Disorganised speech is one of the three most prominent symptoms of schizophrenia. According to the DSM-5, two symptoms out of the following must be present for at least one month (Tandon et al., 2013): delusions, hallucinations, and disorganised speech. Language use in schizophrenia shows systematic abnormalities at various linguistic levels from the single word to the connected complex utterance, suggesting its role even as a potential biomarker for schizophrenia (e.g., Covington et al., 2005; de Boer et al., 2020a,b; Voppel et al., 2021). Problems in lexical retrieval (word finding) are reflected in lower scores in verbal fluency tasks (e.g., Juhasz et al., 2012; Creyaufmüller et al., 2020) and in neologisms (involuntary creations of novel or nonsense words) and semantic paraphasias (selection of the wrong word from the same semantic domain, e.g., “cat” for “dog”). Also, the learning of new words is impaired (Tunkel et al., 2014). At the sentence/text level, narrative and pragmatic problems are found (tangentiality, derailment): persons with schizophrenia (PwS) lose track of their topic, follow chains of loose associations, repeat utterances or ignore rules of turn-taking etc. (e.g., Covington et al., 2005; Marini et al., 2008; Colle et al., 2013; Bambini et al., 2016). These problems in language productions go along with disturbances in understanding irony, humour, or metaphors (e.g., Kuperberg, 2010a; Perlini et al., 2018; Tavano et al., 2008). Syntactic problems also occur in production (incomplete sentences, simpler syntactic structures: Stassen et al., 1995; Perlini et al., 2012; Zimmerer et al., 2017) and comprehension (difficulties with complex syntactic structures; e.g., Condray et al., 2002; Kuperberg, 2010a; Perlini et al., 2012; Delvecchio et al., 2019).

Despite the relevance of linguistic deviations in schizophrenia, hallucinations are commonly regarded as a far more prototypical sign of the disease. In fact, 70–80% of PwS develop acoustic, visual and/or tactile hallucinations (Gaser et al., 2004), often subsequent to other prodromal Axis-I diagnoses (DSM-IV terminology; cf. the seminal work of the Klosterkötter group, e.g., Salokangas et al., 2012). The acoustic hallucinations are of particular relevance as they are often perceived as imperative, commenting, or dialogic voices. Conde et al. (2016) give an overview of the four major theoretical accounts and their behavioural and neuroscientific evidence: memory-based; reality-monitoring; auditory vivid-imagery; and verbal self-monitoring. These accounts draw upon the well-documented alterations in three aspects of voice processing: semantic content in the voice, identity (self vs. other) transported via the voice, and affective information contained in the voice. Each of these aspects is associated with functional alterations in different parts of the brain. Whereas the first two accounts may describe primary or even causal mechanisms, the latter might add the predominantly negative valence of the contents of auditory verbal hallucinations. For instance, de Boer et al. (2019) showed stronger effects of verbal than of non-verbal auditory hallucinations. In addition, Rossell and Boundy (2005) provided evidence for the interaction of alterations in both the semantic and the affective domain.

The identity account and the semantic account have in common the notion of a deficit in top-down processing. Auditory verbal hallucinations might be caused by a dysfunction of the speech output monitoring system, i.e., the monitoring of inner speech (Hubl et al., 2004; Vercammen et al., 2008), causing PwS to perceive their own thoughts as external voices (Allen et al., 2007; Cho and Wu, 2014; Pinheiro et al., 2019; Whitford, 2019). Hugdahl and Sommer (2018) suggested that hallucinations may lead to gating effects and/or extended refractory periods in auditory cortex neurons, causing involuntary shifts of attention. Hugdahl (2009) referred to this effect as a failure of top-down control to sensory processes. In a dichotic listening study, Hugdahl et al. (2013) demonstrated that this failure could selectively occur for the one or other ear, depending on which channel had to be attended. Along the same lines, Løberg et al. (2015) related the emergence of auditory-verbal hallucinations to systematic attention deficits (see also Waters et al., 2012). Interestingly, the connection of speaking and auditory hallucinations has in fact a neurobiological basis: the connectivity strength of speech and auditory brain areas is associated with the presence or severity of auditory-verbal hallucinations (Li et al., 2017; Zhang et al., 2018).

It is yet an open issue whether this relationship between hallucinations and disorganised speech works only one-way, i.e., whether a disorder in the speech monitor influences/causes hallucinations, or rather both ways: does the presence of auditory hallucinations have a potential disorganising impact on the speech output system? For instance, the monitoring deficit hypothesis may not fully account for the fact that PwS with auditory-verbal hallucinations have lower scores in semantic verbal fluency than those without (Siddi et al., 2017).

