Hearing voices as a feature of typical and psychopathological experience

Hearing a voice in the absence of any speaker can be a significant feature of psychiatric illness, but is also increasingly acknowledged as an important aspect of everyday, non-pathological experience. This recognition has led to a growth of interest in voice-hearing in individuals without any psychiatric diagnosis, coupled with greater attention to the subjective experience of voice-hearing across diagnostic groups. Research has also focused on the overlap between some aspects of voice-hearing phenomenology and everyday experiences such as ‘hearing’ the voices of fictional characters and spiritual experience. In this Review, we synthesize research on the range of cognitive, neural, personal and sociocultural processes that contribute to voice-hearing as it occurs in clinical, non-clinical and everyday experience, with particular emphasis on linking mechanism to phenomenology. Heterogeneous forms of voice-hearing can be understood in terms of differing patterns of association among underlying mechanisms. We suggest an approach to hallucinatory experience that sees it as partly continuous with everyday inner experience, but which is critical regarding whether continuity of phenomenology across the clinical–non-clinical divide should be taken to entail continuity of mechanism. Hearing voices has long been associated with severe mental illness but also occurs in the general population. In this Review, Toh et al. describe the cognitive, neural, personal and sociocultural processes that contribute to voice-hearing in clinical, non-clinical and everyday experience, with emphasis on linking mechanism to phenomenology.

Hearing a voice in the absence of any speaker has long been associated with severe mental illness. Voice-hearing has been proposed as a cardinal symptom of the schizophrenia spectrum disorders, with up to 80% of those with a diagnosis of schizophrenia reporting hearing voices at some point over their lifetime 1 . Voice-hearing is increasingly recognized as a feature of other psychiatric conditions, including bipolar and depressive disorders 2 , post-traumatic stress disorder 3 and borderline personality disorder 4 . Syndromes with neurological origins, such as Parkinson disease 5 , temporal lobe epilepsy 6 and migraine 7 , also involve hallucinatory experiences (although not always in the auditory realm).
Notably, voice-hearing also occurs in a small but significant portion of the general population without a mental health diagnosis 8 . In the absence of distress, voice-hearing can be a significant feature of life for these individuals. Many features of voice-hearing also have important commonalities with aspects of typical consciousness that would not be considered signs of mental disorder, such as engaging with fictional entities and imaginary companions 9 . This move towards recognizing that hearing voices is not necessarily pathological has led to the emergence of the term 'voice-hearing' as a replacement for the more medically oriented 'auditory verbal hallucination' , which has potential pejorative connotations.
Historically, there has been a focus on voices as primarily auditory experiences, likely due to the strong association between auditory hallucinations and the (controversial) diagnostic category of schizophrenia. However, the experience of voice-hearing is rarely confined exclusively to the auditory modality. A significant proportion of voices are described as 'soundless' voices, or as difficult to distinguish from regular thoughts 10,11 . Voice-hearing is also often accompanied by sensed presences (that is, feelings that someone is present in the immediate environment, often without clear sensory content) and experiences in other sensory modalities, including visual, somatic-tactile, olfactory, autoscopic and kinesthetic 12 . The term 'voice-hearing' thus encompasses a heterogeneous group of experiences that can differ widely in their auditory phenomenology. Although we favour this term, we will use the more conventional term 'hallucination' when discussing clinically significant experiences that are specific to non-auditory modalities.
In parallel with the growing recognition of the phenom enological complexity of voice-hearing, there has been an increase in research activity examining the causes and maintenance of the voice-hearing experience. In contrast to the 'deficit model' of traditional biomedical psychiatry, which sees voice-hearing as resulting from a dysfunction in normal brain processes, voice-hearing is increasingly understood as a varied phenomenon which can emerge from distinct causal pathways. This variability has important implications for treatment in

Autoscopic
The psychic duplication of a part or the whole of one's body in external space; in other words, an out-of-body experience.
Hearing voices as a feature of typical and psychopathological experience Wei Lin Toh 1

, Peter Moseley 2 and Charles Fernyhough 3 ✉
Abstract | Hearing a voice in the absence of any speaker can be a significant feature of psychiatric illness, but is also increasingly acknowledged as an important aspect of everyday, non-pathological experience. This recognition has led to a growth of interest in voice-hearing in individuals without any psychiatric diagnosis, coupled with greater attention to the subjective experience of voice-hearing across diagnostic groups. Research has also focused on the overlap between some aspects of voice-hearing phenomenology and everyday experiences such as 'hearing' the voices of fictional characters and spiritual experience. In this Review, we synthesize research on the range of cognitive, neural, personal and sociocultural processes that contribute to voice-hearing as it occurs in clinical, non-clinical and everyday experience, with particular emphasis on linking mechanism to phenomenology. Heterogeneous forms of voice-hearing can be understood in terms of differing patterns of association among underlying mechanisms. We suggest an approach to hallucinatory experience that sees it as partly continuous with everyday inner experience, but which is critical regarding whether continuity of phenomenology across the clinical-non-clinical divide should be taken to entail continuity of mechanism. 0123456789();: cases where voice-hearing experiences are distressing 13 . Research has accordingly begun to focus on voices as an interesting and important aspect of the diversity of typical human inner experience rather than as necessarily the product of a 'broken brain' .
In this Review, we examine the state of research on voice-hearing that incorporates experiences ranging from the distressing, disabling hallucinations of severe mental illness to everyday experiences. We begin by describing the varied kinds of voice-hearing experience, why they are sometimes distressing and how their phenomenological features might differ across clinical and non-clinical categories. The full variety of voice-hearing experiences is then explored in light of the cognitive, neural, personal and sociocultural mechanisms that have been implicated in its causes, maintenance, phenomenological variability and clinical significance, including some mechanisms likely to operate across sensory modalities. We examine the present state of the 'continuum' hypothesis, whereby such experiences exist on a spectrum from mild, transient experiences to frank clinical disorder, with significant variability in between. Finally, we illustrate how considering these three categories of voice-hearingclinical, non-clinical and everyday voice-hearing -can help researchers understand the multiple possible pathways to voice-hearing, and recommend an agenda for future research on this complex, varied and significant feature of human experience.

Phenomenology of voice-hearing
The characteristics of voice-hearing in clinical and non-clinical cohorts have been well documented 8,11,[14][15][16][17][18][19][20][21][22][23][24][25][26] . Phenomenological features associated with voice-hearing can differ somewhat across these categories, but there are actually more shared commonalities than differences 27 . These phenomenological similarities and differences, organized into perceptual-sensory, cognitive and affective domains, are summarized in Table 1. Prevalence estimates for voice-hearing in the general population differ, with lifetime figures between 0.6% ("Have you ever heard sounds, music, or voices which other people can't hear?") 28 and 84% (at least one lifetime experience of "brief, auditory hallucinations of the voice type", including sleep-related experiences) 29 . These estimates depend on what the participant believes is being asked of them, signalling a need for clear and consistent terminology.
Throughout this Review, we broadly categorize voices into three groups: clinical voices associated with distress or need for clinical care or treatment; non-clinical voices that are frequent and often recurring but not associated with distress or mental health diagnoses; and everyday voices encompassing experiences such as inner speech or erroneously hearing one's name being called. We do not claim that these are 'true' categories or natural kinds; rather, we suggest that they provide a useful way to think about variability in these experiences. The prevalence statistics cited here do not always clearly map onto one of these categories. For example, voice-hearing prevalence in the general population typically includes prevalence of experiences described in both our non-clinical category and our everyday category.

