44.1 Phonetic Adaptation in Interactive Language Use
It is a truism that speech is a means of shared communicative exchange. Yet, neurocognitive accounts of (adult) speech production have often adopted a unidirectional perspective, modeling speakers as isolated agents (Guenther, Reference Guenther2016; Hickok, Reference Hickok2012; Parrell et al., Reference Parrell, Lammert, Ciccarelli and Quatieri2019). In these models, adult speech is described as a closed, fully developed system of internal feedforward and feedback control mechanisms operating on largely invariant motor and sensory phonological systems (Hickok, Reference Hickok2012) or speech sound maps (Guenther, Reference Guenther2016) acquired during childhood. Feedback control processes in these accounts are designed to ensure that the motor plans or programs, once established, “stay tuned” over the course of a lifetime, despite noisy motor implementation or changes in vocal tract anatomy (Guenther, Reference Guenther2016, p. 108; Parrell et al., Reference Parrell, Lammert, Ciccarelli and Quatieri2019). In the absence of modules mediating auditory input from other speakers in these single-person models, a speaker’s internal sensory targets are sealed from existing phonetic variation in their language community and remain unchanged throughout life.
However, phonetic plasticity, that is, the gradual adaptation of sound patterns to others’ speech, does not end with the completion of language acquisition in adolescence, as even adult speakers are known to adjust their pronunciation to the speech of others in their environment. Evidence for this is based on observations discussed in terms such as phonetic entrainment, convergence, alignment, accommodation, imitation, or adaptation (Wynn and Borrie, Reference Wynn and Borrie2022). Phonetic adaptation occurs at different temporal and social scales, that is, at the most proximal scale of adjacent conversational turns (Pardo, Reference Pardo2006), across short periods of exposure to other dialects (Delvaux and Soquet, Reference Delvaux and Soquet2007), over periods of a few months in contained groups of individuals (Harrington et al., Reference Harrington, Gubian, Stevens and Schiel2019a), over a lifetime in individuals who are exposed to new dialectal variants (Harrington et al., Reference Harrington, Palethorpe and Watson2000; Riverin-Coutlée and Harrington, Reference Riverin-Coutlée and Harrington2022), or as manifestations of ongoing gradual sound change spreading within social groups or regional communities (e.g., Bukmaier et al., Reference Bukmaier, Harrington and Kleber2014; Labov, Reference Labov2014). Finally, at the most distal end, there is the diachronic sound change languages undergo over centuries, which can ultimately be traced back to the perception–production dynamics of interactive language use that trigger distinctive phonological alternations through propagation and phonologization of gradual phonetic changes (Harrington et al., Reference Harrington, Kleber, Reubold, Katz and Assmann2019b).
Apart from phonetic convergence, there is a second widely discussed paradigm of interactive language use, that is, the moment-to-moment temporal coordination of interlocutors (Levinson, Reference Levinson2016). Speakers engaged in a conversation almost universally tend to avoid overlapping talk and strive to minimize silent gaps between conversational turns (Stivers et al., Reference Stivers, Enfield and Brown2009). Of note, turn transition times are typically much shorter than the time required for speech planning, which implies that perception and comprehension of incoming speech overlaps and is interwoven with the linguistic and motor processes involved in preparing a response (Levinson and Torreira, Reference Levinson and Torreira2015). Another important turn-taking issue is to understand how speakers can predict the end of an incoming turn precisely enough to take their own turn with such short delays.
As a summary, there is a large body of linguistic data that cannot be explained by a single-speaker approach to speech motor control and therefore call for an expansion of speech production models to account for the impact of external input from other speakers during conversation (Bradshaw and McGettigan, Reference Bradshaw and McGettigan2021; Sato et al., Reference Sato, Grabski and Garnier2013). Furthermore, the neural underpinnings and the clinical aspects of speech production in interactive contexts are still under-researched, calling for a “second-person” approach in neurophonetics.