An empirical test of the question whether the presence of auditory-verbal hallucinations affects language production is difficult in real PwS: people with acute psychosis only show a limited tolerance to extensive neuropsychological and neurolinguistics testing. Later, on medication, their hallucinations may decline, making it difficult to trace any residual connections to language. Moreover, it might be difficult singling out the individual effects of hallucinations from all the other symptoms in order to assess their selective influence on speaking. One potential solution to this problem is the use of mimicking/simulation paradigms in healthy volunteers in which states are induced that bear a relevant resemblance to those of PwS in their acute phase of a disease (for a review on schizophrenia cf. Ando et al., 2011; for other mimicking studies cf. the review by Heim, 2013). Such mimicking/simulation accounts have demonstrated their usefulness in modelling the behavioural and neural mechanisms underlying developmental dyslexia (Tholen et al., 2011; Heim et al., 2014) and aphasia (Meffert et al., 2011; Grande et al., 2012). Even though this approach has limitations, in particular that the complex constellation of a disorder is at best approximated but never fully realised, there are several potential advantages: larger, well-described samples with clearly defined inclusion criteria can be recruited systematically; the participants can endure longer testing sessions at lower stress levels; individual factors can be isolated and manipulated individually, allowing the systematic experimental control with increasing complexity; and replication studies are easier feasible. A mimicking/simulation account cannot provide a final truth, but it allows pursuing or generating hypotheses to be tested in real PwS who were spared from participating in all the trials and errors in the piloting phases.

Importantly, in the field of schizophrenia, several such studies simulating/mimicking visual (Rastelli et al., 2022) and predominantly auditory hallucinations (Brown, 2010; Ando et al., 2011; Evans et al., 2015; Kidd et al., 2015; Kepler et al., 2016; Skoy et al., 2016; Silva et al., 2017; Kim and Wojnar, 2019; Riches et al., 2018, 2019; Duprey et al., 2021) were published in the past years. The major goal was to raise awareness and empathy for the particular situation of PwS who suffer from such hallucinations. Simulations were applied to students of different health professions, and their self-reports on their experiences and also effects on behavioural performance measures were assessed. For instance, in the study of Silva et al. (2017), auditory-verbal hallucinations were mimicked based on transcripts of reports from schizophrenia patients. Two actors recorded audio tracks which were presented to healthy volunteers. Real human voices instead of computer-generated stimuli were used to increase the authenticity. Kim and Wojnar (2019) obtained data for disturbance in the domains of attention (alertness, concentration, train of thoughts) and memory (memorise and recall), for increase of unpleasant emotions and somatic changes (e.g., racing heart, shallow breath, higher blood pressure and heart rate), and odd behaviour (e.g., fidgeting, nail biting, fiddling with hair and clothes, repeated time checking, inappropriate laughing, swearing and self-talking). In fact, for 90% of their participants, there were changes in their degree of functioning and performance. The authors thus concluded that these mimicked hallucinations “closely resembled the voice hearers’ actual experience” (Abstract, p: 240). “Conceivably students may have reacted to the voice-hearing experience in a manner similar to patients’ experiences of post-psychosis emotional disturbance (Birchwood, 2003). […] The distinctive difference between patients and students was the use of self-reminders and reassurances by faculty that VHS would end soon and the option of stopping the voices was available at any time.” (p. 245). Diederen et al. (2012) investigated the neural similarity of mimicked hallucinations, actual hearing under auditory stimulation, and imagination of voices using positron emission tomography. The activation patterns for hallucinations and auditory stimulation were very similar to each other and quite distinct from imagination.