Clinical voices.
In the clinical group (often, but not always, involving a diagnosis of schizophrenia), patients with psychosis tend to report hearing speech, often with some non-verbal sounds, with voice onset typically in adolescence or early adulthood 27 . These experiences occur on a continuum ranging from fully to 'mixed' auditory percepts (comprising features common to both speech percepts and regular thoughts) to soundless or thought-like voices 11 . Perceptual-sensory characteristics of clinical voices tend to exhibit high variability from person to person as well as within a single person. Frequency and duration can fluctuate from constant, daily voices to those that arise only during acute psychosis; volume can range from a whisper to shouting, although conversational levels are common; clarity can range from incoherent to sharp; a single voice can predominate, but multiple voices (up to a crowd) are possible; localization can comprise internal or external voices or both; linguistic complexity can differ, although repetition of phrases is typical; and second-person or third-person address is most likely 20,24,30,31 .
Cognitive characteristics of clinical voices seemingly exhibit less variation relative to perceptual-sensory charac teristics. Levels of conviction as to the veracity of voices tend to be high; there is limited perceived controllability; voices often have an autonomous, non-self quality; and beliefs regarding voice origin may incorporate a mixture of psychological and environmental factors. There is some variability in when and how patients interact and comply with their voices, but vivid personification of known (or familiar) identities is common 21,24,25 .
Affective characteristics of clinical voices primarily relate to facets of content, distress and functional interference, for which there appears to be broad consistency in patient experience. Voice content can involve a rich array of themes (including neutral or positive voices), but critical, derogatory or threatening voices are most widespread 32 , and typically have substantial negative life impacts 20,21,24 . Somatic or bodily sensations, such as tingling or burning on the skin, might also accompany these experiences 11 .
Non-clinical voices. 'Non-clinical voice-hearing' (often used interchangeably with 'healthy voice-hearing' in the literature) refers to frequent, recurring experiences of complexity and/or persistence comparable to those in clinical groups but associated with little distress and no mental health diagnosis. The lifetime prevalence of non-clinical and everyday voices has been estimated as somewhere between 5% and 15% of the general population having, at minimum, experienced one-off or fleeting voices 29 . However, prevalence estimates can differ depending on the method used and specific questions asked, or even the discipline conducting the study (for instance, psychology versus anthropology). Of note, non-clinical (relative to clinical) voice-hearing typically has an earlier age of onset, occurs less frequently, is associated with greater perceived control, involves less negative content and causes negligible emotional and functional disturbance 8,19,22,23,33,34 . Variability in other phenomenological features implies significant overlaps in these experiential facets across cohorts, with little utility for predicting mental health status 27 .
There have been calls for research to investigate how non-clinical voice-hearers can temper negative emotions and sustain functional behaviours in the face of what many would perceive as aberrant, frightful events, and whether voices may serve some adaptive function that negates the need to seek psychological care 19,33 . A minority of individuals in this category do transition to psychotic illness, but this is often underpinned by secondary precipitants, such as pre-existing anxiety or depression, or ongoing life stressors 14,18,20,35 . Many non-clinical voice-hearers report spiritual beliefs, and voices are often interpreted as communication from deceased individuals [36][37][38] .
Other research has focused on the extent to which spiritual practice and, in particular, control over the experience influence affective aspects of voice-hearing 39 . In particular, it was found that increased control was associated with reduced distress and improved functioning, although clearer delineation between discrete types of control is needed. Some spiritual practices involve cultivating and controlling voice-hearing experiences that might have started spontaneously 40 ; in other cases, voice-hearing experiences may have started as a result of such practices 38 . Examining non-clinical voice-hearing, unfettered by the confounds of broader psychopathology and medication effects, offers an opportunity to discern the fundamentals of this distinctive phenomenon, as well as its transdiagnostic operation across the psychosis continuum.

Everyday voices.
What we refer to here as 'everyday voices' typically encompass common and benign misperception of auditory events, such as the occasional fleeting, unexplained voice, or experiences involving imaginary companions. These occurrences can be sporadic, with their incidence heightened (as in psychosis) by certain internal factors such as sleep disruption 41 and stress 42 , as well as external conditions, including noisy settings or even silence 43 . Associated content is typically mundane or negligible, and there is little emotional significance attached. A possible exception lies in the case of bereavement hallucinations (perceptual experiences of a person, usually a loved one, who has passed away, potentially involving multiple senses, such as hearing that person's voice or seeing that person's image), which can be associated with psychological distress 44 .
Many everyday voice-hearing experiences include elements that appear alien to the self. For example, inner speech (covert, self-directed speech or verbal thinking), in particular its misattribution to an external source, is considered to have a significant mechanistic role in cognitive and neural models of voice-hearing. There is heterogeneity in inner speech 45 , but one replicated finding is that it is not uncommon for individuals to report hearing the voices of other people in their inner speech (for example, by endorsing an item such as "I hear other people's actual voices in my head, saying things that they have never said to me before. ") 46 .
Another example of everyday voice-hearing is 'hearing' the voices of fictional characters when reading. A study of more than 1,500 respondents predominantly from English-speaking countries (such as the UK, the USA, Australia, Canada and Ireland) found that around one in seven experienced such voices with as much vivid ness as hearing an actual person 47 . More than half of a sample of professional writers reported 'hearing' the voices of their characters 48 . A further example is the experience of engaging with imaginary companions. Although imaginary companions are usually associated with childhood, 7% of the large sample of English-speaking adults mentioned above reported currently having an imaginary companion (versus 41% having had such a companion in childhood). For almost half of participants with any experience of having an imaginary companion, engaging with such a companion involved regularly hearing the imaginary companion's voice 49 . Hearing the voices of others in the absence of any speaker thus appears unexceptional as an aspect of human inner experience.
Taken together, these phenomena are consistent with the dominant view that regular sensory perception involves a process of 'controlled hallucination' 50 : the nervous system actively predicts what is in the environment rather than passively receiving information about it. Reports of such everyday voice-hearing phenomena also align with growing evidence that hallucinations can be readily elicited in healthy individuals in experimental situations 51 .

Mechanisms of voice-hearing
Research has investigated the cognitive (Table 2), neural, personal and sociocultural mechanisms that might generate voice-hearing across the three categories of voice-hearing experiences described in the previous section. Although these mechanisms are presented here as categorically distinct, they are not entirely conceptually www.nature.com/nrpsychol separable, and typically exert multifarious influences on one another.