This chapter will focus mainly on phonetic aspects of between-speaker adaptation, although most of the reported evidence is not confined to the phonetic level but also includes other linguistic domains. We first outline some of the prevailing models of interactive speech (Section 44.2) and then discuss assumptions about underlying neural mechanisms using data from neuroimaging studies in neurotypical participants (Section 44.3) and from clinical populations (Section 44.4). A particular focus is on speech rhythm, which plays a crucial role in mediating inter-speaker temporal alignment (see Chapter 29) and may itself be an object of implicit gradual convergence.
44.2 Behavioral Mechanisms and Models
44.2.1 Phonetic Mechanisms
The core of the phonetic changes that occur across different temporal and social/regional scales are the individual episodes of auditory–motor interaction of speakers during conversations. Exemplar theory assumes that speakers taking part in conversations are exposed to the phonetic variation that exists in a community and thereby continuously update the knowledge they have of their language (Blevins, Reference Blevins2004; Harrington et al., Reference Harrington, Kleber, Reubold, Katz and Assmann2019b). Like in other domains of motor action (e.g., Cracco et al., Reference Cracco, Bardi and Desmet2018), there is a tendency also in speech to covertly imitate others’ actions, leading to alignment of fine phonetic details of interlocutors’ speech in conversational interaction (Garrod and Pickering, Reference Garrod and Pickering2009). Implicit imitation may include virtually all phonetic aspects, for example, vocal pitch (Bradshaw and McGettigan, Reference Bradshaw and McGettigan2021), vowel quality (Delvaux and Soquet, Reference Delvaux and Soquet2007), plosive consonant articulation (Shockley et al., Reference Shockley, Sabadini and Fowler2004), speech rate (Schultz et al., Reference Schultz, O’Brien and Phillips2016), or prosody (Weise et al., Reference Weise, Levitan, Hirschberg and Levitan2019), especially speech rhythm (Polyanskaya et al., Reference Polyanskaya, Samuel and Ordin2019). Changes are usually subtle and only measurable using acoustic parameters or sensitive perceptual paradigms.
The degree of phonetic adaptation depends on the type of interaction. Experimental paradigms of joint speech have often been based on semi-interactive laboratory tasks, such as repetition or close shadowing of prerecorded speech, and it is still unclear if phonetic convergence in naturalistic interactions and in such experimental settings rely on the same mechanisms (Pardo et al., Reference Pardo, Urmanche and Wilman2018). Other factors reported to play a role are social proximity and preference (e.g., Babel, Reference Babel2012; see below). Most importantly, the time and frequency of exposure to a specific phonetic variant and the density of interactions in a community have an impact on how salient and persistent a phonetic change will be (Blevins, Reference Blevins2004). Agent-based computational modeling has been used to simulate the propagation of phonetic change and predict the macroscopic changes that emerge from subtle phonetic variation within an ensemble of interacting agents when phonetic or social variables are controlled experimentally (for a recent overview, see Harrington et al., Reference Harrington, Kleber, Reubold, Katz and Assmann2019b).
44.2.2 Social versus Cognitive Models
As outlined above, there is evidence that the degree to which interlocutors involved in a conversation converge or diverge in their pronunciation may depend on their social closeness or mutual liking (e.g., Pardo et al., Reference Pardo, Gibbons, Suppes and Krauss2012; for an overview and discussion, see Ruch et al., Reference Ruch, Zürcher and Burkart2018). Phonetic aspects of pronunciation interact with social patterns and often have a social meaning, which has led several authors to identify social pressure as a driving force of sound change (Eckert, Reference Eckert2012; Labov, Reference Labov1963; Polyanskaya et al., Reference Polyanskaya, Samuel and Ordin2019). Based on such evidence, communication accommodation theory (CAT) (Giles and Ogay, Reference Giles, Ogay, Whaley and Samter2007; Giles et al., Reference Giles, Taylor and Bourhis1973) considers linguistic convergence or divergence as specific communication strategies that speakers use to signal their attitudes towards other individuals and create, maintain, or decrease social distance in interaction.
