in a memory recognition test to see if they remembered the intelligible trials. They were then exposed to the clear speech targets and tested on their identification of speech Susceptibility to auditory hallucinations 5 Table 1. Signal detection outcomes for the SWS discrimination task in experiment 1a Before After M SD M SD Change statistic T(Z) P d(r) d 1.47 0.77 2.17 0.81 −8.63 4.809e−12 −0.90 C 0.19 0.45 −0.37 0.44 −6.25 4.212e−10 0.81 β 1.63 1.87 0.60 0.51 −5.74 9.202e−09 0.74 Note: Higher values of d indicate increased sensitivity to detect speech. Scores below 0 for C and 1 for beta indicate a bias to indicate speech is present. Wilcoxon tests were used for C and beta due to non-parametric data. Experiment 1b, run in parallel to the first, was designed to address potential concerns about the level and specificity of hallucination-proneness. First, as an alternative to the CAPS, we used a revised version of the Launay–Slade Hallucination Scale (Morrison et al. 2000), with a specific focus on auditory experi- ences (McCarthy-Jones and Fernyhough 2011). Second, intending to gather a wider range of unusual experiences, we explicitly set out to recruit individuals with a history of having imagi- nary companions (ICs). Engaging with imaginary companions has been proposed to bear commonalities with hallucinatory experi- ences (Pearson et al. 2001), even though there is no good evidence that they are a developmental marker for later psychopathology (Taylor 1999; Maijer et al. 2019). Specifically, there is evidence to Figure 2. Comparing discrimination pre- and post-template exposure (A) suggest that having an IC as a child is associated with both ele- and the relation of performance change to hallucination-proneness (B) vated hallucination-proneness and bias in auditory signal detec- tionskillsasanadult(Fernyhoughetal.2019).Inaddition,children the original sentence templates, participants consistently per- with ICs are more likely to hear words amidst jumbled speech, formed better in terms of their speech vs. non-speech discrimi- whichissimilarinmanywaystoeffectsseenforSWS(Fernyhough nation and increased their bias to state that speech was present et al. 2007). (across intelligible and unintelligible SWS stimuli). None of these Finally,wealsoattemptedtoprovideasecondtestofthemodu- performance changes were related to hallucination-proneness lationofexpectation,bypriminghalftheparticipantswithashort scores on the CAPS or delusion ratings on the PDI. listening activity (listening to a recording of a conversation in a Two limitations are important to consider. The first is that busy room) before attempting the same task as Experiment 1a, using hallucination-proneness in non-clinical analogue samples i.e. discrimination before and after exposure to the sentence tem- hasbeenquestionedforitsabilitytoidentifyindividualswithtruly plates. Reasoning that directing participants to listen for speech hallucinatory experiences (Stanghellini et al. 2012). If valid, this under suboptimal conditions should prime both the expectation couldleadtotheconcernthatcorrelationsbetweentasksandself- of speech and top-down templates for speech, we predicted that report will be very low and very hard to capture, given the low primed participants would go on to show greater speech discrim- base level and minimal variation in proneness scores. The spread ination of SWS in the subsequent task, even before template of CAPS scores shown in Fig. 2 is comparable to prior research of exposure. If this could be demonstrated, it would represent a this kind and not insubstantial when compared to clinical data more naturalistic modulation of expectation, by indirectly prim- (Bell et al. 2006). Nevertheless, directed recruitment of members ing generic speech templates that could assist in the disambigua- of the general population with higher levels of hallucination- tion of the SWS stimuli. The design for Experiment 1b, therefore, proneness could provide greater variation and more opportunity mixed a between-groups approach (prime vs. no-prime) and a to examine how changes in expectation relate to unusual sensory within-subjects approach (before vs. after template exposure). As experiences. in Experiment 1b, we hypothesized that greater improvements in A second concern is that measures of hallucination-proneness discrimination scores on the SWS task would be associated with can yield inconsistent results, and there is currently no ‘gold greater LSHS scores. standard’ for assessing such experiences in the general popu- lation. For the purposes of replication, we used the CAPS but couldhaveinsteadincludedtheLaunay–SladeHallucinationScale Method (Bentall and Slade 1985b) which is arguably a more commonly Participants used scale in prior research on hallucinations. Moreover, the Sixty participants (age M(SD)=23.22 (4.76), range 18–43years, CAPS asks about hallucinations across a range of modalities, 14 male) were recruited from university settings, social media, whereas prevalence rates for AVH—and the auditory nature of and via word-of-mouth. Exclusion criteria were identical to the SWS task—may warrant a more specific measure of auditory Experiment 1a. Within the 60, it was possible to recruit 22 hallucination-proneness. people with a history of having imaginary companions as children, of whom 14 were able to provide parental Experiment 1b verification of their childhood IC—a validation step consid- Modulating prior knowledge with a wider range of hallucination- ered good practice in IC research (Fernyhough et al. 2007, proneness. 2019). 6 Alderson-Day et al. Materials and procedure The same procedure and task structure were used for the SWS discrimination task as in Experiment 1a. In addition, half of the participants completed a priming activity before the discrimina- tion task. The CAPS and PDI were replaced with a version of the Revised Launay–Slade Hallucination Scale. Listening prime task Thirty participants were asked to complete the priming activity before the SWS discrimination task. Participants were given a worksheet and were asked to circle words that they heard being mentioned in a 3-minute pre-recorded conversation between five girls. The recording was layered with white noise to increase the Figure 3. Change in discrimination pre- and post-exposure divided by difficulty in discerning what was being said. The remaining 30 priming group (A) and relation to hallucination-proneness (B) participants were instructed to close their eyes and count their breaths for 3minutes in silence, as timed by the experimenter. before, there was a significant increase in discrimination follow- Questionnaires ing exposure. Pairwise t-tests showed that this was also the case Experiment 1b included a version of the Revised Launay–Slade for both measures of bias (i.e. lower scores, indicating a greater Hallucination Scale (Morrison et al. 2000; McCarthy-Jones and tendency to say speech is present). Fernyhough 2011). Since the development of the original scale by The overall mean LSHS-A score of 9.40 places this sample as Bentall and Slade (1985b), numerous versions of the LSHS have having a mean level of hallucination-proneness comparable to been used to assess hallucination-proneness in the general popu- other samples with childhood histories of ICs, and slightly higher lation. Here we used a five-item version in which all of the items than large samples without any IC history (M=8.76; Fernyhough related specifically to auditory experiences, which participants et al. 2019). As would be expected, participants with an IC ratedforfrequencyonascalefrom1(Never)to4(AlmostAlways). in the present sample had significantly higher LSHS-A scores This version was developed by McCarthy-Jones and Fernyhough (IC M(SD)=10.55 (2.89); No IC M(SD)=8.74 (2.41); Z=−2.56, (2011) following a revision by Morrison et al. (2000). This version P=0.011, r=0.33). has satisfactory internal reliability (typically alpha => 0.7) and However, few relations between SWS task outcomes and has been used to explore task-to-questionnaire relations in vari- questionnaire scores were observed even with higher rates of ousstudiespreviously(e.g.Garrisonetal.2017;Alderson-Dayetal. hallucination-proneness in the sample. A stronger relationship 2019). was evident between the improvement in d and LSHS-A scores (r=0.22, see Fig. 3B), but this was still non-significant on a Spear- man’s test (P=0.089) and Bayesian analysis little evidence for the Results and discussion hypothesized effect (one-sided BF =0.81, or 1.25 in favour of the Testing the priming effect null). Partial correlation, controlling for baseline d scores, also A 2×2 mixed ANOVA (prime group × pre-/post-exposure) was showed no clear association between hallucination-proneness used to assess the effect of the priming condition on discrim- and improvement in performance (r=0.07, P=0.611). ination, plus any interaction it had with exposure to the sen- Relations to bias tence templates (i.e. before vs. after). As in Experiment 1a, there was a clear increase in discrimination following expo- Correlations with bias metrics (C and β) were non-significant sure to the template sentences (F (1,58)=81.31, P=1.24e-12, (all r<0.21, all P>0.08). Pre-exposure discrimination actually neg- η =0.58). But despite the manipulation, no main effect of atively correlated with LSHS scores (r=−0.34, P=0.008, uncor- priming was observed on the discrimination task following the rected), suggesting more hallucination-prone people were worse listening activity (F (1,58)=0.98, P=0.327, η =0.02, nor any at discriminating speech than controls before hearing the sen- interaction effect between priming group and template exposure tence templates. 2 3 (F (1,58)=0.53, P=0.471, η =0.009; see Fig. 3A). Pre-exposure The results of Experiment 1b, therefore, provided a further d scores were very similar in each group [Primed M(SD)=1.68 test of the modulation hypothesis in the auditory domain, but (0.75); Control M(SD)=1.45 (0.77)], as were post-exposure scores could not support it: improved discrimination following template [Primed M(SD)=2.39 (0.79); Control M(SD)=2.29 (0.66)]. Compar- exposure did not significantly relate to auditory hallucination- isonsofpre-exposurebias(both β andC)werealsonon-significant proneness even when including individuals with an IC—therefore (all P>0.300; see Supplementary Materials, Experiment 2). sampling a broader continuum of hallucination susceptibility. In addition, priming expectation for speech with a further manipu- lationyieldednodifferenceindiscriminationperformance. Taken Testing the modulation effect together, the results of experiments 1a and 1b failed to show Given the lack of differences on any task measure, we subse- an effect in the auditory domain comparable to Teufel et al.’s quently combined the priming groups to facilitate comparison modulation effect. with Experiment 1a. Table 2 shows the SWS task outcomes. As Our findings may suggest that a sensitivity to directed mod- ulation of expectation for speech is not part of the putative A Bayesian mixed ANOVA in JASP indicated strong evidence for an expo- mechanism underlying hallucinations. If correct, this would sure effect (BF=5.90) and weak evidence for either an effect of priming group raise problems for a ‘strong priors’ account of the phenomenon (BF=0.44) or an exposure*prime interaction (BF=0.43; BF inclusion statistics reported). (Corlett et al. 2019).However,thefindingsmaystillreflectunusual Susceptibility to auditory hallucinations 7 Table 2. Signal detection outcomes for the SWS discrimination task in experiment 1b Before After M SD M SD Change statistic T(Z) P d(r) d 1.56 0.76 2.34 0.72 −9.05 9.298e-13 −1.05 C 0.29 0.63 −0.42 0.52 9.65 9.59e-14 1.22 β 2.58 3.04 0.88 1.62 −5.79 6.741e-09 0.75 Note: higher values of d indicate increased sensitivity to detect speech. Scores below 0 for C and 1 for beta indicate a bias to indicate speech is present. Wilcoxon tests were used for beta due to non-parametric data. propertiesofSWSitself.Forexample,SWScomprehensionfollow- Materials and procedure ing exposureand trainingcan show something akin to a ‘pop-out’ Auditory stimuli effect, where suddenly new stimuli can be easily understood. In SVS is similar to SWS. However, rather than tracking only the addition, they offer potentially different opportunities for percep- first three formants of speech, the sinewaves are synthesized at tuallearning,comparedtoMooneyimages(Mooney1957). Forthe the centre frequency of a logarithmically spaced bank of filters latter, the level of visual noise is high compared to the level of spanning a broad frequency range (up to 5kHz). Like SWS, SVS repeated signal across trials, meaning that template exposure is sentences can be rendered intelligible and recognizable as speech required per image. For SWS, template exposure typically permits when participants are aware that it is a speech stimulus (Souza more generalized improvements in discrimination across trials and Rosen 2009). SVS can also be rendered unintelligible by flip- (due to similarities in the underlying structure of speech sounds). ping the frequency mapping of the original sentence (e.g. pushing It is therefore important to tightly control for pre-exposure lev- energy in high-frequency bands into low bands and vice versa), els of performance and ensure that difficulty levels are main- providing an ideal control stimulus, with similar complexity and tained for speech discrimination. In the following experiment, we acoustic structure. introduced a new stimulus that could be used to address these The BKB sentences (Bench et al. 1979) were recorded by a male issues. speaker at a sample rate of 22.05kHz. Each sentence was digitally filtered using either 8 or 16 bands, with sixth-order Butterworth IIR filters in MATLAB. Filter spacing was based on equal basilar Experiment 2 membrane distance (Greenwood 1990) across a frequency range of 100–5000Hz. Next, the output of each band was half-wave rec- Modulating prior knowledge of sine-vocoded speech (SVA) with tified and low-pass filtered (fourth-order Butterworth) at 30Hz varying levels of difficulty. to extract the amplitude envelope. The envelope was then mul- Adapting and extending the SWS developed by Rosen et al. tiplied by a tone carrier at the band centre frequency for each (2011) is challenging, as the stimuli were originally hand-edited filter. The resulting signal (envelope×carrier) was filtered using to closely map the formant contours of speech. To address the the same bandpass filter as for the first filtering stage. RMS level factors described above, we deployed a different auditory stimu- was adjusted at the output of the filter to match the original lusinExperiment2—SVS—andmixeditwithunintelligibleSVSto analysis, and the signal was summed across bands. add a source of auditory noise. This provided: (i) a way to auto- When larger numbers of filter bands (e.g. 16 vs. 8) are used to mate the generation of a degraded speech stimulus (rather than synthesizeaspokensentencethisincreasesthespectralinforma- using hand-crafted SWS), allowing the use of a larger number tion in the signal with a resulting increase in intelligibility (Souza of spoken sentences, and (ii) greater control over manipulating and Rosen 2009). task difficulty. We tested pre- and post-exposure identification Sine-vocoding was used to make two types of stimulus: an of degraded speech in a larger sample and its association with intelligible and unintelligible SVS condition. For the Intelligible hallucination-proneness at an increased level of difficulty com- SVS condition, intelligible SVS sentences were mixed with an pared to the previous experiments. The task was administered in unintelligible sine-vocoded sentence that acted as a competing prior knowledge exposure cycles of ten trials, rather than using noise source. This was designed to make stimulus identification a one-shot exposure approach. This was done in part to attempt more difficult reducing learning in the pre-exposure phase and greater parity with the kind of procedure used by Teufel and col- ensuring greater dynamic range in the prior knowledge advan- leagues, and to provide greater access to top-down information tageprovidedbyhearingclearspeechtemplates. TheRMSlevelof after exposure within each cycle. the intelligible sentence was rendered at different levels of inten- sityrelativetotheunintelligiblesentencebeforemixingthem[e.g. a differing signal-to-noise ratio (SNR) was used]. This, alongside Method manipulating the number of vocoding bands, provided a way to Participants manipulate the difficulty of speech