Cognitive mechanisms.
Although individuals with a diagnosis of schizophrenia typically show wide-ranging cognitive impairments 52 , it is difficult to establish which, if any, are specific to voice-hearing because of co-morbid symptoms and the confounding effects of antipsychotic medication 53 . One study found some impairments in executive function, working memory and verbal intelligence, but not in other cognitive measures, for non-clinical voice-hearers compared with non-voice-hearing controls 17 .
Other studies have taken a narrower, theoretically driven approach. For example, early cognitive models of voice-hearing suggested that clinical voice-hearing might result from misattributing internal mental events (such as inner speech) to an external source 54 . This externalizing bias in source monitoring was supported by evidence from a source memory task in which participants were asked to recall the origin (self or non-self) of previously presented verbal stimuli; hallucinating patients with psychosis were more likely to externally misattribute a self-generated word as not self-generated than non-hallucinating patients with psychosis or non-clinical controls 54,55 . Numerous studies have reported similar findings, with meta-analytic evidence suggesting that such external misattributions are associated with hallucinations in both clinical and non-clinical samples 55 . However, several studies have failed to replicate these findings [56][57][58] , with the discrepancy possibly caused by variability between studies, such as in the specific attributes of the tasks used. Source-monitoring (or self-monitoring) theories of voice-hearing have been linked to 'comparator model' accounts. These accounts posit that dysfunctional efference copies (internal copies of motor signals, sent between motor and sensory brain regions to predict and typically suppress sensory cortical activity) could lead to feelings of alienness or that one's own actions are not self-generated 52,59,60 . Source-monitoring theories are also consistent with accounts of hallucination proneness linked to population-wide variability in the vividness of mental imagery 61 , predicting that vivid mental imagery would be more likely to be externally misattributed.
Externalizing bias accounts of hallucinations are also supported by evidence from auditory signal detection tasks in which participants must detect speech embedded in noise (Fig. 1a). Clinical and non-clinical voice-hearers have a lower threshold for responding that Source monitoring (also self-monitoring or reality monitoring) Source memory: participants speak or hear a series of words, and must later recall which words were self-generated versus non-self-generated Voice-hearing individuals make more external misattributions, recalling self-generated words as non-self-generated Early studies showed an association between external misattributions and voice-hearing 54,55,158 ; however, more recent studies have failed to replicate this finding 56,58 Expectation/strong priors for speech Auditory signal detection: participants listen to bursts of white noise, and are asked to respond whether speech was present or absent ( Fig. 1) Voice-hearing individuals make more false alarms (respond that they heard a voice when none was present), reflecting a bias towards detection of speech Early studies showed an association between an increase in the number of false alarms and voice-hearing in psychosis and in the general population 62 . More recent work has replicated this finding, and has shown a similar effect in non-clinical voice-hearers 58,175 Conditioned hallucinations: participants learn an association between a visual stimulus and an auditory stimulus, and later have to detect the auditory stimulus Voice-hearing individuals make more false alarms in response to the auditory stimulus when cued by the associated visual stimulus There is an association between conditioned hallucinations and voice-hearing in clinical and non-clinical groups, with no effect in non-voice-hearing patients with psychosis 67 Sine-wave speech: participants listen to manipulated unrecognizable speech, and are asked to indicate whether speech is present or absent Voice-hearing individuals are more likely to detect speech in ambiguous stimuli, even before a training phase There is some evidence that clinical and non-clinical voice-hearers detect more speech than healthy non-voice-hearing groups even before training via exposure to speech samples 53,68 Memory inhibition Intentional inhibition of currently irrelevant memories: in a continuous recognition paradigm, participants are instructed to inhibit a response to a previously presented stimulus Voice-hearing individuals are less able to inhibit a previously presented stimulus, and therefore make more errors in later task blocks There is evidence for an association between lower memory inhibition and voice-hearing in clinical groups 71 and in the general population 74 . However, another study did not observe this effect in a group of non-clinical voice-hearers 58 Language lateralization, attentional control Dichotic listening: participants are presented with conflicting auditoryverbal stimuli in each ear, and must report which stimuli they heard most clearly. In subsequent trials participants are instructed to attend to one ear at a time only Voice-hearing individuals show reduced right-ear correct responses compared with controls, and are less able to orient their attention when instructed There is evidence for reduced right-ear responses in clinical voice-hearers 162 but little evidence for this effect in non-clinical or everyday voice-hearers 58,176 there was a voice in the noise compared with non-voice hearers 62 , and therefore are more likely to respond that a voice was present on noise-only trials (Fig. 1b). These results have been interpreted to suggest that biases towards responding that signals are external are associated with hallucinations. However, evidence is mixed regarding whether this response bias is accompanied by a decrease in detection sensitivity (the perceptual ability to distinguish speech signals from noise) 63 .
Although some researchers assume that biases in source monitoring and auditory signal detection tasks reflect a common underlying cognitive mechanism 55 , there is little evidence for this interpretation. Indeed, source-monitoring and signal detection explanations of voice-hearing operate, to some extent, at different explanatory levels that are not mutually exclusive 64 . Whereas source-monitoring tasks assess memory-related decision-making in relation to relevant mnemonic cues 65 , biases in auditory signal detection might reflect overweighting of top-down processes. For example, according to predictive processing accounts, the nervous system makes predictions about the external world and updates these predictions on the basis of discrepancies between these predictions and observation (prediction error). In a predictive processing framework, biases in auditory signal detection tasks might be due to overweighting expectation relative to observation 66 .
The role of top-down processes in voice-hearing is also suggested by evidence that voice-hearing participants, regardless of clinical status, are more susceptible to 'conditioned hallucinations' than non-voice-hearing participants. That is, voice-hearing participants are more likely to report hearing tones that are not present in response to a visual stimulus after an auditory stimulus and a visual stimulus are repeatedly paired 67 .
Other studies have shown that prior expectations for speech, as assessed by responses to ambiguous speech stimuli, might be present in voice-hearers even without experimental induction via conditioning 53 . These prior expectations are associated with performance in the conditioned hallucinations paradigm, such that clinical voice-hearers who were more likely to detect speech in ambiguous stimuli were also more susceptible to conditioned hallucinations 68 .
Bottom-up perceptual processes might also contribute to voice-hearing. For example, people with peripheral sensory deficits such as hearing impairment are predisposed to hallucinatory experiences 69 . Promising computational models of voice-hearing therefore take into account the integration of top-down and bottom-up signals 70 .
Finally, voices may result from failures to inhibit and place into context intrusions from episodic memory due to failures in intentional inhibition 71,72 . In typical assessments of intentional inhibition, participants complete a variant of a continuous recognition task in which, during later task blocks, they are asked to inhibit recognition responses on the basis of stimuli presented earlier in the experiment. Both voice-hearing patients with psychosis 71,72 and participants in the general population reporting relatively high levels of hallucinatory experiences 73,74 exhibit impairments in intentional inhibition, suggesting that voice-hearing might reflect uncontrolled intrusive memories. However, one study found no impairment in intentional inhibition in a group of non-clinical voice-hearers 58 . This raises the possibility that impaired performance on intentional inhibition tasks might be associated with voices only in psychosis.
Thus, as outlined in this section, a number of cognitive mechanisms have been linked to voice-hearing  In some trials, a speech stimulus is presented at a threshold level (panel 2). Participants are instructed to respond, with a button press, as to whether they think a voice was present in the noise or not present (panel 4). b | Everyday, non-clinical and clinical voices are associated with biased performance towards endorsing that a voice was present (that is, a reduced response criterion; red line) compared with non-voice-hearers (grey line). This results in more hits (correctly responding that a voice was present) and more false alarms (responding that a voice was present when it was not). Evidence is mixed regarding whether disproportionately high false alarm rates in voice-hearers are also caused by reduced sensitivity (ability to perceptually differentiate voice signals (blue curve) from noise (yellow curve)).
www.nature.com/nrpsychol in clinical and non-clinical groups. Further research is needed to clarify which cognitive mechanisms are specific to clinical (as opposed to non-clinical) voice-hearing, as well as how different mechanisms may be associated with phenomenologically different kinds of voices.
Neural mechanisms. A variety of neural mechanisms have been proposed to account for voice-hearing in clinical and non-clinical groups. Neuroimaging studies that record brain activity while clinical and non-clinical participants hear voices (known as symptom-capture studies) have shown activation in regions including the inferior frontal gyrus, superior temporal gyrus, insula and (para)hippocampus 75 during voice-hearing. These regions overlap with brain areas involved in inner speech 76 , auditory signal detection 77 and auditory imagery 78,79 .
Studies examining functional connectivity point to aberrant frontotemporal connectivity in both clinical and non-clinical voice-hearing individuals compared with non-voice-hearing controls. These findings suggest a functional model of atypical efference copy signalling between the inferior frontal speech production region and speech perception areas in the temporal lobe 80,81 . This model is supported by atypical N1 event-related potentials (a marker of suppression of activity in cortical speech perception areas) in response to self-generated (but not non-self-generated) actions in people with a diagnosis of schizophrenia 82 . However, most auditory event-related potential findings show an association with psychosis, but do not show specific associations with voice-hearing 83 .
Studies of structural connectivity confirm atypical white matter connectivity linking frontal and temporal areas, particularly in relation to the arcuate fasciculus 84 , in voice-hearers with a diagnosis of schizophrenia. However, findings regarding this white matter tract have been inconsistent 85 . Other structural neuroimaging studies find variation in patterns of brain folding in individuals with schizophrenia who experience hallucinations in different modalities 86 , and there is evidence that decreased paracingulate sulcus length (known to be significant in source monitoring) is associated with a greater susceptibility to hallucinations in patients with schizophrenia 87 . Taken together, these findings suggest that brain structure may be altered in clinical populations in ways that are specifically linked to hallucinations.
Voice-hearing is associated with atypicalities in brain networks activated when participants are not engaged in an explicit task (resting-state networks). Specifically, there are differences in patterns of interaction among the default mode network and networks linked to cognitive control and salience, as well as those relevant to specific modalities of perceptual processing 88,89 . Resting-state auditory cortex hyperactivation has been implicated in voice-hearing in both clinical 90 and non-clinical 91 groups. These findings suggest that mechanistic explanations of voice-hearing may be sought in patterns of brain activation observable even when participants are not reporting on their own hallucinations or engaged in a specific task.
Evidence for atypical frontotemporal signalling in clinical and non-clinical voice-hearers is consistent with the idea that voices result from a failure to predict the sensory consequences of one's actions 92 . Within the general predictive processing framework, such models have been understood as congruent with models of hallucination associated with an over-reliance on strong priors 93 , combined with compromised processing of prediction error, leading to non-veridical acceptance of perceptual hypotheses 94 (the brain's 'best guesses' about what is out there in the environment). Predictive-processing approaches to hallucinations also gain support from research on the role of striatal dopamine in perceptual inference 95 . For example, perceptual and reward expectations and striatal dopamine have been linked to behaviour consistent with hallucination-like experiences in mice 96 , in line with models suggesting a role for prior expectations in hallucinations, and with a large literature suggesting elevated dopamine levels in schizophrenia 97 .
Finally, right-hemisphere homologues of speechproduction regions are activated during voice-hearing 98 . Combined with behavioural findings that patients with psychosis with hallucinations do not show a typical right-ear advantage 99 (more accurate reporting of stimuli presented to the right ear than of stimuli presented to the left ear when different speech stimuli are simultaneously presented to both ears), these results implicate atypical patterns of brain lateralization, particularly relating to language functions 100 , in schizophrenia. Such atypical brain lateralization is consistent with explanations of voice-hearing as instances of auditory cortex hyperactivation and decreased cognitive control.
Research into the neural mechanisms underlying voice-hearing using a variety of methods has shown altered patterns of brain processing and structure in clinical voice-hearers. In some cases, these patterns have extended to non-clinical voice-hearers. Particularly where such findings support cognitive models, they offer hope for mechanistic explanations of the phenomenon, although there remain many gaps in current knowledge and there are many reasons to be cautious about assuming continuity of mechanism across the clinical-non-clinical divide.
Personal and sociocultural mechanisms. Certain personality traits, for instance involving heightened absorption (propensity for total immersion within one's thoughts or experiences), schizotypy (box 1) or openness to experience, are known to increase the likelihood of voice-hearing in certain cohorts 40,101 . When voice-hearing involves the experience of personified voices in interaction with the self, it can be a highly social process 102,103 . Some accounts of the social nature of voice-hearing relate to its potential origin in the misattribution of inner speech to an external source. Inner speech is recognized to have interpersonal qualities, particularly a conversational or dialogic nature, proposed to result from the internalization of linguistically mediated social interactions, such as dialogues with carers, during development 104 . Other accounts suggest that voice-hearing is the result of atypical activation of social agent representations 10 , such that voice-hearing Functional connectivity a measure of the temporal correlation of activity in different brain regions.