In a contrasting approach, Pickering and Garrod (Reference Pickering and Garrod2004, Reference Pickering and Garrod2013) postulated that speaker accommodation in conversation is driven by implicit cognitive processes rather than explicit social strategies, with prediction and forward modeling as the core mechanisms conveying adaptation. In their interactive alignment model of speaker–listener interactions, conversation is understood as a production-comprehension process in which speakers compute forward models to predict and possibly adjust their planned utterances to the perceptual needs of their interlocutors, and listeners in turn covertly imitate the speakers’ utterances and derive their own forward models to predict what their counterparts will shortly produce. Thus, in conversation we are implicitly modeling our interlocutors who, in turn, are modeling us in a reciprocal exchange of sensory signals (Friston and Frith, Reference Friston and Frith2015). This interaction is considered to reduce the computational processing load and facilitate the mutual understanding of speakers/listeners and, at the same time, lead to a convergence at semantic, syntactic, lexical, and phonetic-phonological levels and a temporal alignment of their interaction. In phonetic terms, the perceptual forward model that a listener creates of an interlocutor’s utterance is shaped by phonetic details of that utterance, which can then merge with the motor forward model listeners compute for their own response and so lead to phonetic convergence (for discussions, see Bradshaw and McGettigan, Reference Bradshaw and McGettigan2021; Ruch et al., Reference Ruch, Zürcher and Burkart2018).
44.2.3 The Role of Speech Rhythm
Rhythmic stimuli are considered to offer an invitation – or even constitute an affordance – for perceivers to coordinate their behavior with the stimulus, as in dancing or clapping one’s hands to a piece of music (“rhythmic entrainment”; Cummins, Reference Cummins2009). The regular recurrence of a beat facilitates sensorimotor predictions about the occurrence of subsequent events and thereby allows individuals to synchronize their actions with the stimulus. Neural entrainment theories assume that alignment to a rhythmic stimulus is mediated by endogenous oscillatory brain activity at a frequency corresponding to the frequency of the stimulus (Lakatos et al., Reference Lakatos, Gross and Thut2019). Speech, with its recurring sonority peaks of syllabic structure and its alternations between stressed and unstressed syllables, has a quasi-rhythmic envelope that, in line with neural entrainment assumptions, evokes an oscillatory auditory cortical response in the delta (1–3 Hz) and theta range (4–8 Hz) coupled with the frequency modulation of the speech signal (Lakatos et al., Reference Lakatos, Gross and Thut2019; see Chapter 3; for a critical view, see Oganian et al., Reference Oganian, Kojima and Breska2023). Thus, in conversational interactions, endogenous oscillators in the brains of the speaker and the listener are assumed to become entrained to the speaker’s prosodic (delta) and syllabic (theta) rhythm. This rhythmic speech–brain entrainment promotes speech understanding (Riecke et al., Reference Riecke, Formisano, Sorger, Başkent and Gaudrain2018) and enables perceivers to predict upcoming linguistic content (Kösem et al., Reference Kösem, Bosker and Takashima2018). Likewise, it allows listeners to predict the end of the speaker’s turn and prepare for their own seamless turn-taking (Wilson and Wilson, Reference Wilson and Wilson2005). Domain-general mechanisms of reaction speed enhancement through expectancy-driven delta oscillations, as described by Stefanics et al. (Reference Stefanics, Hangya and Hernádi2010), may contribute to the smoothness of turn transitions and regulate the perception–action cycle of conversational interactions (see Chapter 6). As hypothesized in Pickering and Garrod’s interactive alignment model, phonetic adaptation during conversation emerges through mutual covert imitation and internal forward modeling of each other’s speech, interwoven with the speaker’s forward modeling of their own speech (Pickering and Garrod, Reference Pickering and Garrod2013). On this background, rhythmic speech–brain entrainment can be understood as a neural mechanism that establishes temporal synchrony between the speakers’ and listeners’ forward models, thereby strengthening the coupling of production and comprehension.