identification. Asampleof99participantswasrecruitedfromuniversitysettings, For the Unintelligible SVS condition, two sentences were sine- social media, and via word-of-mouth (age M(SD)=21.58 (3.34), vocoded and frequency flipped and mixed in an equivalent SNR range 18–34years, 45 male). Exclusion criteria were identical to as the Intelligible SVS condition, with one unintelligible sentence thepreviousexperiments.Allparticipantsprovidedinformedcon- arbitrarily assigned to be of greater intensity than the other. This sent in accordance with the approval of the relevant ethics com- ensured that the Intelligible SVS and Unintelligible SVS condi- mittee. Due to experimenter error, the questionnaire data were tions were of equivalent complexity and overall intensity. A set not complete for 12 participants: one participant did not have PDI of stimuli were synthesized from +6dB to −6dB in 3dB steps data and 11 did not have CAPS data. Participants received course using 8 and 16 bands. The sentences composing the intelligible credit for taking part. and unintelligible conditions were mutually exclusive. We have 8 Alderson-Day et al. madethisfullsetofstimulusmaterialsavailablehere(eventhose Results and discussion conditions not used in this article) to facilitate future research: Assessing SVS discrimination across trials and at https://osf.io/yrn9j/. different levels of difficulty Aone-wayindependentANOVAwiththegroupasafactorshowed Auditory task that there was no evidence of a difference in d’ between the five data collection groups, so data were pooled for the com- Participants attended to sounds presented using MATLAB on a mon8bands +6dBcondition(F(4,94)=1.83,P=0.130, η =0.07). laptop using Sennheiser HD 206 headphones. They were ran- Speechidentificationaccuracyinthisconditionwasabovechance domly assigned to five groups which each received two different before prior knowledge exposure (t(98)=10.96, P=9.952e-19), but sets of auditory stimuli (approximately 20 participants in each crucially was significantly increased after prior knowledge expo- group). These two sets differed in either SNR and/or the number sure (t(98)=9.34, P=3.217e-15, d=0.44, see Fig. 4A and Table 3). of bands. Informal piloting indicated that the 8 band +6dB con- A2×8repeatedmeasuresANOVAwithfactorspriorknowledge dition provided an appropriately challenging listening level, such exposure(pre-/post)andblock(1–8)wasconductedtounderstand that accuracy would be above chance, but not at ceiling. Each how d changed across the experiment. This indicated a signifi- group was tested on this common 8 band +6dB condition, plus cantexposure×blockinteraction(F(6.07,595.19)=2.23,P=0.038, oneothercondition. Thiscommonconditionwasincludedtopool η =0.02, see Fig. 4B). Follow-up repeated measure one-way the data across groups to test for the relationship between sig- ANOVAs indicated a change in d’ across blocks before exposure nal detection measures and questionnaire responses. The second topriorknowledge(F(6.23, 610.46)=3.33, P=0.003, η =0.03)but condition had a different signal-to-noise ratio and/or a different notafterit(F(6.18,605.54)=0.86,P=0.530, η =0.009).Inthepre- numberofvocodingbandsandwasincludedtoscopehowspeech exposure phase, the change in accuracy increased linearly with detection accuracy was influenced by the SNR and number of block progression (F(1, 98)=14.83, p=2.100e-4, η =0.13). There bands. was also the main effect of block such that accuracy in general Data from these conditions indicated that the SNR level and increasedacrossblocks(F(5.94,582.36)=2.17,P=0.045, η =0.02) the number of bands had a significant effect on participant per- and from pre- to post-exposure (F (1, 98)=64.69, P=2.054e-12, formance and confirmed the observation that the 8 band +6dB η =0.40). Hence, even with these more challenging stimuli, par- condition provided an appropriately difficult listening experience ticipants continued to improve in their ability to detect speech (see Supplementary Materials, Experiment 2). The specific set of in SVS in the pre-exposure phase, demonstrating the learning sentences used in each condition and order of the conditions was opportunity inherent to SWS/SVS. counterbalanced across participants to ensure that participants did not hear repetitions of any sentences across the experimental session and to reduce order effects. Testing for a modulation effect using SVS Testing in each auditory condition took around 20minutes As in the previous studies, there was no evidence of a significant (40minutes total). Before each condition participants received a relationship between change in d’ and the CAPS (rs (88)=−0.02, short training session in which they were introduced to the intel- P=0.852) or the PDI (rs (98)=−0.12, P=0.257), see Fig. 4C). ligible and unintelligible stimuli. They were informed that they One-sided Bayesian correlations assessing a positive association would hear a 50:50 ratio of intelligible to unintelligible trials in between d’ and the questionnaire measures were used to assess the forthcoming experiment and needed to judge whether each the relative evidence for the null as compared to the experimen- trial contained speech or not. tal hypothesis. These tests provided strong evidence for the null In the experiment, sounds were presented in blocks of ten in hypothesis (both one-sided BF >11). Partial correlation analy- a randomized order (five intelligible trials and five unintelligible sis, controlling for baseline discrimination scores, also provided trials). In the pre-exposure phase, on each trial participants indi- no evidence for an association between d’ change and the CAPS cated whether the sentence contained speech or not. They then (rs (88) =−0.03,P=0.815)andPDIscores(rs(98)=−0.12,P=0.255). receivedexposuretopriorknowledge: theyheardeachtargetsen- Wealsotestedtoseeifchangeind correlatedwitheithertheCAPS tence presented in the original clear speech and saw the written orPDIontheadditionalacousticconditions(e.g. thosedifferingin transcriptofthesentence.Inthepost-exposurephase,theyheard number of bands and SNR level), but this was not the case (see the intelligible and unintelligible sentences that they heard in Supplementary Materials, Experiment 3), suggesting that varying the pre-exposure phase in a different randomized order and were the difficulty level made no difference to the lack of association again asked to indicate which sentences contained speech. The with hallucination-proneness. five transcribed target sentences remained on the right-hand side of the screen during the post-exposure phase to reduce memory Assessing relationships with bias demandsandtomaximizethepriorknowledgebenefit. Thiscycle repeated 8 times, each time with a different set of 10 sentences, A decrease in C after prior knowledge exposure reflected an such that 80 trials were presented in each condition (40 intelligi- increased bias to report the presence of speech [C (t (98)=2.99, ble; 40 unintelligible) and 160 trials werepresented in total across P=0.004, d=0.22)]. However, C did not differ significantly the experiment. from zero either before or after prior knowledge exposure (both Followingadministrationoftheauditorytasks, totalfrequency Ps>0.194).Anon-parametricWilcoxontest—toaccountforadevi- scores were collected for the CAPS (Bell et al. 2006) and the PDI ationinnormality—alsoindicatedthatbetavaluesdidnotchange (Peters et al. 2004). Signal detection measures were calculated as significantly ( z=1.58, P=0.113). These findings suggest that the in the previous experiments. The total testing time was around SVS stimuli provide a more controlled modulation of discrimi- 50minutes. nation, while holding bias relatively more constant. Beta and C Susceptibility to auditory hallucinations 9 ′ ′ Figure 4. (A) d (left) and C change (right panel) with prior knowledge exposure, (B) d change over time in the common 8 band, +6dB condition. Note that the grand mean for d values for the by block analyses differs to the main analysis because the adjustment for extreme values was conducted by block in this instance rather than across the whole experiment (Macmillan and Kaplan 1985), (C) showing the lack of evidence in support of a relationship between d change and the CAPS (left) and PDI measures (right panel) Table 3. Signal detection outcomes for the SVS speech detection task for the 8 band, +6dB condition. Wilcoxon tests were used beta due to non-parametric data Before After M SD M SD Change statistic T(Z) P D(r) d’ 1.20 1.09 1.72 1.25 9.34 3.217e-15 0.44 C 0.04 0.48 −0.06 0.45 2.99 0.004 0.22 β 1.28 1.23 1.14 0.98 1.59 0.113 0.16 change did not correlate with either the CAPS or PDI frequency Our difficulties in demonstrating modulation effects in the measures (all Ps>0.369). auditorydomainraisedtheworrythatourpriorobservationswith Therefore even with a stimulus which is harder to learn and SWS may also be challenging to replicate. It could be that that offers a tighter control on learning effects, we found no evidence such stimuli are too unusual or unique in some way, hampering thatexplicitlymodulatingexpectationsforspeechleadstogreater efforts to index top-down effects on perception. The aim of our gains in discrimination for people who are hallucination prone. final experiment therefore was to replicate and extend the ‘naïve Using SVS we were more able to control learning effects—both listening’ effect observed in our study with NCVH (Alderson-Day pre- and post-exposure—and these were still considerable, even et al. 2017), but this time using SVS. In that study, participants with harder stimuli. However, notwithstanding such effects, we with frequent experience of hearing voices reported recognis- were unable to provide evidence for an auditory version of the ing speech in SWS earlier than control participants, despite not modulation hypothesis. being informed that speech was hidden in the stimulus. Those who reported ‘tuning in’ to SWS earlier also reported signifi- Experiment 3 cantly greater levels of AVH in the preceding week—as mea- sured on the ‘physical characteristics’ subscale of the PSYRATS Replicating and extending the naïve listening effect. 10 Alderson-Day et al. (Haddock et al. 1999)—but they performed no differently to con- target was used to maintain attention and provide an incidental trols on a post-exposure discrimination task, in a similar manner task which would discourage participants guessing at the pur- to the other studies reported here. pose, i.e. the potential intelligibility of distractors, hidden in the To reproduce this, we reran the naïve listening procedure SVS. As in Alderson-Day et al. (2017), the ‘scratchy’ sounds were and collected hallucination-proneness measures on the LSHS-A examples of unintelligible SWS (i.e. the frequency and amplitude (n=134) in a larger, healthy sample of individuals. Moreover, we tracks of two separate original sentences combined), that had addedanadditionalproceduretointroduceamoreobjectivemea- been further noise-vocoded, giving them a different timbre and sureofstimulusdecoding.Oneworryaboutouroriginalprocedure soundquality.Onceparticipantscoulddiscriminatethetwokinds wasthatitreliedonparticipantself-report,andthusmaybeopen of sound, they attempted the main listening task, which con- toparticipantsclaimingtheyhadheardspeechorguessingspeech tainedsixblocksof15SVSstimuli(45intelligible,45unintelligible) was present without actually decoding it. Here we introduced a and three targets per block. At the start of each block, a visual memory test that could only be successfully completed if partic- stimulus appeared announcing the start of the block (i.e. block ipants had actually been understanding words in the SVS prior 1, block 2, etc). Stimuli were presented in a predefined pseudo- to the ‘reveal’ that speech was present. This, therefore, would random order with no more than two of the same kind of stimuli extend our initial finding by providing more objective evidence consecutively. of early SVS comprehension. Using this procedure, we predicted Once the participants had listened to all six blocks, they were that people higher in auditory hallucination-proneness would (i) asked by the experimenter (i) if they noticed anything unusual reportrecognisingspeechintheSVSearlierthanothers,and(ii)be about the words, (ii) if they noticed any words and sentences abletoremembersignificantlymorewordshiddenintheSVStask. and crucially (iii) if they knew which round they started notic- Wealsoincludedmeasuresofvisualhallucination-pronenessand ing them (using the visual block markers as means of marking general schizotypal traits—to test for specificity—and a test of out time). Participants’ estimates for the third question were SVS discrimination post-exposure (for comparison with our prior used as the main task outcome, defined as their ‘recognition study). point’. They were then told that in fact there were words present in the stimuli, and asked to complete the memory task for words contained in the SVS. The memory test consisted of 46 Method words, including 18 words included in the SVS (3 per block), Participants 18 words matched for length and complexity that did not fea- We recruited 134 participants (age M(SD)= 21.45 (5.79), range ture in the SVS, five words from the NART, and five non-target 18–59years, 46 male/2 other) via university departments, social words matched to the NART words for their irregular spelling. media, and word-of-mouth. Exclusion criteria were the same as Following recognition memory methodology (Tulving 1985), par- for the previous experiments. Participants were invited to take ticipants were asked to indicate for each word whether they part in a ‘study of auditory perception’ that involved listening to explicitly remembered the word (R), felt like they knew they had some ‘unusual sounds’, but no mention of voices or speech was heard the word at some point (K), or new items that they didn’t included in the study materials. recognize (N). As we were most interested in participants actu- ally being able to decode the words, we focused on remember Materials and procedure scores. New items, in contrast, acted as lures for potential false The procedure for experiment 3 closely followed Alderson-Day positives. etal.(2017).Themeasuresusedarereportedhereintheorderthey AsinAlderson-Dayetal. (2017), participantsalsothenreceived were attempted by participants. training in understanding SVS, and then attempted an accu- racy task (testing their speech/non-speech discrimination and National Adult Reading Test (NART; Nelson 1982) accuracy for understanding keywords in 25 intelligible and 25 The NART is a measure of vocabulary and reading ability which unintelligible SVS trials). Results for these tasks are included in has been used extensively in research on psychosis as an indica- Supplementary Materials. tor of premorbid IQ (e.g. Broome et al. 2012) and was included in Alderson-Day et al. (2017) for group matching. We retained it here Questionnaires to follow that procedure, but also to provide control material for Following the tasks, participants completed the LSHS-A (as in the memory task. By including a small selection of words from Experiment 1b), but also two further measures: the four corre- the NART (plus similar words matched for unusual spelling), we sponding visual questions from the LSHS (i.e. the LSHS-V) and could control for general memory differences between those who the brief version of the Oxford Liverpool Inventory of Feelings did and did not recognise speech in the SVS. and Experiences (OLIFE; Mason and Claridge 2006; Fernyhough et al. 2008). This allowed for specificity testing by comparing audi- SVS naïve listening procedure toryhallucination-pronenesswithvisualexperiencesandgeneral Intelligible and unintelligible SVS stimuli were drawn from the proneness to schizotypal experiences. same set as Experiment 2. In Experiment 3, the 16-band, +6db stimuli were used, as these appeared to be the most similar to Results and discussion SWS in terms of their level of difficulty. Participants were told Assessing naïve listening via self-report that they would be listening out for a scratchy target sound that sounded ‘different’ from the other sounds and began the pro- No participants guessed the purpose of the experiment before cedure by listening to three examples of the target randomly testing. Overall, 83/134 participants (62%) recognized speech presented along with five examples of unintelligible SVS. The being present in the SVS and 51 (38%) did not. Participants that Specifically,participantsweretold: ‘Yourtaskistolistenoutforatargetsound the others. Have a listen to these examples, and press the space bar if you think you hear the unusual sound.’ and press the space bar every time you hear it. It might sound “scratchy” compared to Susceptibility to auditory hallucinations 11 recognized speech did so most frequently in the second block General discussion (mean=2.58, median=3, range=1–5) and of those, 72 were able Our aim in the present paper was to align and reconcile two find- to repeat some of the words encoded in the stimuli. ingsofenhancedtop-downperceptualprocessinginpeopleprone We wanted to assess how well the LSHS-A predicted which to hallucinations. Across four experiments, we were only par- participants recognised the speech without training. In a logis- tiallysuccessful inthis aim. In Experiments1a, 1b and2 wecould tic regression model, LSHS-A scores significantly predicted not provide evidence for a modulation hypothesis in the audi- recognition group (Z=2.206, P=0.027, OR=1.21, CI=1.03–1.44). tory domain. People who were more prone to hallucinations did Specificity analysis swapping LSHS-A for visual items of the not appear to draw upon prior expectation more in their percep- Launay–Slade (LSHS-V) or a measure of general schizotypy (the tion when their expectations were explicitly updated. In contrast, OLIFE) did not result in significant models (see Supplementary in Experiment 3 we provided further evidence for a naïve listen- Materials). ing effect: healthy people who were prone to hallucinations were The ability to decode ambiguous speech stimuli and sim- more able than others to identify speech in SVS without knowing ilar skills (such as extracting speech from noise) are known that speech was present. to vary considerably across individuals. To explore the naïve This pattern of results suggests a spontaneous, rather than listening effect, we ran an exploratory analysis drawing on directed, use of top-down resources in people who are prone to the NART scores collected at the start of the experiment. unusual perceptions, in contrast to prior findings using visual Along with offering a very rough proxy of verbal intelligence, paradigms (Teufel et al. 2015). How to explain this discrepancy? NART scores reflect vocabulary and reading ability—all of which First, it is important to consider the potential differences in the would plausibly contribute to the identification of ambiguous kind of information provided by each stimulus. As already noted, speech. We therefore reran our logistic regression analyses, SWS/SVS appeared to allow for learning across trials, even in a but including NART as an additional predictor. Higher NART naïve state. This represents an opportunity for perceptual learn- scores were associated with identifying speech spontaneously ing(SohogluandDavis2016)thatwouldnotappeartobeapparent (Z=3.78, P<0.001, OR=1.15, CI=1.08–1.25) but the contribu- when learning to discriminate Mooney images. Moreover, it raises tionofLSHS-Ascoreswasnownon-significant( Z=1.80,P=0.072, the possibility that some participants may learn to discriminate OR=1.17, CI=0.99–1.40), despite similar odds ratios for the two speech from non-speech without understanding the content of predictors. the speech (doing so possibly via prosodic or structural cues). As such, discriminatingspeechfromnon-speechinSWS/SVSmaybe Assessing naïve listening via memory posing a different kind of challenge to Mooney images, and may performance not be solely reliant on top-down knowledge. Following the reveal that speech was present participants were This might raise the concern that SWS/SVS stimuli are testedontheirmemoryforhiddenwords,toprovideamoreobjec- just too different from degraded images to explore top-down tive test of spontaneous SVS decoding. Across the whole sample effects in a comparable way. However, when discrimination was (i.e. combining those who did and did not recognize speech spon- made especially challenging and learning effects minimized (in taneously), accuracy for ‘remember’ items on the memory task experiment 2), we still observed no evident relationship between significantly correlated with LSHS-A scores, rs=0.20, P=0.020 hallucination-pronenessandtemplateexposure. Moreover, inour (Spearman’s test). final experiment, people who were more AVH-prone were specifi- This relation could feasibly have been driven by a general ten- cally more able to remember hidden words in the SVS, suggesting dency to endorse words on the memory test (i.e. a false positive they could successfully decode the stimuli (rather than discrimi- bias),orgeneraldifferencesinmemoryscores.Ifso,LSHS-Ascores nating speech from non-speech in a more general way). If the dif- would have also correlated with (i) endorsement of new (i.e. lure) ference between auditory and visual stimuli of this kind is simply items on the memory task, or (ii) recall for NART items included oneofdifficulty, orthenumberofgeneralizablecuesacrosstrials, in the memory task. However, neither new item scores, r=-0.00, thenthiscanbeexploredempiricallyinfuturestudies.Itwouldbe P=0.969,norformemoryforNARTwords,r=0.05,P=0.551,were possibletoparametricallyvarytheperceptibilityofhiddenspeech associated with LSHS-A scores. This suggested that the ability to inSVSbyadaptingthesignal-to-noiseratioandreducingthenum- understand the SVS early, and thus encode more hidden words, ber of vocoding channels. This would dampen such effects and was indeed related to greater hallucination-proneness. could demonstrate modulatory effects that would be comparable Given the apparent role of NART scores in predicting self- to those seen in the visual domain. reported recognition of SVS, we used partial correlation tests to If the lack of any modulation effect is genuine, however, it assess what role it played in the relationship between LSHS-A might suggest that healthy people prone to unusual experiences scoresand‘remember’scoresforwordshiddenintheSVS.LSHS-A are not necessarily susceptible to momentary effects—such as scores positively correlated with words remembered when con- suggestibility, or demand characteristics—and instead possess trolling for the NART, r=0.18, P=0.035), but no significant rela- perceptual biases that are somehow more ingrained. In predic- tion between memory performance and NART scores was evident tive processing terms, this could constitute a higher-order belief when controlling for LSHS-A, r=0.13, P=0.136. about the world (e.g. ‘the universe is full of hidden meanings’) The results of Experiment 4 therefore supported the original which directs the individual response in ambiguous situations to findings of Alderson-Day et al. (2017), but added to it by provid- explorepotentiallyimportantsignalsdespitetheinstructiontolis- ing a more objective test of participants being able to recognize ten for the target sound. Exploring differences in attention during speech early—namely, actual improved recall for words hidden naïvelisteningwillbevital, assomeparticipantsattitudetowards in the SVS. Moreover, this was specific to auditory hallucination- unusual and ambiguous stimuli may lead them to allocate more proneness and not visual hallucination-proneness, or general attentiontoSVS,thusgivingthemagreateropportunitytoengage schizotypy. in explicit or implicit processes of perceptual learning. Recent 12 Alderson-Day et al. research on primal world beliefs represents a promising avenue butclosercontrolofthesefactorswouldhavebeentotesthearing for this line of research (Clifton et al. 2019). skills in all participants. Alternatively, hallucinations may relate to a very different An open question for future research is how people with fre- level in the processing hierarchy and be expressed in how audi- quent and distressing hallucinations perform under naïve listen- tory objects are represented. A recent reframing of the predictive ingconditionsandhowaperceptualadvantageinsomeconditions approach by Teufel and Fletcher (2020) has proposed to sepa- translates into non-veridical experiences in other contexts. A ratelong-term, contextinvariant‘constraints’onperceptionfrom recent small-scale study by Kafadar and colleagues (2020)—using context-specific, temporary ‘expectations’ that shape immedi- the structure of Experiment 1a but the same stimuli as Exper- ate, moment-to-moment sensation. Our data point towards a iments 2 and 3—found that people at clinical high risk of psy- long-term constraint for some individuals in how they recog- chosisaremarkedmorebytheirpre-exposurebiastohearspeech, nise auditory objects when faced with ambiguous speech-like rather than discrimination. They also found no differences in stimuli, rather than sensitivity to changes in situation-specific post-exposure discrimination, suggesting again that the explicit expectations. Support for this argument comes from recent work modulation of expectation was less relevant to understanding on the templates that people use to make judgements about how they perceived SVS. It may be that healthy individuals who speech. By creating ‘speechiness’ kernels for individual partic- have unusual experiences have very slight biases in their percep- ipants, Erb and colleagues (2020) have demonstrated that peo- tion that facilitate a top-down advantage, akin to the concept of ple high in hallucination-proneness utilize qualitatively differ- encoding style for internal over external meanings (Valerie et al. ent speech templates when discriminating speech sounds, such 2011). Conversely, individuals with clinical hallucinations may be that lower frequencies typical of speech are attended to less, more fixed in their expectations of finding signal in noise, leading compared to higher frequency auditory information. If so, this tonon-veridicalexperiencesmoregenerally. Insuchasituation, a suggests a fundamental long-term alteration to how speech is slightbiasmightyieldanadvantage—especiallyifitcanbeapplied recognisedandprocessedinthosepronetohallucinatoryandillu- selectively—whereas a strong bias would lead to false positives sory experiences—rather than a dynamic, moment-to-moment (i.e. hearing speech in unintelligible SWS/SVS). volatility in how expectations are managed. Variation of naïve lis- In conclusion, the experiments that we present here refine tening effects with SVS tailored to different speech kernels could our understanding of how top-down expectations shape speech perception for people prone to auditory hallucinations. Directly be used to explore this further. updatingexpectationwouldnotappeartoconferanadvantagefor Finally, such effects are likely to be shaped by a considerable peoplehigherinhallucination-proneness—atleastforthesekinds number of individual differences, including auditory experience of stimuli—but being susceptible to hallucinations would appear and cognitive factors such as verbal IQ (Gwilliams and Wallisch to be associated with responding to ambiguous speech stimuli in 2020). We explored the role of NART scores in our final exper- a differential way, facilitating speech identification. This impli- iment for this reason. The observed relationship between those cates longer term and potentially lower-level constraints on how scoresandspontaneousrecognitionsuggestsitmayhaveasignif- speechisrecognizedinsuchpopulations,ratherthanatemporary icant impact on SVS discrimination, with implications for future sensitivity to expectation. studies. Importantly, the same relationship was not evident for performance on the memory task—suggesting that verbal IQ dif- ferences could not fully account for the relationship between SVS Supplementary data decoding and hallucination-proneness. Nevertheless, the range Supplementary data is available at NCONSC online. of variables guiding individual comprehension skills for degraded speech, noise-vocoded speech and speech-in-noise are vast, com- Data availability plex and yet to be clearly pinpointed (McGettigan et al. 2012). Somelimitationsofourgeneralapproachmustbenoted. First, Data and analysis code for each of the experiments are available we tested exclusively university and general population-based via HYPERLINK https://osf.io/yrn9j/ OSF. samples, rather than either clinical or NCVH (i.e. those with very regular hallucination-like experiences; Johns et al. 2014). Acknowledgements We therefore cannot rule out that participants with more fre- We thank Marissa Thomas, Kamal Duran, Charlotte Newman, quent experiences would not show specific modulation effects Paris Brown, Uzma Farooqui, Ellie Maycock and Summer Zhu on their perception. However, predictive approaches to hallu- for helping to collect data for this project. This research was cinations are often explicitly framed as models of both clin- funded in whole, or in part, by the Wellcome Trust [Grant num- ical and non-clinical phenomena, with the former resulting ber WT108720]. For the purpose of open access, the author has from an accentuation of mechanisms underlying normal, veridi- applied a CC BY public copyright licence to any Author Accepted cal perception rather than being specific to clinical disorders Manuscript version arising from this submission. (Corlett et al. 2019). Our data therefore would appear relevant to understanding perceptual mechanisms across the psychosis continuum. Funding Second, none of our experiments deployed basic tests of audi- This research was funded in whole, or in part, by the Wellcome tory processing or hearing ability, which could contribute to low- Trust [Grant number WT108720]. level differences in how SWS/SVS are recognized and processed. Thereisemergingevidencethatsubtledifferencesinhearingabil- Conflict of interest statement itycanbeassociatedwithhallucinationsforsomepeople(Linszen et al. 2016). Participants with a self-declared hearing difficulty None declared. were not recruited to the study, and tests of speech intelligibil- ity do not pose the same challenges to hearing as other tasks usedinhallucinationresearch(suchasauditorysignaldetection), Kristiina Kompus (U. Bergen) is thanked for this comparison. Susceptibility to auditory hallucinations 13 Greenwood DD. A cochlear frequency-position function for several References species—29 years later. J Acoust Soc Am 1990;87:2592–605. Alderson-DayB,LimaCF,EvansSetal.Distinctprocessingofambigu- Griffin JD, Fletcher PC. Predictive processing, source monitoring, and ousspeechin peoplewith non-clinicalauditoryverbalhallucina- psychosis. Annu Rev Clin Psychol 2017;13:265–89. tions. Brain 2017;140:2475–89. GwilliamsL,WallischP.Immediateambiguityresolutioninspeechpercep- Alderson-Day B, Smailes D, Moffatt J et al. Intentional inhibition tion based on prior acoustic experience. 2020. 10.31234/osf.io/yptfr. but not source memory is related to hallucination-proneness HaddockG,McCarronJ,TarrierNetal.Scalestomeasuredimensions and intrusive thoughts in a university sample. Cortex 2019;113: of hallucinations and delusions: the psychotic symptom rating 267–78. scales (PSYRATS). Psychol Med 1999;29:879–89. BauerSM,SchandaH,KarakulaHetal.Cultureandtheprevalenceof Hohwy J. The Predictive Mind. Oxford: Oxford University Press, 2014. hallucinationsinschizophrenia.ComprPsychiatry2011;52:319–25. Jardri R, Denève S. Circular inferences in schizophrenia. Brain Bell V, Halligan PW, Ellis HD. The Cardiff Anomalous Perceptions 2013;136:3227–42. Scale (CAPS): a new validated measure of anomalous perceptual Johns LC, Kompus K, Connell M et al. Auditory verbal hallucina- experience. Schizophr Bull 2006;32:366–77. tions in persons with and without a need for care. Schizophr Bull Bench J, Kowal Å, Bamford J. The BKB (Bamford-Kowal-Bench) 2014;40:S255–S64. sentence lists for partially-hearing children. Br J Audiol 1979; Kafadar E, Mittal VA, Strauss GP et al. Modeling perception and 13:108–12. behaviorinindividualsatclinicalhighriskforpsychosis: support BentallRP.Theillusionofreality:areviewandintegrationofpsycho- for the predictive processing framework. Schizophr Res 2020;226: logical research on hallucinations. Psychol Bull 1990;107:82–95. 