Structural connectivity
a group of measures of how much different parts of the brain are anatomically connected.
Default mode network a network of brain regions showing significant activation when the participant is not engaged in any task, thought to be related to self-referential processing.
experiences reflect hallucinations of communicative agents rather than simply auditory experiences. Agent representation may be particularly important for the role of social threat in distressing voice-hearing experiences 148 . Relatedly, loneliness and social isolation have been linked to voice-hearing, with voices and other psychosis-like experiences in the absence of social interaction playing a role analogous to the experience of phantom limbs in the absence of an amputated limb 105 . However, there is some evidence that only negatively appraised social exclusion ('social defeat') triggers hallucinations 106 .
The social nature of voice-hearing is also apparent in its connections to trauma and dissociation (discontinuity or separation between one's thoughts and emotions, or even facets of self-identity). Dissociation is hypothesized to exert a twofold influence between voice-hearing and trauma, via fragmentation of self components 107,108 (resulting from traumatic life events, including interpersonal losses and stressors) and as a mediator (where depersonalization explains a specific portion of the variance in voice-hearing following childhood abuse) 109,110 . The involvement of dissociation is underscored by the interpersonal nature of specific types of trauma, as well as the social relationships that some people share with their voices 103,111 . Moreover, early trauma has been shown to be a risk factor for psychosis [112][113][114][115] , and in particular hallucination severity 116 . However, other reviews have been more critical about drawing firm aetiological links between trauma and voice-hearing due to existing conceptual and methodological limitations, such as diagnostic ambiguities, consensus regarding the definition and measurement of childhood trauma, underpowered studies and the lack of longitudinal research 117,118 . Nevertheless, trauma is often reported in conjunction with voice-hearing in clinical groups 119,120 and population-based studies 121,122 .
However, closer inspection of mechanistic factors potentially related to trauma elicits more questions than answers. One model outlining possible pathways from trauma to psychosis has highlighted interactions among personal vulnerabilities and various kinds of intrusion 123 , but the empirical evidence for memory intrusions is currently limited 124 . Although certain risk factors for voice-hearing have been documented (for example, childhood sexual abuse) 121 , less is known about individual differences and protective factors (such as attachment or resilience) that may determine why some people with similar life and trauma experiences do not go on to develop voices (or vice versa).
Sociodemographic factors have also been linked to the propensity to hear voices, although it is often difficult to establish specificity to voices (as opposed to psychotic experiences more broadly). Some evidence suggests that, within psychosis, women are more likely to report voices 125,126 . Other sociodemographic factors, such as ethnicity and migration status, are well-established risk factors for psychosis 127 , but their specific contribution to voice-hearing is less clear. Several studies have linked ethnic minority status to hallucinations 128,129 ; one study indicated that migration was linked only to delusional ideation, not hallucinations 128 . Longitudinal research has found that cannabis use, childhood trauma and ethnic minority status are linked to persistence of psychotic experiences over a period of 6 years 129 , but these findings are not specific to voice-hearing.
A final set of mechanisms relate to cultural or subcultural influences. For example, the Fang ethnic group in Equatorial Guinea describe mibili as possession by evil spirits that manifests itself as auditory and visual hallucinations 130 . Cultural influences are complex and multifaceted, involving myriad interfaces along geographical, ethnic and religious lines. Cultural expectations profoundly shape the definition, prevalence, experiential facets and significance of voice-hearing, as well as levels of psychopathology and concomitant outcomes 131,132 . For example, one study found distinct differences in the phenomenological experiences of voice-hearing in the context of schizophrenia across the USA, Ghana and India 133 . The US sample mostly had an adversarial relationship with their voices, which involved themes of violence, and this was readily pathologized as indicative of mental illness; the Ghanian sample often linked their voices to a benevolent God, with whom they shared a social and positive relationship; and the Indian sample personified their voices, who were often those of kin, and these voices provided an adaptive function by dispensing advice or instruction 133 .
Most existing studies examining cultural influences on voice-hearing have relied on rudimentary