44.3 Neural Mechanisms
The hypothesis that listeners create forward models of a speaker’s utterances was addressed in a seminal multi-brain imaging study by Stephens et al. (Reference Stephens, Silbert and Hasson2010). In this investigation, the functional magnetic resonance imaging (fMRI) activities of individuals listening to a story were modeled, off-line, with the blood-oxygenation-level-dependent (BOLD) signal of the speaker telling the story. Stephens et al. (Reference Stephens, Silbert and Hasson2010) found that listeners’ brain activity was spatially and temporally coupled with the speaker’s activity, predominantly with a delay, but to some extent also anticipating it. Interestingly, the degree of similarity and predictive anticipatory coupling with the storyteller’s brain activation was correlated with the accuracy of story comprehension, supporting the concept that listeners’ forward modeling of a talker’s utterances facilitates their comprehension (Pickering and Garrod, Reference Pickering and Garrod2004, Reference Pickering and Garrod2013).
Since then, several other multi-brain studies of speaker–listener interactions have been conducted (e.g., Dikker et al., Reference Dikker, Silbert, Hasson and Zevin2014; Silbert et al., Reference Silbert, Honey, Simony, Poeppel and Hasson2014). Interactive alignment theory predicts that speaker–listener coupling can involve different processing levels, from phonetic to syntactic and lexical-semantic, which should then be reflected in the spatial pattern of between-brain alignment. This prediction was confirmed by Dikker et al. (Reference Dikker, Silbert, Hasson and Zevin2014), who used a paradigm where listener accommodation was specifically dependent on lexical-semantic factors. These authors found brain-to-brain synchrony predominantly in left posterior superior-temporal gyrus, which they interpreted as a sign of speaker–listener alignment at the lexical-semantic level. For a review of this literature, see Schoot et al. (Reference Schoot, Hagoort and Segaert2016).
The studies outlined so far have focused on how listeners’ brain activation resonates with a speaker’s activation patterns, as the basis for linguistic convergence that may unfold between the two. However, they disregard if adaptation actually occurs and give no specific account of the brain areas involved in the imitation of aspects of another speaker’s phonetic repertoire. In fact, only few studies have so far focused on the neural underpinnings of phonetic (or, more generally, linguistic) adaptation to others’ speech. Peschke et al. (Reference Peschke, Ziegler, Kappes and Baumgaertner2009) used a close-shadowing task requiring participants to overtly repeat pseudowords produced by model speakers, with the instruction to start speaking at the shortest possible delay. Adaptation to stimulus fundamental frequency (F0) and stimulus duration (i.e., speaking rate) was examined in 20 participants who performed the task in an fMRI scanner. Significant brain activation was found for the degree of speech rate imitation, but not for pitch imitation. Greater imitation of rate was associated with greater activation in right inferior-parietal cortex near the temporo-parietal junction. In one explanation, this result was interpreted as a sign of higher auditory attention in participants with a greater tendency to imitate, rather than as a correlate of imitation per se. In an alternative explanation, the observed right temporo-parietal activation was considered to reflect automatic processing of paralinguistic details of a speech stimulus, such as speech rate (Peschke et al., Reference Peschke, Ziegler, Kappes and Baumgaertner2009). A later fMRI study by Garnier et al. (Reference Garnier, Lamalle and Sato2013) studied F0 imitation in a vowel production task and found a similar, though bilateral, activation of inferior-parietal and posterior superior-temporal regions. However, in this study, brain activation was inversely correlated with the degree of adaptation. Although these data point at a role of temporo-parietal cortex in the adaptation to others’ speech, there are inconsistencies that may be due to differences in the methods used to elicit imitation, that is, production of isolated vowels versus shadowing of single words.