Whereas the goal of these simulation studies was to increase empathy in unaffected people (e.g., in the education of nurses or pharmacists) with the situation of PwS, Kim and Wojnar (2019) also reported a wide range of behavioural effects. Thus, the procedure might likewise be used to investigate the selective effects of the presence of commenting, imperative or dialogue-like voices on verbal behaviour, i.e., the speech-output system in healthy volunteers. Consequently, the present study used this approach, which in consequence may contribute to generating novel hypotheses about the potential loci in the cognitive system at which real hallucinations in real PwS might act. The paradigm is thus an extension of the well-established picture-word interference task in which the presence of distractor stimuli may alter the speed or accuracy of the intended utterance (e.g., Klaus et al., 2017 or Creyaufmüller et al., 2020 for application in both healthy persons and PwS). It is feasible and valid on the grounds that PwS with auditory-verbal hallucinations fail to distinguish external from internal voices, at least in those stages of the disease in which the PwS have no sense of distance to the hallucinations. Moreover, if it mimics disorders in the speech output, these effects will go beyond the standard Lombard effect (Lombard, 1911) that speakers raise their voice when in noisy environments, because here noise serves as the baseline. In the present study, all speech output levels from single-word retrieval (verbal fluency, picture naming) to sentence production and discourse were addressed, thus offering a systematic and comprehensive investigation of the effect.

The study was approved by the local ethics committee at the Medical Faculty, RWTH Aachen University (EK080/19).

The average performance on each of the parameters in the different tasks and conditions is displayed in Figure 1. The results of the 2 × 2 ANOVAs with factors Condition and Order are summarised in Table 1.

The set of 2 × 2 ANOVAs for verbal fluency revealed a main effects of Condition in the switch but not in the simple condition. Main effects of Order were not significant. The interaction term was significant for the Simple but not for the Switch condition. Fewer words were generated under the mimicked hallucinations than under noise.

The follow-up 2 × 2 × 2 ANOVA combining the two analyses for the Simple and Switch variants of verbal fluency into one large analysis yielded the following findings: In line with the differential pattern in the preceding analyses of the individual tasks, there was a significant interaction of Task and Condition (F[1;38] = 5.100, p = 0.030). The other effects were as follows. There was a main effect for Task (F[1;38] = 28.930, p < 0.001) but not for Order (F[1;38] = 0.325, p = 0.572) or Condition (F[1;38] = 2.177, p = 0.148). The interaction of Order with Task was significant (F[1;38] = 51.557, p < 0.001), but not the interaction of Order with Condition (F[1;38] = 0.061, p = 0.806). Finally, the 3-way interaction was significant (F[1;38] = 35.916, p < 0.001).

The set of 2 × 2 ANOVAs for picture naming revealed no significant main effects of Condition A main effects of Order was found for Accuracy and Duration, but not for Latency. Interaction terms were not significant. As accuracy did not differ between the two conditions, no additional qualitative analysis of the individual speech errors was performed.

The set of 2 × 2 ANOVAs for sentence production revealed a main effect of Condition for Duration but not for Accuracy or Latency. Main effects of Order were not significant. Interaction terms significant for Accuracy but not for Latency or Duration. Sentences under auditory-verbal hallucinations were spoken more quickly than in the noise condition. Again, as accuracy did not differ between the two conditions, no additional qualitative analysis of the individual speech errors was performed.

The set of 2 × 2 ANOVAs for discourse revealed no main effects of Condition or Order. The interaction term was significant for TIU but not for LIU.

Finally, a Pearson correlation analysis looked into the potential relationship of the significant Condition effects for the verbal fluency Switch condition and the sentence production Duration. The difference effects for Condition (hallucination vs. noise) for verbal fluency Switch and sentence production Duration were not correlated (r = 0.189; p = 0.243). Looking at the individual correlation effects of each simulation variant (hallucination, noise) in each task (fluency Switch, fluency Simple), a significant correlation of the sentence duration under both conditions emerged (r = 0.611; p < 0.001; Figure 2). The other correlations were not significant (sentence_Duration_hallucination × fluency_Switch_hallucination: r = −0.206; p = 0.203; sentence_Duration_hallucination × fluency_Switch_noise: r = 0.001; p = 0.996; fluency_Switch_hallucination × fluency_Switch_noise: r = 0.240; p = 0.136). Please note that, even though no value of p correction for multiple comparisons was applied, the one significant effect would remain significant even after a strict Bonferroni correction.

The present study investigated the influence of mimicked auditory-verbal hallucinations (vs. noise) on different aspects of language production at the word, sentence, and text level. There was a significant influence of mimicked auditory-verbal hallucinations on single word production only in the most difficult task (fluency, switch condition). The distinction of the two fluency variants was confirmed by the significant interaction effect of Task and Condition. Moreover, sentences under mimicked hallucinations were spoken more rapidly than under noise. These effects of the simulation at word vs. sentence level seem to be of differential nature as there was no correlation whatsoever across linguistic levels.