Box 1 | Schizotypy and hallucination-proneness
voice-hearing has typically been associated with a diagnosis of schizophrenia, which involves multiple positive symptoms (such as hallucinations, delusions or disorganized speech) and negative symptoms (such as social anhedonia; that is, a disinterest in and a lack of pleasure from social situations) combined with a decreased level of functioning 177 . The constructs of schizotypy and hallucination-proneness were formulated to explore the distribution of traits that may underlie psychotic-like experiences in the general population, with the assumption that they share at least some components (phenomenological, aetiological or cognitive) with experiences reported by people with schizophrenia.
Schizotypy is generally posited as a set of personality characteristics or latent traits that vary across the population, are assumed to be relatively stable over time and are possibly associated with an individual's liability to develop schizophrenia (or psychosis more broadly) 151 . Therefore, schizotypy includes the propensity to experience perceptual aberrations, magical ideation and social anhedonia, among other symptoms 178,179 . The construct of schizotypy encompasses, but is not specific to, voice-hearing. although assessments of schizotypy include items relating to perceptual aberrations, these typically form their own factor in principal component analyses, indicating that this aspect of schizotypy is separately measurable and conceptually distinct from other facets, such as social anhedonia and cognitive disorganization 178,179 .
The construct of hallucination-proneness, although strongly associated with schizotypy, attempts to assess trait-like variation in susceptibility to a broader range of perceptual experiences. The construct is more tightly focused on variation in hallucinatory experiences, and often breaks down into factors associated with specific sensory modalities (for example, auditory, visual and tactile) [180][181][182] .
The term 'hallucination' has itself been the subject of debate, and typically includes perceptual experiences that occur in the absence of external stimulation, have a compelling sense of reality, are not amenable to voluntary control and occur in the awake state 183,184 . However, some experiences typically labelled as hallucinations do not meet all of these criteria, such as experiences triggered by external sensory stimulation (for example, hearing a voice directed at oneself in a noisy crowded room) 185 and hallucinations that occur as a person is on the borders of sleep 28 . In this review, our focus is on voice-hearing experiences that, whether or not meeting the full definition of hallucination, are crucial to understanding the phenomenology and mechanisms involved in such experiences.
www.nature.com/nrpsychol divides across national borders or between Western and non-Western cultures [133][134][135][136][137][138] . The lack of nuanced considerations inherent in such divisions is problematic for the field because globalization and the mass movement of people across borders over time means that homogeneous cultural settings are becoming increasingly rare. Outside a Western medical disease model, voice-hearing may be less negatively regarded or pathologized, possibly due to fluid delineations between reality and fantasy as well as self and other 19,131 , referred to as 'porosity' 38 .
From a subcultural perspective, understanding voicehearing as a personally significant aspect of human experience has taken on political force with the growth of the international Hearing Voices Movement 139 . This grassroots advocacy initiative focuses on personal empowerment and the depathologization of voices 140 . Its emphasis on positive acceptance and individual meaning-making denotes significant departures from, but also some overlaps with, conventional views espoused in psychiatry and psychology. Other examples of subcultural or religious influences on voice-hearing include the repositioning of the hallucinatory experiences of historical figures 141,142 , or studying their attribution to jinn (invisible spirits) in Islam 143 .
Religious affinity might serve as both a risk factor and a protective factor in shaping voice-hearing 144,145 . For example, tentative evidence suggests that religious activity might serve as a coping mechanism for voicehearing in Brazil 146 , whereas in the Netherlands religious affiliation has been theorized to add to the specificity and burden of this experience 147 . Relatedly, unusual sensory experiences can be deliberately sought out in certain spiritual settings, such as indigenous use of psycho tropic agents to invoke communication with the spirit realm 131,132 or contemporary rituals performed by psychics and mediums 40,148,149 . These practices stand apart from clinical voice-hearing in that they are highly valued, often replete with vivid detail and accepted within shared cultural norms. Different pathways might therefore underlie these different voice-hearing experiences 150 . What cultural or spiritual beliefs such as mibili and jinn may have in common is a collective pursuit of meaning by adapting interpretations of anomalous perceptual events to fit the broader cultural milieu. More research is needed to understand the mechanisms underlying such experiences.
It is therefore evident that voice-hearing experiences do not occur in isolation but are shaped by factors including early-life trauma, sociodemographic and personality variables, and cultural and religious influences. The endeavour to understand voice-hearing experiences in their full complexity will benefit from situating them within the broad personal, social and cultural contexts in which they occur.