167–75. Bentall RP, Slade PD. Reality testing and auditory hallucinations: a Kim S. 2015. ppcor: Partial and Semi-Partial (Part) Correlation. signal detection analysis. Br J Clinial Psychol 1985a;24:159–69. https://CRAN.R-project.org/package=ppcor (1 July 2020, date last Bentall RP, Slade PD. Reliability of a scale measuring dispo- accessed). sition towards hallucination: a brief report. Pers Individ Dif 1985b;6:527–9. Linszen MMJ, Brouwer RM, Heringa SM et al. Increased risk of psy- Brookwell ML, Bentall RP, Varese F. Externalizing biases and hal- chosis in patients with hearing impairment: review and meta- lucinations in source-monitoring, self-monitoring and signal analyses. Neurosci Biobehav Rev 2016;62:1–20. detection studies: a meta-analytic review. Psychol Med 2013;43: Macmillan NA, Kaplan HL. Detection theory analysis of group data: 2465–75. estimating sensitivity from average hit and false-alarm rates. Broome MR, Day F, Valli I et al. Delusional ideation, manic symp- Psychol Bull 1985;98:185–99. tomatology and working memory in a cohort at clinical high-risk Maijer K, Hayward M, Fernyhough C et al. Hallucinations in children for psychosis: a longitudinal study. Eur Psychiatry 2012;27:258–63. and adolescents: an updated review and practical recommenda- Clark A. Whatever next? Predictive brains, situated agents, and the tions for clinicians. Schizophr Bull 2019;45:S5–S23. future of cognitive science. Behav Brain Sci 2013;36:181–204. Mason O, Claridge G. The Oxford-Liverpool Inventory of Feelings and CliftonJDW,BakerJD,ParkCLetal.Primalworldbeliefs.PsycholAssess Experiences (O-LIFE): further description and extended norms. 2019;31:82–99. Schizophr Res 2006;82:203–11. Corlett PR, Horga G, Fletcher PC et al. Hallucinations and strong Mason O, Claridge G, Jackson M. New scales for the assessment of priors. Trends Cogn Sci 2019;23:114–27. schizotypy. Pers Individ Dif 1995;18:7–13. Erb J, Kreitewolf J, Pinheiro AP et al. Aberrant perceptual judgements McCarthy-Jones S, Fernyhough C. The varieties of inner speech: onspeech-relevantacousticfeaturesinhallucination-proneindi- links between quality of inner speech and psychopathological viduals. Schizophr Bull Open 2020;1:sgaa059. variables in a sample of young adults. Conscious Cogn 2011;20: ´ ´ ´ ´ EsquirolJ.Surlesillusionsdessenschezlesalienes.ArchivesGenerales 1586–93. de Medecine 1832;2:23. McGettiganC,EvansS,RosenSetal.Anapplicationofunivariateand Feinberg I. Efference copy and corollary discharge: implications for multivariate approaches in fMRI to quantifying the hemispheric thinking and its disorders. Schizophr Bull 1978;4:636–40. lateralization of acoustic and linguistic processes. J Cogn Neurosci Fernyhough C, Bland K, Meins E et al. Imaginary companions and 2012;24:636–52. youngchildren’sresponsesto ambiguousauditorystimuli: impli- Mintz S, Alpert M. Imagery vividness, reality testing, and cations for typical and atypical development. J Child Psychol Psy- schizophrenic hallucinations. J Abnorm Psychol 1972;79:310–6. chiatry 2007;48:1094–101. Mooney CM. Age in the development of closure ability in chil- Fernyhough C, Jones SR, Whittle C et al. Theory of mind, schizotypy, dren. Can J Psychol/Revue Canadienne de Psychologie 1957;11: and persecutory ideation in young adults. Cogn Neuropsychiatry 219–26. 2008;13:233–49. Morrison AP, Haddock G, Tarrier N. Intrusive thoughts and audi- Fernyhough C, Watson A, Bernini M et al. Imaginary companions, tory hallucinations: a cognitive approach. Behav Cogn Psychother inner speech, and auditory verbal hallucinations: what are the 1995;23:265–80. relations? Front Psychol 2019;10:1665. Morrison AP, Wells A, Nothard S. Cognitive factors in predispo- Fletcher PC, Frith C. Perceiving is believing: a Bayesian approach sition to auditory and visual hallucinations. Br J Clin Psychol to explaining the positive symptoms of schizophrenia. Nat Rev 2000;39:67–78. Neurosci 2009;10:48–58. Nelson HE. National Adult Reading Test. Test Manual. Windsor: NFER- Frith C. The Cognitive Neuropsychology of Schizophrenia. New York: Nelson, 1982. Psychology Press, 1992. Pearson D, Burrow A, FitzGerald C et al. Auditory hallucinations in Garrison JR, Moseley P, Alderson-Day B et al. Testing continuum normal child populations. Pers Individ Dif 2001;31:401–7. models of psychosis: no reduction in source monitoring ability Peters E, Joseph S, Day S et al. Measuring delusional ideation: the in healthy individuals prone to auditory hallucinations. Cortex 21-item Peters et al. Delusions Inventory (PDI). Schizophr Bull 2017;91:197–207. 2004;30:1005–22. 14 Alderson-Day et al. Peters E, Joseph SA, Garety PA. Measurement of delusional ideation Stanghellini G, Langer AI, Ambrosini A et al. Quality of halluci- in the normal population: introducing the PDI (Peters et al. Delu- natory experiences: differences between a clinical and a non- sions Inventory). Schizophr Bull 1999;25:553–76. clinical sample. World Psychiatr: Off J World Psychiatr Assoc (WPA) Peters E, Ward T, Jackson M et al. Clinical, socio-demographic 2012;11:110–3. and psychological characteristics in individuals with persistent Stanislaw H, Todorov N. Calculation of signal detection theory mea- psychotic experiences with and without a “need for care”. World sures. Behav Res Methods Instrum Comput 1999;31:137–49. Psychiatry 2016;15:41–52. Taylor M. Imaginary Companions and the Children Who Create Them. Pickering MJ, Clark A. Getting ahead: forward models and their place Oxford: Oxford University Press, 1999. in cognitive architecture. Trends Cogn Sci 2014;18:451–6. Teufel C, Fletcher PC. Forms of prediction in the nervous system. Nat Powers AR, Kelley M, Corlett PR. Hallucinations as top-down Rev Neurosci 2020;21:231–42. effects on perception. Biol Psychiatr: Cognit Neurosci Neuroimaging Teufel C, Subramaniam N, Dobler V et al. Shift toward prior 2016;1:393–400. knowledge confers a perceptual advantage in early psychosis Remez RE, Rubin PE, Pisoni DB et al. Speech perception without and psychosis-prone healthy individuals. Proc Nat Acad Sci traditional speech cues. Science 1981;212:947–9. 2015;112:13401–6. Rosen S, Wise RJS, Chadha S et al. Hemispheric asymmetries in Tulving E. Memory and consciousness. Can Psychol/Psychologie Cana- speech perception: sense, nonsense and modulations. PLoS One dienne 1985;26:1–12. 2011;6:e24672. ValerieRV,BelayachiS,VanderLindenM.Internalencodingstyleand Seal ML, Aleman A, McGuire PK. Compelling imagery, unanticipated schizotypy in a sub-clinical sample. Eur Psychiatry 2011;26:1519. speech and deceptive memory: neurocognitive models of audi- Waters F, Badcock JC, Maybery MT et al. Inhibition in schizophre- tory verbal hallucinations in schizophrenia. Cogn Neuropsychiatry nia: association with auditory hallucinations. Schizophr Res 2004;9:43–72. 2003;62:275–80. Senn S. Change from baseline and analysis of covariance revisited. Waters F, Collerton D, Ffytche DH et al. Visual hallucinations Stat Med 2006;25:4334–44. in the psychosis spectrum and comparative information from SohogluE,DavisMH.Perceptuallearningofdegradedspeechbymin- neurodegenerative disorders and eye disease. Schizophr Bull imizing prediction error. Proc Nat Acad Sci 2016;113:201523266. 2014;40:S233–S45. Souza P, Rosen S. Effects of envelope bandwidth on the intel- Wilkinson S. Accounting for the phenomenology and varieties of ligibility of sine- and noise-vocoded speech. J Acoust Soc Am auditory verbal hallucination within a predictive processing 2009;126:792–805. framework. Conscious Cogn 2014;30:142–55. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Neuroscience of Consciousness Oxford University Press

Susceptibility to auditory hallucinations is associated with spontaneous but not directed modulation of top-down expectations for speech

Download PDF
14 pages

Loading next page...
 
/lp/oxford-university-press/susceptibility-to-auditory-hallucinations-is-associated-with-nMbPwD0xfX

References (69)

Publisher
Oxford University Press
Copyright
© The Author(s) 2022. Published by Oxford University Press.
eISSN
2057-2107
DOI
10.1093/nc/niac002
Publisher site
See Article on Publisher Site

Abstract

Auditoryverbalhallucinations(AVHs)—orhearingvoices—occurinclinicalandnon-clinicalpopulations,buttheirmechanismsremain unclear.Predictiveprocessingmodelsofpsychosishaveproposedthathallucinationsarisefromanover-weightingofpriorexpectations inperception.Itisunknown,however,whetherthisreflects(i)asensitivitytoexplicitmodulationofpriorknowledgeor(ii)apre-existing tendency to spontaneously use such knowledge in ambiguous contexts. Four experiments were conducted to examine this question in healthy participants listening to ambiguous speech stimuli. In experiments 1a (n=60) and 1b (n=60), participants discriminated intelligible and unintelligible sine-wave speech before and after exposure to the original language templates (i.e. a modulation of expectation). No relationship was observed between top-down modulation and two common measures of hallucination-proneness. Experiment 2 (n=99) confirmed this pattern with a different stimulus—sine-vocoded speech (SVS)—that was designed to minimize ceiling effects in discrimination and more closely model previous top-down effects reported in psychosis. In Experiment 3 (n=134), participants were exposed to SVS without prior knowledge that it contained speech (i.e. naïve listening). AVH-proneness significantly predicted both pre-exposure identification of speech and successful recall for words hidden in SVS, indicating that participants could actually decode the hidden signal spontaneously. Altogether, these findings support a pre-existing tendency to spontaneously draw upon prior knowledge in healthy people prone to AVH, rather than a sensitivity to temporary modulations of expectation. We propose amodelofclinicalandnon-clinicalhallucinations, acrossauditoryandvisualmodalities, withtestablepredictionsforfutureresearch. Keywords: consciousness; ketamine anesthesia; EEG markers of consciousness; perturbational complexity index Introduction 1992). Although not always framed as a ‘top-down’ model of hallucinatory experience, this grounded much research in the Hallucinations have long been considered a product of top-down metacognitive domain, consistent with cognitive approaches to processes: what the mind brings to our perception of the world, psychosis in clinical practice (Morrison et al. 1995). not the other way round (Esquirol 1832). Auditory verbal hal- lucinations (AVHs) in particular have been studied extensively Recent interest in predictive processing approaches has because of their association with schizophrenia, occurring in reframedtheputativeroleoftop-downprocessesinhallucination. 60–90% of cases (Bauer et al. 2011) and at rates that are often Underthepredictiveprocessingframework(PPF),allofperception double those seen for other modalities (Waters et al. 2014). AVHs and cognition is the result of a trade-off between generative mod- have been proposed to result from various internal sources such elsoftheworld,shapedbypriorexpectationsandpredictionerror, as memories, imagery, and self-talk or inner speech (Mintz and i.e. the gap between expectation and sensory input (Clark 2013; Alpert 1972; Waters et al. 2003; Seal et al. 2004). Difficulties in dis- Hohwy 2014). Hallucinations have been posited as an imbalance tinguishing the internal from external were interpreted as a prob- betweenpriorexpectationandpredictionerror(FletcherandFrith lemwith‘realitymonitoring’,inwhichdisruptionstosourcemon- 2009; Jardri and Denève 2013; Powers et al. 2016). Such accounts itoring could explain how self-generated cognitive states could have been argued to be consistent with source-monitoring theo- becomeperceptualexperiences(Feinberg1978;Bentall1990;Frith ries (Wilkinson 2014; Griffin and Fletcher 2017 ; Corlett et al. 2019) Received: 1 October 2021; Accepted: 13 January 2022 © The Author(s) 2022. Published by Oxford University Press. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. 2 Alderson-Day et al. and may even reflect a generalization of prediction mechanisms we followed a similar test-train-test procedure to Teufel and inherent in earlier theories (Pickering and Clark 2014). Neverthe- colleagues, using the CAPS (Bell et al. 2006) and PDI (Peters et al. less,theyinvolveashiftinemphasisawayfromthemetacognitive 2004) to measure unusual perceptual experiences and delusional monitoring of self, focusing instead on expectation and learning beliefs.Experiment1b(n=60),runinparallel,usedanalternative as being central to hallucination. measure specific to AVH-proneness: a version of the Launay– Supporting evidence for a PPF approach to hallucinations was Slade Hallucination Scale-Revised (Bentall and Slade 1985b; Mor- providedbyTeufeletal.(2015),inastudyofindividualswithanat- rison et al. 2000). It also included participants intentionally riskmentalstateforpsychosis.Patientsandhealthycontrolswere recruited to expand the potential range of individual differences asked to discriminate monochrome Mooney (1957) images, before in hallucination-proneness (specifically, people with a history of andafterexposuretotheiroriginaltemplates(picturesofhumans imaginary companions; Fernyhough et al. 2019), and an added and animals). While both groups improved their discrimination condition that sought to further prime potential templates for after viewing the templates, clinical participants showed signifi- speech. Based on the modulation hypothesis, we hypothesized cantly enhanced discrimination compared to controls, consistent thatimprovementsindiscriminationfollowingtemplateexposure with top-down information being given greater weight in their should be associated with higher hallucination-proneness. perceptual processing. Teufel and colleagues then replicated this One problem with SWS is that some participants attempting finding in a sample of 40 healthy participants rated for psychosis- to understand it may go from not understanding it all before pronenessonmeasuresofhallucination-likeexperiences(r=0.42, training to suddenly understanding it all, while others may spon- theCardiffAnomalousPerceptionsScale;Belletal.2006)anddelu- taneously learn to decode it. This learning profile can make it sional traits (r=0.33, the Peters Delusion Inventory; Peters et al. hard to compare with perceptual learning for Mooney images. So, 2004), with higher scores on these scales being associated with a in Experiment 2 (n=99), we tested the same modulation effect greater improvement in discrimination following exposure to the but with a new stimulus, sine-vocoded speech (SVS). We devel- templates (Teufel et al. 2015). opedthisparticularstimulussetwiththeaimofofferingatighter These findings speak to visual processes—but what of voices, control on some of the potential learning effects inherent to the most common kind of hallucination in psychosis? Various SWS comprehension—making it more comparable to Teufel et al. source-monitoring studies have demonstrated biases in auditory (2015). Asinexperiments1aand1b, weexpectedthatmodulation signal detection in people with hallucinations—often on white of discrimination would be related to hallucination-proneness. noise tasks (Bentall and Slade 1985a)—but facilitatory effects like Finally, having tested modulation effects using SVS, we those described by Teufel and colleagues have not typically been returned to the behavioural design from Alderson-Day et al. studied. (2017),examiningSVSperceptionundernaïvelisteningconditions Analogous to Mooney images, sine-wave speech (SWS; Remez (Experiment 3, n=134). According to the naïve listening hypothe- et al. 1981; Rosen et al. 2011) is a perceptually ambiguous stimu- sis, we predicted that hallucination-proneness would be higher in lus derived from speech that allows for exploration of top-down those who were quicker to recognize that SVS contained hidden effects on perception. SWS is not usually identified as intelligible speech. After the naïve listening procedure, we also tested them speech by naïve listeners; instead, it requires prior training to be ontheirmemoryforthehiddenwords,thereforeprovidingamore recognized and understood. In a recent study, a sample of non- objectivetestofspontaneousdecodingofthehiddenspeech.Data clinicalvoice-hearers(NCVH)—individualswithfrequentAVHbut and analysis code for each of the experiments are available via noneedforclinicalcare(Johnsetal. 2014; Petersetal. 2016)—were OSF. scanned in fMRI while naïvely listening to SWS (Alderson-Day and Lima et al. 2017). Instead of being told to listen for speech, Experiment 1a participants were instructed to listen for an unintelligible target Modulating prior knowledge of sine-wave speech. sound amidst a range of SWS stimuli. Despite this, a majority Theaimofourfirstexperimentwastodevelopamodulationof of the NCVH group identified speech in the SWS spontaneously expectationintheauditorymodalityandtoseehowthisrelatedto and without any training. When asked to estimate the point at hallucination-proneness scores. In contrast to Teufel et al. (2015), which they recognized the hidden speech (visual markers had who used 12 blocks of before/after trials, we chose to play all been displayed indicating numbered ‘rounds’ during the scan), 90 trials, train on the whole set, and then retest for all trials NCVH participants reported doing so significantly earlier than a (a ‘one-shot’ procedure). This was chosen to minimize any poten- matched control group. Subsequent tests of discrimination fol- tialtrainingeffectsoccurringacrossmultipleblocksoftestingand lowing the ‘reveal’ that speech was present, failed to identify training. We predicted that higher CAPS scores would be associ- any group differences. This suggested that voice-hearers may ated with greater increases in discrimination following template automatically draw upon top-down resources—such as speech exposure. We also explored this effect for delusion-proneness templates—whenfacedwithambiguoussensoryinput(Alderson- scores on the PDI. Day et al. 2017). Both experiments are consistent with top-down processing Method being linked to hallucinations, but they highlight contrasting Participants effects: a modulatory effect (Teufel et al. 2015) and a naïve listen- A convenience sample of 60 participants was recruited from a ingeffectonperception(Alderson-Dayetal. 