These results indicate that the presence of mimicked auditory-verbal hallucinations, i.e., a linguistically meaningful source of distraction, can exert influence on the language production process over and above that of mere background noise, as in the Lombard effect which predominantly concerns the loudness and perhaps pitch of the speaker’s voice. Moreover, the overall pattern of results is largely in line with reports in the literature about the linguistic performance of real PwS with auditory-verbal hallucinations. For single word production, there is ample evidence for deficits in verbal fluency (e.g., Hintze et al. 2022; Wei et al., 2022; for a review cf. the meta-analysis by Doughty and Done, 2009). These verbal fluency deficits must be distinguished from a rather fair performance in simpler tasks of lexical retrieval such as picture naming (e.g., Al-Uzri et al., 2004; Tan and Rossell, 2017; but see Vogel et al., 2009). Thus, rather than assuming the locus of the effect (only) in lexical access per se (e.g., Heim, 2020), a link to other cognitive functions such as executive functions (e.g., working memory; processing speed) must be considered as potential causes for these specific deficits in schizophrenia (Ojeda et al., 2008; Berberian et al., 2016), as these functions are associated with verbal fluency and cognitive reserve in neurotypical persons and (at least in male) PwS (Marsh et al., 2019; Kubota et al., 2022). Also, the dissociation of verbal fluency deficits but no speech-onset latency effects in sentence production is in line with observations in real PwS vs. healthy controls (Creyaufmüller et al., 2020). Finally, the increases talking speed under mimicked hallucinations in the present study could be associated with reports of logorrhoea and concurrent hallucinations (Lee, 2004), a set of symptoms responding to the dosage of Lorazepam.

To conclude, the simulation paradigm employed here caused specific effects with the following properties: (1) They go beyond those of the presence of white noise; (2) They reflect findings in the literature about symptoms in real auditory-verbal hallucinations; (3) Over and above these reports in the literature, the effects observed here have a clear direction of causality: the auditory stimulation caused the speech effects and not vice versa (as was previously investigated in studies looking for the source of hallucinations in the impaired speech-output monitor).

The simulation applied here did not cause any further deficits in the tasks of connected speech (sentence production, discourse). This lack of effects was unexpected (cf. e.g., the systematic study by Vogel et al., 2009 that did report more errors in a sentence production task for PwS). Also, the differential pattern for deficits in the switch condition of the fluency task but not in the simple condition is not immediately reflecting the state-of-the-art: PwS tend to be affected already in the simpler version of the task. However, they are certainly affected in the version with category switches (Vogel et al., 2009; Creyaufmüller et al., 2020). For the sentence condition, one explanation could be that the type of sentences was too simple to be sensitive to the simulation, as it were only main clauses of the S-P-O type. Very little syntactic planning was necessary here – perhaps too little (Moro et al., 2015; see Creyaufmüller et al., 2020 for similiarly simple sentences with similar effects). Likewise, in the discourse task, the challenge may have been too easy since all pictures were presented for the full amount of time, and in the correct order: very much structure was provided so that the lexical and thematic processing (which was analysed here in terms of LIU and TIU) had enough context (see Kuperberg, 2010b) on the role and relevance of context for linguistic processing in schizophrenia.

This, however, is exactly the advantage of a simulation study, as outlined in the introduction: one can derive novel hypotheses which can be tested with healthy volunteers and, later, with people with schizophrenia. Two alternative hypotheses present themselves: (1) The mimicking method is not strong enough to produce meaningful effects beyond the single word level. – (2) Auditory-verbal hallucinations have a differential impact on language processing at the different linguistic levels: they interfere with lexical retrieval under cognitively demanding conditions but are less relevant for connected speech (leading to the next question which other pathological mechanisms, then, are interfering at the level of connected speech). Both alternatives can be pursued, (1) by using less structured materials and different methods of speech elicitation, and (2) by systematically juxtaposing the statistical effect of severity of auditory-verbal hallucinations in PwS with other carefully obtained and well-structured clinical scales and measures of cognitive and linguistic functioning. Such insights might help formulate more complex models of language processing that take into account processing levels beyond (psycho-)linguistics.

In any case, this study is among the first to suggest a potential causal link between auditory-verbal hallucinations and (a part of the) linguistic problems in language production in people with schizophrenia instead of the other way around. With the first preliminary knowledge obtained here, more far-reaching future studies, both behaviourally and in combination with neuroimaging, become feasible. Kuperberg (2010b, p: 600) asked, “Is it possible to neuroanatomically dissociate language networks engaged in different types of psycholinguistic operations? How and at what stage do these networks interact? Are these neural mechanisms reciprocally linked such that an over-engagement of one network is necessarily accompanied by an under-engagement of another? The study of language processing in a widespread functional disorder such as schizophrenia may help us begin to address some of these questions.”