Continuity and discontinuity
In the previous section we outlined a range of cognitive, neural, personal and sociocultural mechanisms that might underlie voice-hearing experiences across clinical and non-clinical categories. A key question is the extent to which the transient and non-distressing voice-hearing experiences of individuals without need for psychiatric care lie along a continuum with recurring and distressing experiences in psychosis. According to diagnostic discontinuous models, non-clinical voice-hearing is explicable only in terms of phenomenology and processes also occurring in non-voice-hearers, and these are distinct from those underlying distressing experiences that characterize clinical voice-hearing (Fig. 2a). By contrast, the continuum hypothesis proposes that non-clinical voice-hearing is simply an attenuated version of psychotic experience, fundamentally of the same kind and perhaps reflecting an underlying continuous mechanism, phenotype or 'proneness' 1 . In continuous models, psychotic experiences might be directly related to distress and the need for treatment (Fig. 2b) or could occur independently of distress and the need for treatment (Fig. 2c).
Here we do not aim to debate whether psychotic disorders represent an extreme manifestation of a continuous trait or a latent factor in the population (such as schizotypy) 151 ; instead we focus on which specific aspects of voice-hearing might be continuous across the population in phenomenology or mechanism. A related question is the extent to which aspects of phenomenology and mechanism are linked, and via which facets of phenomenology and mechanism such linkages occur. A crucial question is whether any continuity of experience reflects common underlying cognitive and neural processes; in other words, whether continuity of frequency and pheno menology translates to continuity of mechanism. The answer to this question will be informative about whether voice-hearing in different groups should be considered as representing the same kind of experience.

Continuity of experience.
At its most basic level, voice-hearing could be continuous in frequency across the population, with some people reporting regular voice-hearing and others reporting only occasional experiences. Epidemiological data provide evidence for occasional psychotic-like experiences in the general population, with one meta-analysis suggesting a median lifetime prevalence of 7.2% (and an annual incidence of 2.5%) 151 , involving similar risk factors (such as cannabis use and stress) 127 as in psychosis. Numerous standardized self-report assessments show variability in the frequency of hallucinations in the general population at lower rates than in psychosis 152 . Although it has been argued that such scales may ask about fundamentally different experiences 153 , it is not contentious to state that some forms of voice-hearing are present in the general population, with varying frequencies.
There may also be continuous variation in aspects of the voice-hearing experience itself, such as associated distress (Fig. 2b). Recent research with non-clinical voice-hearers shows that, as in psychosis, distress can vary independently of voice-hearing frequency 22,36,37 , and scales administered in the general population show some variability in ratings of distress associated with hallucinations 154 . Others have argued that distress might be discontinuous between clinical and non-clinical populations 19,153 (Fig. 2c), appearing predominantly in those with a diagnosis of (or at risk of) psychosis.
Other aspects of the phenomenology of voices might also vary across the population. While volitional control is typically low or non-existent in psychosis, non-clinical voice-hearers often report at least some level of control 39 . Additionally, some studies have shown that personification varies within clinical and non-clinical categories 155,156 , with individuals within both groups reporting different levels of 'person-like' voices. Although no research has assessed personification of infrequent hallucinatory experiences in the general population, proneness to such experiences has been associated with the tendency to assign personality characteristics to neutral auditory-verbal stimuli 157 . These findings suggest that attributions of social agency might vary within and between groups who report voices.

Continuity of mechanism.
Evidence regarding continuity of mechanism across different forms of voice-hearing is mixed. To take an oft-cited example, voice-hearing might result from biases in source monitoring or auditory signal detection 54,55 . Evidence of such biases in voice-hearers with psychosis is often reported 62,158,159 . Signal detection biases have been found in non-clinical populations 55,57 ; however, evidence for source-monitoring biases is more mixed [55][56][57] . Evidence is also mixed regarding the roles of attention 160,161 and intentional inhibition 58,71,73,74 across clinical and non-clinical voice-hearing, with some studies reporting differences between voice-hearing and non-voice-hearing groups, and some not. One interpretation of these discrepancies is that only some cognitive mechanisms associated with voice-hearing vary continuously, whereas others are discontinuous and contribute only to some forms of voice-hearing.
The validity of the continuum hypothesis as applied to voice-hearing therefore seems to depend on the answer to the following question: a continuum of what? A related question concerns whether any such continuum has an explanatory function. Similar to debates surrounding the utility of psychiatric diagnostic practices 162 , it is important to consider whether investigating continuity is useful clinically or valid scientifically. One solution has been to suggest multiple continua across which voice-hearing might vary, such as frequency, distress or vividness, allowing an exploration of variance and covariance within and between these continua 163 . This approach could also be clinically helpful for developing treatments for people with distressing voice-hearing experiences (for example, improving detection of people at risk of developing psychosis, or more precise targeting of therapeutic options for voices). Another option is to focus less on putative continua and more on underlying commonalities: that is, factors that might not vary continuously between individuals but rather play a common causal role in voice-hearing across different kinds of experiences.