2017). Theyalsodiffer university cohort (age M(SD)=21.22 (3.11), range 18–32years, 18 in design and stimuli, making it challenging to directly compare male). Individuals were invited to take part if they were native them. In this paper, we aimed to draw together these effects, English speakers with no hearing impairments or any previous adapting the SWS procedure across a series of experiments with psychiatricorneurologicaldiagnoses.Participantsreceivedcourse healthy participants to explore top-down effects on audition. We began with the original SWS stimuli used in Alderson-Day et al. While a convenience sample, this number was nevertheless sufficiently (2017) deployed in two parallel experiments examining modula- powered to identify an effect in a similar range to Teufel and colleagues’ tion effects (Experiments 1a and 1b). In Experiment 1a (n=60), observed effect size for the CAPS (r=0.42, 90% power, min. sample=50). Susceptibility to auditory hallucinations 3 credit or a gift voucher in recognition of their time. For this The Peters Delusion Inventory—21 item version (PDI; Peters et al. and the remaining experiments, written informed consent was 2004) is a shortened adaptation of the original 40-item PDI (Peters obtained for all participants and all procedures were approved by et al. 1999). Both measures have been used extensively as a mea- a university ethics committee. sure of proneness to unusual beliefs in the general population, have good convergent validity with other measures of schizo- typy, and have strong internal reliability (e.g. alpha>0.8). The Materials and procedure PDI has an identical structure to the CAPS (the latter being mod- SWS discrimination task elledontheformer).Frequencyofbeliefwasincludedasthemain SWS is created by tracking and modeling the formant tracks of outcome. spoken sentences using a sine-wave tone. This procedure can Our analytic approach sought to first assess changes in dis- be used to create potentially intelligible stimuli (in which the crimination and bias variables following the exposure phase, frequency and amplitude tracks of the same original sentence using paired t-tests. We then (i) followed Teufel and colleagues’ arecombined)—orunintelligiblestimuli(combiningthefrequency analysis by testing the relationship between CAPS scores and and amplitude information of two different sentences). Both are d-prime improvement using correlational analysis, and (ii) used typically perceived as unintelligible, but potentially intelligible partial correlation to test this association while controlling for SWS typically becomes comprehensible following training and confounds such as baseline performance. Correlations with exposure to 2–3 template sentences (Rosen et al. 2011). Experi- other change scores (i.e. beta and C), relations to pre-exposure ment1usedthesameSWSstimuliasinAlderson-Dayetal.(2017), 2 performance and associations with the PDI were included for which were first developed by Rosen et al. (2011). The original exploratory purposes. sentences were taken from the Bamford-Kowal-Bench (BKB) sen- tence set (Bench et al. 1979). Participants completed the task in a quiet university room. The task was presented using Psych- Results and discussion toolbox in MATLAB 2016 on a Windows PC with a 17'' monitor, Table 1 shows signal detection outcomes for the SWS discrim- using Sennheiser headphones for stimulus delivery. See Fig. 1 ination task. As would be expected, performance significantly for a summary of the design of this and the other reported improved following exposure to the original (i.e. non-masked) experiments. sentences, as indicated by an increase in d . However, bias also The SWS discrimination task was divided into two runs of 90 significantly increased, with participants being more likely to trials (45 intelligible SWS, 45 unintelligible SWS) occurring before say that speech was present after template exposure [before hits and after participants heard each of the original sentences that M(SD)%=67.9% (21.1%), false alarms M(SD)=20.6% (12.3%); after theintelligibleSWStrialswerebasedon(‘templateexposure’).On hits M(SD)%=88.6% (14.2%), false alarms=26% (14.9%)]. eachrun,participantslistenedto2.5sclipsofSWSandwereasked to decide whether the speech was present or not for each trial, Testing a modulation effect via bivariate allowing for signal detection measures to be calculated based on correlation hitrates(intelligibletrialsmarkedascontainingspeech)andfalse Following Teufel et al. (2015), we first tested for bivariate correla- alarm rates (unintelligible trials marked as containing speech). tionsbetweenchangeind ,CAPS,andPDIscores.Despitetheclear Signal detection theory (Stanislaw and Todorov 1999) was used to change in discrimination following exposure, no correlation was calculate discrimination (d ), plus two measures of bias: criterion observed between CAPS scores and change in d (Pearson’s prod- (C) and beta (β), the measure most typically used in source mon- uct, r (58)=0.02, P=0.864, 95% CI=−0.23:0.28), contrary to the itoring research on hallucinations (Brookwell et al. 2013). Where modulation hypothesis (see Fig. 2). A one-sided Bayesian analysis hitratesandfalsealarmswere0and1,theMacmillanandKaplan (usingJASPv.0.8.6withdefaultpriors)indicatedaBFof0.19forthe (1985) method was used (i.e. zero scores replaced with 0.5/n and 1 experimental hypothesis and 5.39 for the null (i.e. good evidence replaced with (n−0.5)/n). foralackofanyeffectofinterest).Asimilarresultwasevidentfor PDI scores (r (58)=0.11, P=0.400). Questionnaires In each experiment, questionnaires were collected after task- Controlling for baseline performance and other based measures were taken. confounds The Cardiff Anomalous Perceptions Scale (CAPS; Bell et al. 2006) is Analysisofdifferencescorescanbeaffectedbythevaluesatbase- a commonly used scale of hallucination-proneness that assesses line (Senn 2006) and are often thought to be less reliable than a range of unusual perceptual experiences—including auditory, the measures they derive from. To account for this, we also ran visual and gustatory phenomena—across 32 items. It correlates partial correlation tests controlling for baseline discrimination with other measures of schizotypy and hallucinations, such scores, using the ppcor package in R (Kim 2015). This adjustment as the Oxford Liverpool Inventory of Feelings and Experiences made no difference to the results (r=−0.04, P=0.758) suggesting (OLIFE; Mason et al. 1995), and has strong internal reliability that the overall null result was unlikely to be driven by baseline (alpha=0.87). Participants are asked to indicate whether they performance differences. haveeverhadaspecificexperience,andifso,howdistressing,how intrusive, and how frequent the experience was (on a 1–5 scale). Relations to bias To assess the general tendency to experience hallucinations, here Finally, we also tested for any relations between bias (C and we used the total frequency as the main CAPS outcome. β) and proneness to psychotic experiences. No significant cor- relations were observed between task and questionnaire scores (all P>0.10; see Supplementary Materials). Rosen and colleagues’ SWS stimuli were also noise-vocoded. This step wasomittedinouruseofthestimuliforthepresentpaperandinAlderson-Day The results of Experiment 1a, therefore, did not support etal. (2017), asnoise-vocodingcaninduceaneffectakintowhispering, andcan the idea of a modulatory effect of expectation being related to maketheunderlyingsinewavescohere—bothofwhichcouldpotentiallyreveal the underlying speech signal. hallucination-proneness. When exposed to new information via 4 Alderson-Day et al. Figure 1. Overview of experiments. (A) Experiment 1a: Participants heard 90 trials comprised of potentially intelligible and unintelligible sounds and judged whether each sound contained speech (pre-exposure). They were then exposed to the target clear speech exemplars from which the intelligible trials were made (exposure) and then asked again to judge which trials contained speech (post-exposure). (B) Experiment 1b: Participants took part in the same paradigm as Experiment 1a but half the participants were primed by listening to a busy auditory scene and the other half were not. (C) Experiment 2: Participants heard blocks of 10 trials using the same pre-exposure, exposure, post-exposure cycles in Study 1. (D) Experiment 3: Participants took part in a naïve listening experiment in which they were tasked with identifying sounds with a specific acoustic quality (noise-vocoded sounds). They were not informed that some sounds contained speech. They were then asked whether they had heard any speech in the naïve listening task and took part in a memory recognition test to see if they remembered the intelligible trials. They were then exposed to the clear speech targets and tested on their identification of speech Susceptibility to auditory hallucinations 5 Table 1. Signal detection outcomes for the SWS discrimination task in experiment 1a Before After M SD M SD Change statistic T(Z) P d(r) d 1.47 0.77 2.17 0.81 −8.63 4.809e−12 −0.90 C 0.19 0.45 −0.37 0.44 −6.25 4.212e−10 0.81 β 1.63 1.87 0.60 0.51 −5.74 9.202e−09 0.74 Note: Higher values of d indicate increased sensitivity to detect speech. Scores below 0 for C and 1 for beta indicate a bias to indicate speech is present. Wilcoxon tests were used for C and beta due to non-parametric data. Experiment 1b, run in parallel to the first, was designed to address potential concerns about the level and specificity of hallucination-proneness. First, as an alternative to the CAPS, we used a revised version of the Launay–Slade Hallucination Scale (Morrison et al. 2000), with a specific focus on auditory experi- ences (McCarthy-Jones and Fernyhough 2011). Second, intending to gather a wider range of unusual experiences, we explicitly set out to recruit individuals with a history of having imagi- nary companions (ICs). Engaging with imaginary companions has been proposed to bear commonalities with hallucinatory experi- ences (Pearson et al. 2001), even though there is no good evidence that they are a developmental marker for later psychopathology (Taylor 1999; Maijer et al. 2019). Specifically, there is evidence to Figure 2. Comparing discrimination pre- and post-template exposure (A) suggest that having an IC as a child is associated with both ele- and the relation of performance change to hallucination-proneness (B) vated hallucination-proneness and bias in auditory signal detec- tionskillsasanadult(Fernyhoughetal.2019).Inaddition,children the original sentence templates, participants consistently per- with ICs are more likely to hear words amidst jumbled speech, formed better in terms of their speech vs. non-speech discrimi- whichissimilarinmanywaystoeffectsseenforSWS(Fernyhough nation and increased their bias to state that speech was present et al. 2007). (across intelligible and unintelligible SWS stimuli). None of these Finally,wealsoattemptedtoprovideasecondtestofthemodu- performance changes were related to hallucination-proneness lationofexpectation,bypriminghalftheparticipantswithashort scores on the CAPS or delusion ratings on the PDI. listening activity (listening to a recording of a conversation in a Two limitations are important to consider. The first is that busy room) before attempting the same task as Experiment 1a, using hallucination-proneness in non-clinical analogue samples i.e. discrimination before and after exposure to the sentence tem- hasbeenquestionedforitsabilitytoidentifyindividualswithtruly plates. Reasoning that directing participants to listen for speech hallucinatory experiences (Stanghellini et al. 2012). If valid, this under suboptimal conditions should prime both the expectation couldleadtotheconcernthatcorrelationsbetweentasksandself- of speech and top-down templates for speech, we predicted that report will be very low and very hard to capture, given the low primed participants would go on to show greater speech discrim- base level and minimal variation in proneness scores. The spread ination of SWS in the subsequent task, even before template of CAPS scores shown in Fig. 2 is comparable to prior research of exposure. If this could be demonstrated, it would represent a this kind and not insubstantial when compared to clinical data more naturalistic modulation of expectation, by indirectly prim- (Bell et al. 2006). Nevertheless, directed recruitment of members ing generic speech templates that could assist in the disambigua- of the general population with higher levels of hallucination- tion of the SWS stimuli. The design for Experiment 1b, therefore, proneness could provide greater variation and more opportunity mixed a between-groups approach (prime vs. no-prime) and a to examine how changes in expectation relate to unusual sensory within-subjects approach (before vs. after template exposure). As experiences. in Experiment 1b, we hypothesized that greater improvements in A second concern is that measures of hallucination-proneness discrimination scores on the SWS task would be associated with can yield inconsistent results, and there is currently no ‘gold greater LSHS scores. standard’ for assessing such experiences in the general popu- lation. For the purposes of replication, we used the CAPS but couldhaveinsteadincludedtheLaunay–SladeHallucinationScale Method (Bentall and Slade 1985b) which is arguably a more commonly Participants used scale in prior research on hallucinations. Moreover, the Sixty participants (age M(SD)=23.22 (4.76), range 18–43years, CAPS asks about hallucinations across a range of modalities, 14 male) were recruited from university settings, social media, whereas prevalence rates for AVH—and the auditory nature of and via word-of-mouth. Exclusion criteria were identical to the SWS task—may warrant a more specific measure of auditory Experiment 1a. Within the 60, it was possible to recruit 22 hallucination-proneness. people with a history of having imaginary companions as children, of whom 14 were able to provide parental Experiment 1b verification of their childhood IC—a validation step consid- Modulating prior knowledge with a wider range of hallucination- ered good practice in IC research (Fernyhough et al. 2007, proneness. 2019). 