In future studies, one aspect which might be considered is that of a potential habituation to the auditory stimulation during the tasks. We controlled for order effects by counter-balancing the occurrence of the mimicked hallucinations and that of white noise across participants. Statistically, the order effects were prominent in many cases (even though the actual experimental manipulation was strong enough to cause detectable effects). Also, subsequent studies might skip the picture naming task and perhaps start immediately with those tasks requiring longer and more complex connected speech, thus taking advantage of the simulation while habituation is still low. Finally, future studies might assess the degree of proneness to auditory hallucinations, as this may significantly modulate the amount to which the presence of hallucination-like stimuli actually affects behavior (cf. the recent study by Laloyaux et al., 2022).

Based on earlier evidence for the usefulness of employing simulations to model language performance in patient samples, this study demonstrates that mimicking auditory-verbal hallucinations provokes language symptoms comparable to those reported in the literature. A refinement of the methods is suggested. In future, the simulation approach might help piloting studies with real patient samples, allowing more precise manipulations and thus increasing the impact of clinical studies with real PwS. In particular, the simulation may help formulating more detailed hypotheses about the (neuro-)cognitive foundations of the first-rank symptoms in schizophrenia and their interplay with the human cognitive apparatus.

For clinical purposes, the present findings might have implications for speech-language therapy and/or cognitive-behavioural therapy. For instance, they might inspire the development of meta-cognitive strategies of control over one’s own utterances (including semantic content and syntactic structure) in the presence of distracting auditory verbal hallucinations: these could be tested and evaluated in neurotypical participants hearing mimicked auditory-verbal hallucinations before application to PwS. Success in the management of one’s own voice and communication behaviour might be one key feature to interrupt the vicious circle in which PwS find themselves trapped (Heim, 2020): neurobiological alterations in the brain affect directly and indirectly (via auditory-verbal hallucinations) communication success, both in the personal life (reduced social participation) and in the psychotherapeutic setting (confuse or diffuse interaction with the therapist). Even if the neurobiological basis cannot be changed pharmaceutically, or if this change progresses slowly, improved control over one’s utterances despite hallucinations might sooner lead to more efficient communication, and hence improved participation and therapeutical exchange and, in consequence, quality of life (Joyal et al., 2016; Heim, 2020). “The field of SLT can contribute at characterizing speech and language impairments across life span of patients with schizophrenia in a detailed manner, as well as identifying and developing efficient approaches to treat speech and language impairments with evidence based data.” (Joyal et al., 2016, p. 93). In their review, Joyal et al. (2016) demonstrated the potential usefulness of various therapeutic approaches for improving language and communication skills, in particular pragmatic skills, in PwS, indicating that language abilities can really be modified by interventions. At the same time, the large heterogeneity of these approaches (operant conditioning; meta-comprehension/meta-learning; cognitive therapy; rehabilitative approaches), their settings (group or individual), and their parameters (duration of session: 15–90 min; total duration of intervention: 2 weeks up to 2 years; frequency of therapy 1x per week up to 2x per day) also illustrates that there is a large need for setting standards and for model-guided interventions. In a most recent RCT, Bambini et al. (2022) demonstrated that their novel approach PragmaCom, which included both production and comprehension elements, improved pragmatic communication skills of PwS after 12 weeks of training and also after a 3-months follow-up. As such RCTs can only be run after sufficient piloting, the paradigm used in the present study might prove useful for such piloting of intervention strategies in neurotypical persons before the most promising ones are actually applied to PwS, thus improving their chances to benefit from the trial they enrol in.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by Ethikkommission an der Medizinischen Fakultät der RWTH Aachen. The patients/participants provided their written informed consent to participate in this study.

SH: study concept and design, supervision of the creation of materials/stimuli, data analysis and supervision of data analysis, discussion and manuscript. SP: creation of materials/stimuli, recruitment of participants, data acquisition, data analysis, and contribution to and revision of the manuscript. KH: study concept and design, supervision of the creation of materials/stimuli, supervision of the linguistic aspects of data analysis, discussion of statistical data analysis and results, and revisions of the manuscript. All authors contributed to the article and approved the submitted version.

This work is funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – 491111487.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.