Multiple pathways to voice-hearing
In this section, we illustrate how multifactorial consideration of the varying phenomenology of voice-hearing in the everyday, non-clinical and clinical categories can generate testable hypotheses about how various building blocks underlying voice-hearing might interrelate across the range of experiences. We do not set out to present a comprehensive model of voice-hearing 164 ; there are too many gaps in current knowledge to propose a feasible integrated model linking cognitive and neural mechanisms to phenomenological features. Rather, we hope to offer a functional approach that can be used to set a research agenda to advance the field.
To this end, box 2 describes three putative instances of voice-hearing, one from each of our main categories: an individual with relatively frequent experiences of the voice of a deceased loved one in the absence of distress (everyday voice-hearing); an individual with particular spiritual beliefs within a specific cultural context (spiritualism/mediumship; non-clinical voice-hearer); and an individual hearing voices in the context of psychosis (clinical voice-hearer). These examples have been chosen to illustrate how common and distinct processes interact in generating the varied phenomenology of voice-hearing, including in complex ways that remain to be fully elucidated. An important point to be reiterated is that the overlap of phenomenology across groups is considerable, but should not be taken as necessarily entailing continuity of mechanism 163 . Indeed, we propose considerable variation in mechanism even within our groups, with differing patterns of association among top-down and bottom-up mechanisms potentially underlying different phenomenological clusters. The examples should therefore be read as indicative of how constituent processes might interrelate in line with select phenomenological facets of voice-hearing.
Comparisons across the three examples can help illustrate our arguments about continuity and discontinuity across mechanisms and phenomenological features. The subjective experience of voice-hearing in our everyday example is as vivid, clear and complex as the distressing hallucinations experienced in our clinical case. In our non-clinical example, voices are shaped by underlying cultural beliefs in a way that is less apparent in the everyday and clinical examples. Deficits in peripheral sensory systems, such as hearing loss, play a role in the everyday example but not in the non-clinical and clinical cases. To be clear, the argument is not that hearing loss always features in everyday voice-hearing and never in clinical voice-hearing; rather, it is that causal factors underly ing voice-hearing can dip in and out across groups in ways not easily described as continuous or discontinuous with phenomenology.
Another interesting point is that the predictive processing framework is implicated in all three examples but takes different forms in each. In the everyday example, there is a high expectation of hearing a particular voice. In the non-clinical example, there are stronger priors for hearing speech stimuli in all auditory stimuli 53 which are enhanced by learning and cultivation processes relevant to the subcultural context. In the clinical example, atypical processing of prediction error, in addition to strong speech priors, is implicated. Although the same general mechanism is thus involved in each example (predictive processing), the biases involved take different forms and shape phenomenology in distinct ways.
Comparison of phenomenological and mechanistic features across voice-hearing groups is instructive in other ways. Our example of non-clinical voice-hearing is particularly dependent on processes that manifest themselves at the personal-sociocultural level of explanation rather than at the neural or cognitive levels. This example highlights how paying attention to neural and cognitive processes at the expense of personal and sociocultural ones -as research on voice-hearing has tended to do -might make it more challenging to account for the range of voice-hearing experiences.
Our three examples in box 2 also shed light on ongoing debates about continuity in voice-hearing experiences. Constituent processes might be involved across the three groups with no simple gradation of mechanism and some degree of commonality in phenomenology. That said, one pattern that emerges is the increase in the number of relevant features as we progress from everyday voice-hearing

Box 2 | Case examples of everyday, non-clinical and clinical voice-hearing
Here, we provide illustrative case examples of everyday, non-clinical and clinical voice-hearing. These specific examples should not be taken to suggest that all voice-hearing within these groups can be explained in the same way, or that specific individuals only ever have experiences associated with one of these groups.
Everyday voice-hearing our case of everyday voice-hearing involves a recently bereaved older person. That person expects to hear the voice of their deceased partner (strong prior), which lowers the threshold for non-veridical perceptions. voice-hearing is intensified by vivid mental imagery, personality traits such as absorption, and sensory deprivation due to social isolation. active social representations arise from the manifestation of a known, personified agent. Phenomenologically, the single voice has a specified gender and identity, is experienced with a high degree of clarity and is localized externally within familiar environs. The origins of the voice are well understood, and its content is mostly positive or neutral.
Non-clinical voice-hearing our case of non-clinical voice-hearing involves a spiritualist who holds particular religious beliefs and who regularly engages in the cultivated practice of seeking out voices. The spiritualist expects to hear one or more voices, with increased susceptibility conferred by spontaneous hyperactivity within the auditory cortex 91 . The spiritualist has personality traits related to elevated openness to experiences, and active social agent representations take the form of known spiritual identities recognized within their specific subculture. The active pursuit of voices transpires via an altered dissociative state, which at times is aided by the deliberate consumption of psychoactive stimulants. Phenomenologically, hearing these voices occurs with high frequency, typically when they are sought out. beliefs regarding the voices' origins predominantly involve cultural and spiritual interpretations, with personification aligned to culturally sanctioned figures or spirits. Controllability is desired, and is variable, but typically improves with practice.
Clinical voice-hearing our case of clinical voice-hearing involves an individual experiencing psychosis. In addition to mechanisms that also underlie non-clinical voice-hearing (dominance of speech priors, auditory cortex hyperactivity, dissociative experiences, social agent representation and personality traits, in this case heightened schizotypy), further mechanisms significantly alter the nature of voice-hearing experiences. Cognitive processes encompass diminished inhibition of irrelevant stimuli, alongside source-monitoring difficulties. Neural mechanisms include frontotemporal dysconnectivity and elevated dopamine levels. These influences are further exacerbated by a history of childhood trauma and existing sleep disruptions. Phenomenologically, the individual frequently experiences hearing voices that cannot be controlled, with a prolonged duration per episode, and typically in third-person address. The individual has a strong conviction that the voices are real and tends to comply with command hallucinations. voices are generally experienced as intrusive, and are associated with a designated purpose. valence of content is mixed, but is often negative, involving critical or derogatory themes. emotional distress and functional interference are significant, and are accompanied by bodily sensations, such as temperature changes or tingling.
to clinical voice-hearing. This increase in the number of relevant features could suggest that clinical voice-hearing is indeed characterized by a greater number of underlying mechanisms (and hence, perhaps, phenomenological features). Alternatively, it could be that there has been less research on everyday and non-clinical voice-hearing, in which case the absence of evidence should not be taken as evidence of absence. The present state of research is not sufficient to distinguish between these two possibilities.
Owing to a lack of relevant evidence, our proposed links between mechanism and phenomenology in this section are largely speculative, and highlight areas where intensive research will be particularly valuable. Indeed, an overarching question for future research concerns the extent to which researchers might ever expect to be able to map phenomenology onto mechanism. At a minimum, researchers will need to address the complexity of interactions among processes at different levels of explanation, and how this might lead to different kinds of voice-hearing.