6 Alderson-Day et al. Materials and procedure The same procedure and task structure were used for the SWS discrimination task as in Experiment 1a. In addition, half of the participants completed a priming activity before the discrimina- tion task. The CAPS and PDI were replaced with a version of the Revised Launay–Slade Hallucination Scale. Listening prime task Thirty participants were asked to complete the priming activity before the SWS discrimination task. Participants were given a worksheet and were asked to circle words that they heard being mentioned in a 3-minute pre-recorded conversation between five girls. The recording was layered with white noise to increase the Figure 3. Change in discrimination pre- and post-exposure divided by difficulty in discerning what was being said. The remaining 30 priming group (A) and relation to hallucination-proneness (B) participants were instructed to close their eyes and count their breaths for 3minutes in silence, as timed by the experimenter. before, there was a significant increase in discrimination follow- Questionnaires ing exposure. Pairwise t-tests showed that this was also the case Experiment 1b included a version of the Revised Launay–Slade for both measures of bias (i.e. lower scores, indicating a greater Hallucination Scale (Morrison et al. 2000; McCarthy-Jones and tendency to say speech is present). Fernyhough 2011). Since the development of the original scale by The overall mean LSHS-A score of 9.40 places this sample as Bentall and Slade (1985b), numerous versions of the LSHS have having a mean level of hallucination-proneness comparable to been used to assess hallucination-proneness in the general popu- other samples with childhood histories of ICs, and slightly higher lation. Here we used a five-item version in which all of the items than large samples without any IC history (M=8.76; Fernyhough related specifically to auditory experiences, which participants et al. 2019). As would be expected, participants with an IC ratedforfrequencyonascalefrom1(Never)to4(AlmostAlways). in the present sample had significantly higher LSHS-A scores This version was developed by McCarthy-Jones and Fernyhough (IC M(SD)=10.55 (2.89); No IC M(SD)=8.74 (2.41); Z=−2.56, (2011) following a revision by Morrison et al. (2000). This version P=0.011, r=0.33). has satisfactory internal reliability (typically alpha => 0.7) and However, few relations between SWS task outcomes and has been used to explore task-to-questionnaire relations in vari- questionnaire scores were observed even with higher rates of ousstudiespreviously(e.g.Garrisonetal.2017;Alderson-Dayetal. hallucination-proneness in the sample. A stronger relationship 2019). was evident between the improvement in d and LSHS-A scores (r=0.22, see Fig. 3B), but this was still non-significant on a Spear- man’s test (P=0.089) and Bayesian analysis little evidence for the Results and discussion hypothesized effect (one-sided BF =0.81, or 1.25 in favour of the Testing the priming effect null). Partial correlation, controlling for baseline d scores, also A 2×2 mixed ANOVA (prime group × pre-/post-exposure) was showed no clear association between hallucination-proneness used to assess the effect of the priming condition on discrim- and improvement in performance (r=0.07, P=0.611). ination, plus any interaction it had with exposure to the sen- Relations to bias tence templates (i.e. before vs. after). As in Experiment 1a, there was a clear increase in discrimination following expo- Correlations with bias metrics (C and β) were non-significant sure to the template sentences (F (1,58)=81.31, P=1.24e-12, (all r<0.21, all P>0.08). Pre-exposure discrimination actually neg- η =0.58). But despite the manipulation, no main effect of atively correlated with LSHS scores (r=−0.34, P=0.008, uncor- priming was observed on the discrimination task following the rected), suggesting more hallucination-prone people were worse listening activity (F (1,58)=0.98, P=0.327, η =0.02, nor any at discriminating speech than controls before hearing the sen- interaction effect between priming group and template exposure tence templates. 2 3 (F (1,58)=0.53, P=0.471, η =0.009; see Fig. 3A). Pre-exposure The results of Experiment 1b, therefore, provided a further d scores were very similar in each group [Primed M(SD)=1.68 test of the modulation hypothesis in the auditory domain, but (0.75); Control M(SD)=1.45 (0.77)], as were post-exposure scores could not support it: improved discrimination following template [Primed M(SD)=2.39 (0.79); Control M(SD)=2.29 (0.66)]. Compar- exposure did not significantly relate to auditory hallucination- isonsofpre-exposurebias(both β andC)werealsonon-significant proneness even when including individuals with an IC—therefore (all P>0.300; see Supplementary Materials, Experiment 2). sampling a broader continuum of hallucination susceptibility. In addition, priming expectation for speech with a further manipu- lationyieldednodifferenceindiscriminationperformance. Taken Testing the modulation effect together, the results of experiments 1a and 1b failed to show Given the lack of differences on any task measure, we subse- an effect in the auditory domain comparable to Teufel et al.’s quently combined the priming groups to facilitate comparison modulation effect. with Experiment 1a. Table 2 shows the SWS task outcomes. As Our findings may suggest that a sensitivity to directed mod- ulation of expectation for speech is not part of the putative A Bayesian mixed ANOVA in JASP indicated strong evidence for an expo- mechanism underlying hallucinations. If correct, this would sure effect (BF=5.90) and weak evidence for either an effect of priming group raise problems for a ‘strong priors’ account of the phenomenon (BF=0.44) or an exposure*prime interaction (BF=0.43; BF inclusion statistics reported). (Corlett et al. 2019).However,thefindingsmaystillreflectunusual Susceptibility to auditory hallucinations 7 Table 2. Signal detection outcomes for the SWS discrimination task in experiment 1b Before After M SD M SD Change statistic T(Z) P d(r) d 1.56 0.76 2.34 0.72 −9.05 9.298e-13 −1.05 C 0.29 0.63 −0.42 0.52 9.65 9.59e-14 1.22 β 2.58 3.04 0.88 1.62 −5.79 6.741e-09 0.75 Note: higher values of d indicate increased sensitivity to detect speech. Scores below 0 for C and 1 for beta indicate a bias to indicate speech is present. Wilcoxon tests were used for beta due to non-parametric data. propertiesofSWSitself.Forexample,SWScomprehensionfollow- Materials and procedure ing exposureand trainingcan show something akin to a ‘pop-out’ Auditory stimuli effect, where suddenly new stimuli can be easily understood. In SVS is similar to SWS. However, rather than tracking only the addition, they offer potentially different opportunities for percep- first three formants of speech, the sinewaves are synthesized at tuallearning,comparedtoMooneyimages(Mooney1957). Forthe the centre frequency of a logarithmically spaced bank of filters latter, the level of visual noise is high compared to the level of spanning a broad frequency range (up to 5kHz). Like SWS, SVS repeated signal across trials, meaning that template exposure is sentences can be rendered intelligible and recognizable as speech required per image. For SWS, template exposure typically permits when participants are aware that it is a speech stimulus (Souza more generalized improvements in discrimination across trials and Rosen 2009). SVS can also be rendered unintelligible by flip- (due to similarities in the underlying structure of speech sounds). ping the frequency mapping of the original sentence (e.g. pushing It is therefore important to tightly control for pre-exposure lev- energy in high-frequency bands into low bands and vice versa), els of performance and ensure that difficulty levels are main- providing an ideal control stimulus, with similar complexity and tained for speech discrimination. In the following experiment, we acoustic structure. introduced a new stimulus that could be used to address these The BKB sentences (Bench et al. 1979) were recorded by a male issues. speaker at a sample rate of 22.05kHz. Each sentence was digitally filtered using either 8 or 16 bands, with sixth-order Butterworth IIR filters in MATLAB. Filter spacing was based on equal basilar Experiment 2 membrane distance (Greenwood 1990) across a frequency range of 100–5000Hz. Next, the output of each band was half-wave rec- Modulating prior knowledge of sine-vocoded speech (SVA) with tified and low-pass filtered (fourth-order Butterworth) at 30Hz varying levels of difficulty. to extract the amplitude envelope. The envelope was then mul- Adapting and extending the SWS developed by Rosen et al. tiplied by a tone carrier at the band centre frequency for each (2011) is challenging, as the stimuli were originally hand-edited filter. The resulting signal (envelope×carrier) was filtered using to closely map the formant contours of speech. To address the the same bandpass filter as for the first filtering stage. RMS level factors described above, we deployed a different auditory stimu- was adjusted at the output of the filter to match the original lusinExperiment2—SVS—andmixeditwithunintelligibleSVSto analysis, and the signal was summed across bands. add a source of auditory noise. This provided: (i) a way to auto- When larger numbers of filter bands (e.g. 16 vs. 8) are used to mate the generation of a degraded speech stimulus (rather than synthesizeaspokensentencethisincreasesthespectralinforma- using hand-crafted SWS), allowing the use of a larger number tion in the signal with a resulting increase in intelligibility (Souza of spoken sentences, and (ii) greater control over manipulating and Rosen 2009). task difficulty. We tested pre- and post-exposure identification Sine-vocoding was used to make two types of stimulus: an of degraded speech in a larger sample and its association with intelligible and unintelligible SVS condition. For the Intelligible hallucination-proneness at an increased level of difficulty com- SVS condition, intelligible SVS sentences were mixed with an pared to the previous experiments. The task was administered in unintelligible sine-vocoded sentence that acted as a competing prior knowledge exposure cycles of ten trials, rather than using noise source. This was designed to make stimulus identification a one-shot exposure approach. This was done in part to attempt more difficult reducing learning in the pre-exposure phase and greater parity with the kind of procedure used by Teufel and col- ensuring greater dynamic range in the prior knowledge advan- leagues, and to provide greater access to top-down information tageprovidedbyhearingclearspeechtemplates. TheRMSlevelof after exposure within each cycle. the intelligible sentence was rendered at different levels of inten- sityrelativetotheunintelligiblesentencebeforemixingthem[e.g. a differing signal-to-noise ratio (SNR) was used]. This, alongside Method manipulating the number of vocoding bands, provided a way to Participants manipulate the difficulty of speech identification. Asampleof99participantswasrecruitedfromuniversitysettings, For the Unintelligible SVS condition, two sentences were sine- social media, and via word-of-mouth (age M(SD)=21.58 (3.34), vocoded and frequency flipped and mixed in an equivalent SNR range 18–34years, 45 male). Exclusion criteria were identical to as the Intelligible SVS condition, with one unintelligible sentence thepreviousexperiments.Allparticipantsprovidedinformedcon- arbitrarily assigned to be of greater intensity than the other. This sent in accordance with the approval of the relevant ethics com- ensured that the Intelligible SVS and Unintelligible SVS condi- mittee. Due to experimenter error, the questionnaire data were tions were of equivalent complexity and overall intensity. A set not complete for 12 participants: one participant did not have PDI of stimuli were synthesized from +6dB to −6dB in 3dB steps data and 11 did not have CAPS data. Participants received course using 8 and 16 bands. The sentences composing the intelligible credit for taking part. and unintelligible conditions were mutually exclusive. We have 8 Alderson-Day et al. madethisfullsetofstimulusmaterialsavailablehere(eventhose Results and discussion conditions not used in this article) to facilitate future research: Assessing SVS discrimination across trials and at https://osf.io/yrn9j/. different levels of difficulty Aone-wayindependentANOVAwiththegroupasafactorshowed Auditory task that there was no evidence of a difference in d’ between the five data collection groups, so data were pooled for the com- Participants attended to sounds presented using MATLAB on a mon8bands +6dBcondition(F(4,94)=1.83,P=0.130, η =0.07). laptop using Sennheiser HD 206 headphones. They were ran- Speechidentificationaccuracyinthisconditionwasabovechance domly assigned to five groups which each received two different before prior knowledge exposure (t(98)=10.96, P=9.952e-19), but sets of auditory stimuli (approximately 20 participants in each crucially was significantly increased after prior knowledge expo- group). These two sets differed in either SNR and/or the number sure (t(98)=9.34, P=3.217e-15, d=0.44, see Fig. 4A and Table 3). of bands. Informal piloting indicated that the 8 band +6dB con- A2×8repeatedmeasuresANOVAwithfactorspriorknowledge dition provided an appropriately challenging listening level, such exposure(pre-/post)andblock(1–8)wasconductedtounderstand that accuracy would be above chance, but not at ceiling. Each how d changed across the experiment. This indicated a signifi- group was tested on this common 8 band +6dB condition, plus cantexposure×blockinteraction(F(6.07,595.19)=2.23,P=0.038, oneothercondition. Thiscommonconditionwasincludedtopool η =0.02, see Fig. 4B). Follow-up repeated measure one-way the data across groups to test for the relationship between sig- ANOVAs indicated a change in d’ across blocks before exposure nal detection measures and questionnaire responses. The second topriorknowledge(F(6.23, 610.46)=3.33, P=0.003, η =0.03)but condition had a different signal-to-noise ratio and/or a different notafterit(F(6.18,605.54)=0.86,P=0.530, η =0.009).Inthepre- numberofvocodingbandsandwasincludedtoscopehowspeech exposure phase, the change in accuracy increased linearly with detection accuracy was influenced by the SNR and number of block progression (F(1, 98)=14.83, p=2.100e-4, η =0.13). There bands. was also the main effect of block such that accuracy in general Data from these conditions indicated that the SNR level and increasedacrossblocks(F(5.94,582.36)=2.17,P=0.045, η =0.02) the number of bands had a significant effect on participant per- and from pre- to post-exposure (F (1, 98)=64.69, P=2.054e-12, formance and confirmed the observation that the 8 band +6dB η =0.40). Hence, even with these more challenging stimuli, par- condition provided an appropriately difficult listening experience ticipants continued to improve in their ability to detect speech (see Supplementary Materials, Experiment 2). The specific set of in SVS in the pre-exposure phase, demonstrating the learning sentences used in each condition and order of the conditions was opportunity inherent to SWS/SVS. counterbalanced across participants to ensure that participants did not hear repetitions of any sentences across the experimental session and to reduce order effects. Testing for a modulation effect using SVS Testing in each auditory condition took around 20minutes As in the previous studies, there was no evidence of a significant (40minutes total). Before each condition participants received a relationship between change in d’ and the CAPS (rs (88)=−0.02, short training session in which they were introduced to the intel- P=0.852) or the PDI (rs (98)=−0.12, P=0.257), see Fig. 4C). ligible and unintelligible stimuli. They were informed that they One-sided Bayesian correlations assessing a positive association would hear a 50:50 ratio of intelligible to unintelligible trials in between d’ and the questionnaire measures were used to assess the forthcoming experiment and needed to judge whether each the relative evidence for the null as compared to the experimen- trial contained speech or not. tal hypothesis. These tests provided strong evidence for the null In the experiment, sounds were presented in blocks of ten in hypothesis (both one-sided BF >11). Partial correlation analy- a randomized order (five intelligible trials and five unintelligible sis, controlling for baseline discrimination scores, also provided trials). In the pre-exposure phase, on each trial participants indi- no evidence for an association between d’ change and the CAPS cated whether the sentence contained speech or not. They then (rs (88) =−0.03,P=0.815)andPDIscores(rs(98)=−0.12,P=0.255). receivedexposuretopriorknowledge: theyheardeachtargetsen- Wealsotestedtoseeifchangeind correlatedwitheithertheCAPS tence presented in the original clear speech and saw the written orPDIontheadditionalacousticconditions(e.g. thosedifferingin transcriptofthesentence.Inthepost-exposurephase,theyheard number of bands and SNR level), but this was not the case (see the intelligible and unintelligible sentences that they heard in Supplementary Materials, Experiment 3), suggesting that varying the pre-exposure phase in a different randomized order and were the difficulty level made no difference to the lack of association again asked to indicate which sentences contained speech. The with hallucination-proneness. five transcribed target sentences remained on the right-hand side of the screen during the post-exposure phase to reduce memory Assessing relationships with bias demandsandtomaximizethepriorknowledgebenefit. Thiscycle repeated 8 times, each time with a different set of 10 sentences, A decrease in C after prior knowledge exposure reflected an such that 80 trials were presented in each condition (40 intelligi- increased bias to report the presence of speech [C (t (98)=2.99, ble; 40 unintelligible) and 160 trials werepresented in total across