Summary and future directions
Voice-hearing is a diverse phenomenon experienced in some form by a significant proportion of the population (with or without mental health diagnoses), with multiple cognitive, neural, personal and sociocultural mechanisms influencing its causes, dynamic development and phenomenology. Our approach to understanding how multiple mechanisms can constellate in different patterns between and within our three groups of voice-hearers highlights that we cannot assume any simple continuity of mechanism or experience across the spectrum from everyday through non-clinical to clinical voice-hearing, even when many aspects of subjective experience are shared. Importantly, although our focus here has been on voice-hearing, many of the processes described are modality general 89 and thus likely to feature in hallucinations across other modalities (box 3). Determining which processes generalize across modalities remains an empirical question that should guide future research.
The predominant factor that determines diagnostic group membership is the occurrence of distress. Voice-hearing occurs in a variety of forms that do not significantly influence functioning, and which can even have beneficial, if not merely neutral, effects on the individual. These offer hope for advancing understanding of many aspects of typical, non-pathological psychological functioning, such as the nature of inner speech 165 and cognitive processes such as source monitoring 57 . Pinning down exactly why some voices are distressing is a complex and challenging task, and one that lies beyond the scope of this Review (see reF. 166 for a recent interdisciplinary attempt).
This Review does not cover therapeutic options for distressing voice-hearing 167,168 . However, the fact that many voice-hearing experiences are intensely distressing makes the search for further knowledge of links between phenomenology and mechanism even more pressing. If voice-hearing is as varied an experience as the research reviewed here suggests, psychological therapies will arguably be best targeted at particular subtypes of phenomenology to the extent that they can be tied to underlying mechanisms 169 . These patterns of association are likely to be highly complex, and 'one size fits all' approaches that do not consider the specificity of phenomenology and mechanism might fail to deliver therapeutic benefits. In particular, there is an urgent need to improve understanding of which early-developing manifestations of voice-hearing relate to future distress, and which remain largely benign, so that effective therapeutic interventions for distressing voice-hearing can be developed and deployed.
The co-occurrence of trauma in some voice-hearing individuals leaves several questions unanswered. Although clear associations with childhood sexual abuse exist 116,121 , explicit causal links are unverified 170 , and contradictory findings remain in relation to other forms of trauma 119,120 . Preliminary evidence supports distinct voice-hearing subtypes possibly underpinned by a trauma-related phenotype 31 . Future studies could ask how the precise nature of traumatic events aligns with voice-hearing and its phenomenological correlates 114 .

Box 3 | Hallucinations beyond the auditory
The fact that many voice-hearing experiences do not carry a strong auditory component has triggered growing interest in experiences in modalities beyond the auditory. up to 14 sensory modes have been proposed, each with its own corresponding type of hallucination 12 . Psychosis studies have disproportionately focused on voices, with a limited subset of research exploring the visual, olfactory and somatic-tactile domains. one study noted that the lifetime prevalence of hallucinations in two or more senses was double the prevalence of hallucinations in only a single modality 186 . emerging evidence has also suggested that specific patterns of clustering by sensory domain may occur [187][188][189] . Hallucinations in other modalities were twice as likely in people hearing voices, whereas this likelihood was more than 11 times higher in those who experience visual hallucinations 189 . voice-hearing co-occurs with visual hallucinations, while the latter tend instead to cluster with olfactory and somatic-tactile experiences 187 . In general, experiencing hallucinations in more than one modality is associated with more severe psychopathology and less favourable outcomes 190 . General population studies have also reported multisensory hallucinations, involving hallucinatory experiences across two or more sensory modalities, with heightened variability in sensory involvement that tends to stabilize with age 191 . multiple modalities of hallucination might be associated with more severe voice-hearing and poorer mental health in non-clinical groups 192 .
additional consideration needs to be given to the special case of multimodal hallucinations (as distinct from multisensory hallucinations), where there are known overlaps in time order and/or thematic content of hallucinatory experiences 193,194 . The auditory-visual-tactile combination might be most prevalent in psychosis 195 , with serial, rather than simultaneous, experiences more common 196 . auditory-visual hallucinations are additionally associated with elevated delusionality and negative affect 197 . Some authors have speculated that multimodal hallucinations engender higher levels of conviction and distress due to greater veracity afforded by simultaneous multisensory involvement 9,195 . However, much remains unknown, including whether multisensory hallucinations are widely shared with non-clinical and everyday voice-hearers and, more importantly, possible therapeutic implications, given the hypothesized elevation in distress.
emerging evidence suggests that multisensory hallucinations might have links to socio cultural precipitants: more traumatic events correlate with more modalities involved in subsequent hallucinations [198][199][200] . This growth of interest in modalities beyond the auditory has led to an approach to understanding hallucinations in any particular modality as representing combinations of modality-general mechanisms, such as source-monitoring processes or social agent representations, and modality-specific mechanisms, such as processes relating to peripheral sensory systems 89 . experiences in other sensory modes (including gustatory, sensed presence, kinesthetic-vestibular and autoscopic) and hypnagogic-hypnopompic hallucinations 201 have been less studied, although we do know they are experienced by the general population 191,202 .

www.nature.com/nrpsychol
Complex interactions among dissociation and other aetiological factors, such as memory and cognitive inhibition, also need further study.
In non-clinical voice-hearing, more research is needed on cognitive processes, with particular focus on attentional, inhibitory and source-monitoring abilities. Inconsistencies in the literature relating to this group, for example around paracingulate sulcus morphology 91,171 in different non-clinical samples (possibly reflecting different recruitment methods and/or inclusion criteria), might be resolved by new evidence for previously unseen phenomenological heterogeneity in such groups 163 . More research is also needed on how sociocultural factors such as cultural beliefs interact with cognitive and neural mechanisms to shape experiential facets of non-clinical voice-hearing, and what factors, including age of onset and distress, distinguish clinical and non-clinical voice-hearing groups. For all three of our voice-hearing categories, the distinct contributions of predictive processing models to mechanistic accounts remain to be specified. Such models may be particularly valuable in determining the mechanistic roles of peripheral sensory deficits (such as hearing loss) and social agent representation in everyday voice-hearing.
Another question concerns whether one would expect to see linear associations between relevant cognitive mechanisms and phenomenological features. Such linear associations might be observed only for particular mechanisms. For example, the degree of bias in auditory signal detection could correlate in a linear fashion with certain phenomenological variables, such as frequency or intensity, whereas inhibition might covary non-linearly with those same phenomenological variables.
An approach based on 'functional systems' 172 of dynamically interacting cognitive, neural and personalsociocultural mechanisms will likely prove useful for understanding complex patterns of interaction among a large number of constituent processes and implicated mechanisms. Such an approach has been proposed as a way of explaining how modality-general processes (such as source monitoring) interact with modality-specific processes (such as inner speech) in 'networks of networks' that are differentially sensitive to disruption at different stages of development 89 . Graph-theoretical and other approaches to understanding linkages among symptoms in psychopathology 173 might be valuable as research aims to elucidate the complex picture of voice-hearing across diagnostic groups. Although we have proposed a three-way categorization as a useful heuristic, future research should explore whether other taxonomies of voice-hearing may be more valuable in understanding phenomenology, mechanism and targeting of treatment.
Progress in understanding voice-hearing will depend on the willingness of the research community to embrace principles of open, reproducible, replicable research. Voice-hearing research has previously relied on small sample sizes and non-standardized measures 174 . These factors likely account for inconsistencies in research findings that are apparent across the field. Initiatives that support large, pre-registered, multisite studies, such as the International Consortium on Hallucination Research 57 , have the potential to counter some of the methodological problems that have plagued previous endeavours. As we look ahead to the next decade of research on this complex, heterogeneous and profoundly personally significant experience, there are reasons to be hopeful that the rich and varied phenomenon of voice-hearing will continue to attract the attention of researchers, with the prospect of real benefits for those distressed by their experiences.