P=0.004, d=0.22)]. However, C did not differ significantly the experiment. from zero either before or after prior knowledge exposure (both Followingadministrationoftheauditorytasks, totalfrequency Ps>0.194).Anon-parametricWilcoxontest—toaccountforadevi- scores were collected for the CAPS (Bell et al. 2006) and the PDI ationinnormality—alsoindicatedthatbetavaluesdidnotchange (Peters et al. 2004). Signal detection measures were calculated as significantly ( z=1.58, P=0.113). These findings suggest that the in the previous experiments. The total testing time was around SVS stimuli provide a more controlled modulation of discrimi- 50minutes. nation, while holding bias relatively more constant. Beta and C Susceptibility to auditory hallucinations 9 ′ ′ Figure 4. (A) d (left) and C change (right panel) with prior knowledge exposure, (B) d change over time in the common 8 band, +6dB condition. Note that the grand mean for d values for the by block analyses differs to the main analysis because the adjustment for extreme values was conducted by block in this instance rather than across the whole experiment (Macmillan and Kaplan 1985), (C) showing the lack of evidence in support of a relationship between d change and the CAPS (left) and PDI measures (right panel) Table 3. Signal detection outcomes for the SVS speech detection task for the 8 band, +6dB condition. Wilcoxon tests were used beta due to non-parametric data Before After M SD M SD Change statistic T(Z) P D(r) d’ 1.20 1.09 1.72 1.25 9.34 3.217e-15 0.44 C 0.04 0.48 −0.06 0.45 2.99 0.004 0.22 β 1.28 1.23 1.14 0.98 1.59 0.113 0.16 change did not correlate with either the CAPS or PDI frequency Our difficulties in demonstrating modulation effects in the measures (all Ps>0.369). auditorydomainraisedtheworrythatourpriorobservationswith Therefore even with a stimulus which is harder to learn and SWS may also be challenging to replicate. It could be that that offers a tighter control on learning effects, we found no evidence such stimuli are too unusual or unique in some way, hampering thatexplicitlymodulatingexpectationsforspeechleadstogreater efforts to index top-down effects on perception. The aim of our gains in discrimination for people who are hallucination prone. final experiment therefore was to replicate and extend the ‘naïve Using SVS we were more able to control learning effects—both listening’ effect observed in our study with NCVH (Alderson-Day pre- and post-exposure—and these were still considerable, even et al. 2017), but this time using SVS. In that study, participants with harder stimuli. However, notwithstanding such effects, we with frequent experience of hearing voices reported recognis- were unable to provide evidence for an auditory version of the ing speech in SWS earlier than control participants, despite not modulation hypothesis. being informed that speech was hidden in the stimulus. Those who reported ‘tuning in’ to SWS earlier also reported signifi- Experiment 3 cantly greater levels of AVH in the preceding week—as mea- sured on the ‘physical characteristics’ subscale of the PSYRATS Replicating and extending the naïve listening effect. 10 Alderson-Day et al. (Haddock et al. 1999)—but they performed no differently to con- target was used to maintain attention and provide an incidental trols on a post-exposure discrimination task, in a similar manner task which would discourage participants guessing at the pur- to the other studies reported here. pose, i.e. the potential intelligibility of distractors, hidden in the To reproduce this, we reran the naïve listening procedure SVS. As in Alderson-Day et al. (2017), the ‘scratchy’ sounds were and collected hallucination-proneness measures on the LSHS-A examples of unintelligible SWS (i.e. the frequency and amplitude (n=134) in a larger, healthy sample of individuals. Moreover, we tracks of two separate original sentences combined), that had addedanadditionalproceduretointroduceamoreobjectivemea- been further noise-vocoded, giving them a different timbre and sureofstimulusdecoding.Oneworryaboutouroriginalprocedure soundquality.Onceparticipantscoulddiscriminatethetwokinds wasthatitreliedonparticipantself-report,andthusmaybeopen of sound, they attempted the main listening task, which con- toparticipantsclaimingtheyhadheardspeechorguessingspeech tainedsixblocksof15SVSstimuli(45intelligible,45unintelligible) was present without actually decoding it. Here we introduced a and three targets per block. At the start of each block, a visual memory test that could only be successfully completed if partic- stimulus appeared announcing the start of the block (i.e. block ipants had actually been understanding words in the SVS prior 1, block 2, etc). Stimuli were presented in a predefined pseudo- to the ‘reveal’ that speech was present. This, therefore, would random order with no more than two of the same kind of stimuli extend our initial finding by providing more objective evidence consecutively. of early SVS comprehension. Using this procedure, we predicted Once the participants had listened to all six blocks, they were that people higher in auditory hallucination-proneness would (i) asked by the experimenter (i) if they noticed anything unusual reportrecognisingspeechintheSVSearlierthanothers,and(ii)be about the words, (ii) if they noticed any words and sentences abletoremembersignificantlymorewordshiddenintheSVStask. and crucially (iii) if they knew which round they started notic- Wealsoincludedmeasuresofvisualhallucination-pronenessand ing them (using the visual block markers as means of marking general schizotypal traits—to test for specificity—and a test of out time). Participants’ estimates for the third question were SVS discrimination post-exposure (for comparison with our prior used as the main task outcome, defined as their ‘recognition study). point’. They were then told that in fact there were words present in the stimuli, and asked to complete the memory task for words contained in the SVS. The memory test consisted of 46 Method words, including 18 words included in the SVS (3 per block), Participants 18 words matched for length and complexity that did not fea- We recruited 134 participants (age M(SD)= 21.45 (5.79), range ture in the SVS, five words from the NART, and five non-target 18–59years, 46 male/2 other) via university departments, social words matched to the NART words for their irregular spelling. media, and word-of-mouth. Exclusion criteria were the same as Following recognition memory methodology (Tulving 1985), par- for the previous experiments. Participants were invited to take ticipants were asked to indicate for each word whether they part in a ‘study of auditory perception’ that involved listening to explicitly remembered the word (R), felt like they knew they had some ‘unusual sounds’, but no mention of voices or speech was heard the word at some point (K), or new items that they didn’t included in the study materials. recognize (N). As we were most interested in participants actu- ally being able to decode the words, we focused on remember Materials and procedure scores. New items, in contrast, acted as lures for potential false The procedure for experiment 3 closely followed Alderson-Day positives. etal.(2017).Themeasuresusedarereportedhereintheorderthey AsinAlderson-Dayetal. (2017), participantsalsothenreceived were attempted by participants. training in understanding SVS, and then attempted an accu- racy task (testing their speech/non-speech discrimination and National Adult Reading Test (NART; Nelson 1982) accuracy for understanding keywords in 25 intelligible and 25 The NART is a measure of vocabulary and reading ability which unintelligible SVS trials). Results for these tasks are included in has been used extensively in research on psychosis as an indica- Supplementary Materials. tor of premorbid IQ (e.g. Broome et al. 2012) and was included in Alderson-Day et al. (2017) for group matching. We retained it here Questionnaires to follow that procedure, but also to provide control material for Following the tasks, participants completed the LSHS-A (as in the memory task. By including a small selection of words from Experiment 1b), but also two further measures: the four corre- the NART (plus similar words matched for unusual spelling), we sponding visual questions from the LSHS (i.e. the LSHS-V) and could control for general memory differences between those who the brief version of the Oxford Liverpool Inventory of Feelings did and did not recognise speech in the SVS. and Experiences (OLIFE; Mason and Claridge 2006; Fernyhough et al. 2008). This allowed for specificity testing by comparing audi- SVS naïve listening procedure toryhallucination-pronenesswithvisualexperiencesandgeneral Intelligible and unintelligible SVS stimuli were drawn from the proneness to schizotypal experiences. same set as Experiment 2. In Experiment 3, the 16-band, +6db stimuli were used, as these appeared to be the most similar to Results and discussion SWS in terms of their level of difficulty. Participants were told Assessing naïve listening via self-report that they would be listening out for a scratchy target sound that sounded ‘different’ from the other sounds and began the pro- No participants guessed the purpose of the experiment before cedure by listening to three examples of the target randomly testing. Overall, 83/134 participants (62%) recognized speech presented along with five examples of unintelligible SVS. The being present in the SVS and 51 (38%) did not. Participants that Specifically,participantsweretold: ‘Yourtaskistolistenoutforatargetsound the others. Have a listen to these examples, and press the space bar if you think you hear the unusual sound.’ and press the space bar every time you hear it. It might sound “scratchy” compared to Susceptibility to auditory hallucinations 11 recognized speech did so most frequently in the second block General discussion (mean=2.58, median=3, range=1–5) and of those, 72 were able Our aim in the present paper was to align and reconcile two find- to repeat some of the words encoded in the stimuli. ingsofenhancedtop-downperceptualprocessinginpeopleprone We wanted to assess how well the LSHS-A predicted which to hallucinations. Across four experiments, we were only par- participants recognised the speech without training. In a logis- tiallysuccessful inthis aim. In Experiments1a, 1b and2 wecould tic regression model, LSHS-A scores significantly predicted not provide evidence for a modulation hypothesis in the audi- recognition group (Z=2.206, P=0.027, OR=1.21, CI=1.03–1.44). tory domain. People who were more prone to hallucinations did Specificity analysis swapping LSHS-A for visual items of the not appear to draw upon prior expectation more in their percep- Launay–Slade (LSHS-V) or a measure of general schizotypy (the tion when their expectations were explicitly updated. In contrast, OLIFE) did not result in significant models (see Supplementary in Experiment 3 we provided further evidence for a naïve listen- Materials). ing effect: healthy people who were prone to hallucinations were The ability to decode ambiguous speech stimuli and sim- more able than others to identify speech in SVS without knowing ilar skills (such as extracting speech from noise) are known that speech was present. to vary considerably across individuals. To explore the naïve This pattern of results suggests a spontaneous, rather than listening effect, we ran an exploratory analysis drawing on directed, use of top-down resources in people who are prone to the NART scores collected at the start of the experiment. unusual perceptions, in contrast to prior findings using visual Along with offering a very rough proxy of verbal intelligence, paradigms (Teufel et al. 2015). How to explain this discrepancy? NART scores reflect vocabulary and reading ability—all of which First, it is important to consider the potential differences in the would plausibly contribute to the identification of ambiguous kind of information provided by each stimulus. As already noted, speech. We therefore reran our logistic regression analyses, SWS/SVS appeared to allow for learning across trials, even in a but including NART as an additional predictor. Higher NART naïve state. This represents an opportunity for perceptual learn- scores were associated with identifying speech spontaneously ing(SohogluandDavis2016)thatwouldnotappeartobeapparent (Z=3.78, P<0.001, OR=1.15, CI=1.08–1.25) but the contribu- when learning to discriminate Mooney images. Moreover, it raises tionofLSHS-Ascoreswasnownon-significant( Z=1.80,P=0.072, the possibility that some participants may learn to discriminate OR=1.17, CI=0.99–1.40), despite similar odds ratios for the two speech from non-speech without understanding the content of predictors. the speech (doing so possibly via prosodic or structural cues). As such, discriminatingspeechfromnon-speechinSWS/SVSmaybe Assessing naïve listening via memory posing a different kind of challenge to Mooney images, and may performance not be solely reliant on top-down knowledge. Following the reveal that speech was present participants were This might raise the concern that SWS/SVS stimuli are testedontheirmemoryforhiddenwords,toprovideamoreobjec- just too different from degraded images to explore top-down tive test of spontaneous SVS decoding. Across the whole sample effects in a comparable way. However, when discrimination was (i.e. combining those who did and did not recognize speech spon- made especially challenging and learning effects minimized (in taneously), accuracy for ‘remember’ items on the memory task experiment 2), we still observed no evident relationship between significantly correlated with LSHS-A scores, rs=0.20, P=0.020 hallucination-pronenessandtemplateexposure. Moreover, inour (Spearman’s test). final experiment, people who were more AVH-prone were specifi- This relation could feasibly have been driven by a general ten- cally more able to remember hidden words in the SVS, suggesting dency to endorse words on the memory test (i.e. a false positive they could successfully decode the stimuli (rather than discrimi- bias),orgeneraldifferencesinmemoryscores.Ifso,LSHS-Ascores nating speech from non-speech in a more general way). If the dif- would have also correlated with (i) endorsement of new (i.e. lure) ference between auditory and visual stimuli of this kind is simply items on the memory task, or (ii) recall for NART items included oneofdifficulty, orthenumberofgeneralizablecuesacrosstrials, in the memory task. However, neither new item scores, r=-0.00, thenthiscanbeexploredempiricallyinfuturestudies.Itwouldbe P=0.969,norformemoryforNARTwords,r=0.05,P=0.551,were possibletoparametricallyvarytheperceptibilityofhiddenspeech associated with LSHS-A scores. This suggested that the ability to inSVSbyadaptingthesignal-to-noiseratioandreducingthenum- understand the SVS early, and thus encode more hidden words, ber of vocoding channels. This would dampen such effects and was indeed related to greater hallucination-proneness. could demonstrate modulatory effects that would be comparable Given the apparent role of NART scores in predicting self- to those seen in the visual domain. reported recognition of SVS, we used partial correlation tests to If the lack of any modulation effect is genuine, however, it assess what role it played in the relationship between LSHS-A might suggest that healthy people prone to unusual experiences scoresand‘remember’scoresforwordshiddenintheSVS.LSHS-A are not necessarily susceptible to momentary effects—such as scores positively correlated with words remembered when con- suggestibility, or demand characteristics—and instead possess trolling for the NART, r=0.18, P=0.035), but no significant rela- perceptual biases that are somehow more ingrained. In predic- tion between memory performance and NART scores was evident tive processing terms, this could constitute a higher-order belief when controlling for LSHS-A, r=0.13, P=0.136. about the world (e.g. ‘the universe is full of hidden meanings’) The results of Experiment 4 therefore supported the original which directs the individual response in ambiguous situations to findings of Alderson-Day et al. (2017), but added to it by provid- explorepotentiallyimportantsignalsdespitetheinstructiontolis- ing a more objective test of participants being able to recognize ten for the target sound. Exploring differences in attention during speech early—namely, actual improved recall for words hidden naïvelisteningwillbevital, assomeparticipantsattitudetowards in the SVS. Moreover, this was specific to auditory hallucination- unusual and ambiguous stimuli may lead them to allocate more proneness and not visual hallucination-proneness, or general attentiontoSVS,thusgivingthemagreateropportunitytoengage schizotypy. in explicit or implicit processes of perceptual learning. Recent 12 Alderson-Day et al. research on primal world beliefs represents a promising avenue butclosercontrolofthesefactorswouldhavebeentotesthearing for this line of research (Clifton et al. 2019). skills in all participants. Alternatively, hallucinations may relate to a very different An open question for future research is how people with fre- level in the processing hierarchy and be expressed in how audi- quent and distressing hallucinations perform under naïve listen- tory objects are represented. A recent reframing of the predictive ingconditionsandhowaperceptualadvantageinsomeconditions approach by Teufel and Fletcher (2020) has proposed to sepa- translates into non-veridical experiences in other contexts. A ratelong-term, contextinvariant‘constraints’onperceptionfrom recent small-scale study by Kafadar and colleagues (2020)—using context-specific, temporary ‘expectations’ that shape immedi- the structure of Experiment 1a but the same stimuli as Exper- ate, moment-to-moment sensation. Our data point towards a iments 2 and 3—found that people at clinical high risk of psy- long-term constraint for some individuals in how they recog- chosisaremarkedmorebytheirpre-exposurebiastohearspeech, nise auditory objects when faced with ambiguous speech-like rather than discrimination. They also found no differences in stimuli, rather than sensitivity to changes in situation-specific post-exposure discrimination, suggesting again that the explicit expectations. Support for this argument comes from recent work modulation of expectation was less relevant to understanding on the templates that people use to make judgements about how they perceived SVS. It may be that healthy individuals who speech. By creating ‘speechiness’ kernels for individual partic- have unusual experiences have very slight biases in their percep- ipants, Erb and colleagues (2020) have demonstrated that peo- tion that facilitate a top-down advantage, akin to the concept of ple high in hallucination-proneness utilize qualitatively differ- encoding style for internal over external meanings (Valerie et al. ent speech templates when discriminating speech sounds, such 2011). Conversely, individuals with clinical hallucinations may be that lower frequencies typical of speech are attended to less, more fixed in their expectations of finding signal in noise, leading compared to higher frequency auditory information. If so, this tonon-veridicalexperiencesmoregenerally. Insuchasituation, a suggests a fundamental long-term alteration to how speech is slightbiasmightyieldanadvantage—especiallyifitcanbeapplied recognisedandprocessedinthosepronetohallucinatoryandillu- selectively—whereas a strong bias would lead to false positives sory experiences—rather than a dynamic, moment-to-moment (i.e. hearing speech in unintelligible SWS/SVS). volatility in how expectations are managed. Variation of naïve lis- In conclusion, the experiments that we present here refine tening effects with SVS tailored to different speech kernels could our understanding of how top-down expectations shape speech perception for people prone to auditory hallucinations. Directly be used to explore this further. updatingexpectationwouldnotappeartoconferanadvantagefor Finally, such effects are likely to be shaped by a considerable peoplehigherinhallucination-proneness—atleastforthesekinds number of individual differences, including auditory experience of stimuli—but being susceptible to hallucinations would appear and cognitive factors such as verbal IQ (Gwilliams and Wallisch to be associated with responding to ambiguous speech stimuli in 2020). We explored the role of NART scores in our final exper- a differential way, facilitating speech identification. This impli- iment for this reason. The observed relationship between those cates longer term and potentially lower-level constraints on how scoresandspontaneousrecognitionsuggestsitmayhaveasignif- speechisrecognizedinsuchpopulations,ratherthanatemporary icant impact on SVS discrimination, with implications for future sensitivity to expectation. studies. Importantly, the same relationship was not evident for performance on the memory task—suggesting that verbal IQ dif- ferences could not fully account for the relationship between SVS Supplementary data decoding and hallucination-proneness. Nevertheless, the range Supplementary data is available at NCONSC online. of variables guiding individual comprehension skills for degraded speech, noise-vocoded speech and speech-in-noise are vast, com- Data availability plex and yet to be clearly pinpointed (McGettigan et al. 2012). Somelimitationsofourgeneralapproachmustbenoted. First, Data and analysis code for each of the experiments are available we tested exclusively university and general population-based via HYPERLINK https://osf.io/yrn9j/ OSF. samples, rather than either clinical or NCVH (i.e. those with very regular hallucination-like experiences; Johns et al. 2014). Acknowledgements We therefore cannot rule out that participants with more fre- We thank Marissa Thomas, Kamal Duran, Charlotte Newman, quent experiences would not show specific modulation effects Paris Brown, Uzma Farooqui, Ellie Maycock and Summer Zhu on their perception. However, predictive approaches to hallu- for helping to collect data for this project. This research was cinations are often explicitly framed as models of both clin- funded in whole, or in part, by the Wellcome Trust [Grant num- ical and non-clinical phenomena, with the former resulting ber WT108720]. For the purpose of open access, the author has from an accentuation of mechanisms underlying normal, veridi- applied a CC BY public copyright licence to any Author Accepted cal perception rather than being specific to clinical disorders Manuscript version arising from this submission. (Corlett et al. 2019). Our data therefore would appear relevant to understanding perceptual mechanisms across the psychosis continuum. Funding Second, none of our experiments deployed basic tests of audi- This research was funded in whole, or in part, by the Wellcome tory processing or hearing ability, which could contribute to low- Trust [Grant number WT108720]. level differences in how SWS/SVS are recognized and processed. Thereisemergingevidencethatsubtledifferencesinhearingabil- Conflict of interest statement itycanbeassociatedwithhallucinationsforsomepeople(Linszen et al. 2016). Participants with a self-declared hearing difficulty None declared. were not recruited to the study, and tests of speech intelligibil- ity do not pose the same challenges to hearing as other tasks usedinhallucinationresearch(suchasauditorysignaldetection), Kristiina Kompus (U. Bergen) is thanked for this comparison. Susceptibility to auditory hallucinations 13 Greenwood DD. A cochlear frequency-position function for several References species—29 years later. J Acoust Soc Am 1990;87:2592–605. Alderson-DayB,LimaCF,EvansSetal.Distinctprocessingofambigu- Griffin JD, Fletcher PC. Predictive processing, source monitoring, and ousspeechin peoplewith non-clinicalauditoryverbalhallucina- psychosis. Annu Rev Clin Psychol 2017;13:265–89. tions. Brain 2017;140:2475–89. GwilliamsL,WallischP.Immediateambiguityresolutioninspeechpercep- Alderson-Day B, Smailes D, Moffatt J et al. Intentional inhibition tion based on prior acoustic experience. 2020. 10.31234/osf.io/yptfr. but not source memory is related to hallucination-proneness HaddockG,McCarronJ,TarrierNetal.Scalestomeasuredimensions and intrusive thoughts in a university sample. Cortex 2019;113: of hallucinations and delusions: the psychotic symptom rating 267–78. scales (PSYRATS). Psychol Med 1999;29:879–89. BauerSM,SchandaH,KarakulaHetal.Cultureandtheprevalenceof Hohwy J. The Predictive Mind. Oxford: Oxford University Press, 2014. hallucinationsinschizophrenia.ComprPsychiatry2011;52:319–25. Jardri R, Denève S. Circular inferences in schizophrenia. Brain Bell V, Halligan PW, Ellis HD. The Cardiff Anomalous Perceptions 2013;136:3227–42. Scale (CAPS): a new validated measure of anomalous perceptual Johns LC, Kompus K, Connell M et al. Auditory verbal hallucina- experience. Schizophr Bull 2006;32:366–77. tions in persons with and without a need for care. Schizophr Bull Bench J, Kowal Å, Bamford J. The BKB (Bamford-Kowal-Bench) 2014;40:S255–S64. sentence lists for partially-hearing children. Br J Audiol 1979; Kafadar E, Mittal VA, Strauss GP et al. Modeling perception and 13:108–12. behaviorinindividualsatclinicalhighriskforpsychosis: support BentallRP.Theillusionofreality:areviewandintegrationofpsycho- for the predictive processing framework. Schizophr Res 2020;226: logical research on hallucinations. Psychol Bull 1990;107:82–95. 167–75. Bentall RP, Slade PD. Reality testing and auditory hallucinations: a Kim S. 2015. ppcor: Partial and Semi-Partial (Part) Correlation. signal detection analysis. Br J Clinial Psychol 1985a;24:159–69. https://CRAN.R-project.org/package=ppcor (1 July 2020, date last Bentall RP, Slade PD. Reliability of a scale measuring dispo- accessed). sition towards hallucination: a brief report. Pers Individ Dif 1985b;6:527–9. Linszen MMJ, Brouwer RM, Heringa SM et al. Increased risk of psy- Brookwell ML, Bentall RP, Varese F. Externalizing biases and hal- chosis in patients with hearing impairment: review and meta- lucinations in source-monitoring, self-monitoring and signal analyses. Neurosci Biobehav Rev 2016;62:1–20. detection studies: a meta-analytic review. Psychol Med 2013;43: Macmillan NA, Kaplan HL. Detection theory analysis of group data: 2465–75. estimating sensitivity from average hit and false-alarm rates. Broome MR, Day F, Valli I et al. Delusional ideation, manic symp- Psychol Bull 1985;98:185–99. tomatology and working memory in a cohort at clinical high-risk Maijer K, Hayward M, Fernyhough C et al. Hallucinations in children for psychosis: a longitudinal study. Eur Psychiatry 2012;27:258–63. and adolescents: an updated review and practical recommenda- Clark A. Whatever next? Predictive brains, situated agents, and the tions for clinicians. Schizophr Bull 2019;45:S5–S23. future of cognitive science. Behav Brain Sci 2013;36:181–204. Mason O, Claridge G. The Oxford-Liverpool Inventory of Feelings and CliftonJDW,BakerJD,ParkCLetal.Primalworldbeliefs.PsycholAssess Experiences (O-LIFE): further description and extended norms. 2019;31:82–99. Schizophr Res 2006;82:203–11. Corlett PR, Horga G, Fletcher PC et al. Hallucinations and strong Mason O, Claridge G, Jackson M. New scales for the assessment of priors. Trends Cogn Sci 2019;23:114–27. schizotypy. Pers Individ Dif 1995;18:7–13. Erb J, Kreitewolf J, Pinheiro AP et al. Aberrant perceptual judgements McCarthy-Jones S, Fernyhough C. The varieties of inner speech: onspeech-relevantacousticfeaturesinhallucination-proneindi- links between quality of inner speech and psychopathological viduals. Schizophr Bull Open 2020;1:sgaa059. variables in a sample of young adults. Conscious Cogn 2011;20: ´ ´ ´ ´ EsquirolJ.Surlesillusionsdessenschezlesalienes.ArchivesGenerales 1586–93. de Medecine 1832;2:23. McGettiganC,EvansS,RosenSetal.Anapplicationofunivariateand Feinberg I. Efference copy and corollary discharge: implications for multivariate approaches in fMRI to quantifying the hemispheric thinking and its disorders. Schizophr Bull 1978;4:636–40. lateralization of acoustic and linguistic processes. J Cogn Neurosci Fernyhough C, Bland K, Meins E et al. Imaginary companions and 2012;24:636–52. youngchildren’sresponsesto ambiguousauditorystimuli: impli- Mintz S, Alpert M. Imagery vividness, reality testing, and cations for typical and atypical development. J Child Psychol Psy- schizophrenic hallucinations. J Abnorm Psychol 1972;79:310–6. chiatry 2007;48:1094–101. Mooney CM. Age in the development of closure ability in chil- Fernyhough C, Jones SR, Whittle C et al. Theory of mind, schizotypy, dren. Can J Psychol/Revue Canadienne de Psychologie 1957;11: and persecutory ideation in young adults. Cogn Neuropsychiatry 219–26. 2008;13:233–49. Morrison AP, Haddock G, Tarrier N. Intrusive thoughts and audi- Fernyhough C, Watson A, Bernini M et al. Imaginary companions, tory hallucinations: a cognitive approach. Behav Cogn Psychother inner speech, and auditory verbal hallucinations: what are the 1995;23:265–80. relations? Front Psychol 2019;10:1665. Morrison AP, Wells A, Nothard S. Cognitive factors in predispo- Fletcher PC, Frith C. Perceiving is believing: a Bayesian approach sition to auditory and visual hallucinations. Br J Clin Psychol to explaining the positive symptoms of schizophrenia. Nat Rev 2000;39:67–78. Neurosci 2009;10:48–58. Nelson HE. National Adult Reading Test. Test Manual. Windsor: NFER- Frith C. The Cognitive Neuropsychology of Schizophrenia. New York: Nelson, 1982. Psychology Press, 1992. Pearson D, Burrow A, FitzGerald C et al. Auditory hallucinations in Garrison JR, Moseley P, Alderson-Day B et al. Testing continuum normal child populations. Pers Individ Dif 2001;31:401–7. models of psychosis: no reduction in source monitoring ability Peters E, Joseph S, Day S et al. Measuring delusional ideation: the in healthy individuals prone to auditory hallucinations. Cortex 21-item Peters et al. Delusions Inventory (PDI). Schizophr Bull 2017;91:197–207. 2004;30:1005–22. 14 Alderson-Day et al. Peters E, Joseph SA, Garety PA. Measurement of delusional ideation Stanghellini G, Langer AI, Ambrosini A et al. Quality of halluci- in the normal population: introducing the PDI (Peters et al. Delu- natory experiences: differences between a clinical and a non- sions Inventory). Schizophr Bull 1999;25:553–76. clinical sample. World Psychiatr: Off J World Psychiatr Assoc (WPA) Peters E, Ward T, Jackson M et al. Clinical, socio-demographic 2012;11:110–3. and psychological characteristics in individuals with persistent Stanislaw H, Todorov N. Calculation of signal detection theory mea- psychotic experiences with and without a “need for care”. World sures. Behav Res Methods Instrum Comput 1999;31:137–49. Psychiatry 2016;15:41–52. Taylor M. Imaginary Companions and the Children Who Create Them. Pickering MJ, Clark A. Getting ahead: forward models and their place Oxford: Oxford University Press, 1999. in cognitive architecture. Trends Cogn Sci 2014;18:451–6. Teufel C, Fletcher PC. Forms of prediction in the nervous system. Nat Powers AR, Kelley M, Corlett PR. Hallucinations as top-down Rev Neurosci 2020;21:231–42. effects on perception. Biol Psychiatr: Cognit Neurosci Neuroimaging Teufel C, Subramaniam N, Dobler V et al. Shift toward prior 2016;1:393–400. knowledge confers a perceptual advantage in early psychosis Remez RE, Rubin PE, Pisoni DB et al. Speech perception without and psychosis-prone healthy individuals. Proc Nat Acad Sci traditional speech cues. Science 1981;212:947–9. 2015;112:13401–6. Rosen S, Wise RJS, Chadha S et al. Hemispheric asymmetries in Tulving E. Memory and consciousness. Can Psychol/Psychologie Cana- speech perception: sense, nonsense and modulations. PLoS One dienne 1985;26:1–12. 2011;6:e24672. ValerieRV,BelayachiS,VanderLindenM.Internalencodingstyleand Seal ML, Aleman A, McGuire PK. Compelling imagery, unanticipated schizotypy in a sub-clinical sample. Eur Psychiatry 2011;26:1519. speech and deceptive memory: neurocognitive models of audi- Waters F, Badcock JC, Maybery MT et al. Inhibition in schizophre- tory verbal hallucinations in schizophrenia. Cogn Neuropsychiatry nia: association with auditory hallucinations. Schizophr Res 2004;9:43–72. 2003;62:275–80. Senn S. Change from baseline and analysis of covariance revisited. Waters F, Collerton D, Ffytche DH et al. Visual hallucinations Stat Med 2006;25:4334–44. in the psychosis spectrum and comparative information from SohogluE,DavisMH.Perceptuallearningofdegradedspeechbymin- neurodegenerative disorders and eye disease. Schizophr Bull imizing prediction error. Proc Nat Acad Sci 2016;113:201523266. 2014;40:S233–S45. Souza P, Rosen S. Effects of envelope bandwidth on the intel- Wilkinson S. Accounting for the phenomenology and varieties of ligibility of sine- and noise-vocoded speech. J Acoust Soc Am auditory verbal hallucination within a predictive processing 2009;126:792–805. framework. Conscious Cogn 2014;30:142–55.

Journal

Neuroscience of Consciousness – Oxford University Press

Published: Feb 1, 2022

Keywords: predictive coding; psychosis; speech perception; speech-in-noise; audition

There are no references for this article.