In the reported paradigms of isolated word or vowel production, the quasi-rhythmic pattern of connected speech played no role, which may have diminished the participants’ propensity for entrainment and phonetic adaptation to the model speakers. As outlined in Section 44.2.3, neural entrainment to the slow modulation of the speech envelope constitutes a core mechanism facilitating the prediction of upcoming input. Giraud and Poeppel (Reference Giraud and Poeppel2012) hypothesize that the sampling of quasi-rhythmic speech information by neural oscillations is asymmetric to the extent that theta and delta oscillations parsing timescales of syllable and larger size predominates in the right auditory cortex, while neural computations across timescales encompassing segmental or sub-segmental information rely on high-frequency (“gamma”) oscillations dominating in left auditory cortex (Poeppel, Reference Poeppel2003). A study by Park et al. (Reference Park, Ince, Schyns, Thut and Gross2015) revealed that speech–brain coupling is enhanced by top-down information from left motor and premotor areas, causing a lateralization of speech-induced neural entrainment to the left superior-temporal lobe and a “brain–brain coupling” between auditory and speech motor regions. This coupling is highly speech-specific to the extent that it is restricted to a relatively narrow frequency range around ca. 4.5 Hz, which corresponds to mean syllabic speech rate across languages (Assaneo and Poeppel, Reference Assaneo and Poeppel2018). For an overview of the literature on neural oscillations in speech processing, see Meyer (Reference Meyer2018) and Chapter 3.
In accordance with the idea of a modulating influence of anterior language areas on auditory processing, Scott et al. (Reference Scott, McGettigan and Eisner2009) argued for a specific role of motor cortex in conversation, especially in tracking the speech rate and rhythm of the current talker and thereby organizing smooth turn transitions and interactive alignment. In a more recent study, Castellucci et al. (Reference Castellucci, Kovach, Howard, Greenlee and Long2022) used intracranial electrocorticography in patient volunteers undergoing surgical treatment to investigate the neural dynamics underlying the specific planning processes that enable rapid turn-taking in interactive speech. They identified a frontotemporal network centered on posterior portions of the left inferior and middle frontal gyri to be engaged in response planning during dyadic interactions, providing further evidence for a specific role of auditory–motor neural coupling in conversational interaction.
44.4 Clinical Data
The question of whether neurological conditions affect a speaker’s propensity to align with others’ speech is relevant from three perspectives: First, clinical data can provide insight into which brain regions are involved in linguistic adaptation and thereby contribute to a refinement of models of neural control of speech. Second, entrainment paradigms can offer specific treatment options for patients with neurological speech disorders. Third, disruption of the social mechanisms of conversational adaptation can have adverse consequences for the participation of neurologically impaired individuals that may need to be addressed clinically (Borrie et al., Reference Borrie, Lubold and Pon-Barry2015).
Studies of speaker alignment in interactive or semi-interactive language use in patients with neurological disorders are still rare. The work reviewed here relates to phonetic adaptation and temporal alignment in three clinical groups, that is, post-stroke aphasia with or without apraxia of speech (AOS), Parkinson’s disease (PD), and spinocerebellar ataxia.
44.4.1 Post-Stroke Aphasia and AOS
Infarctions of the left middle cerebral artery, which most often cause language disorder, usually involve damage to components of the dorsal and/or ventral auditory stream, that is, the neural connectivity considered to implement the “perception–action pathway” that is at the center of interactive alignment theories (Pickering and Garrod, Reference Pickering and Garrod2007; Scott et al., Reference Scott, McGettigan and Eisner2009). Research has focused on the impact of lesions to the anterior versus posterior sections of this pathway on phonetic adaptation. In some studies, the anterior–posterior distinction has been implicitly equated with a contrast between “non-fluent” aphasia syndromes, that is, Broca’s aphasia, on the one hand, and “fluent aphasia” with phonological impairment, on the other, although syndrome classifications and the fluent versus non-fluent dichotomy provide only a vague account of impaired language processing components and affected cortical areas (Caramazza and Badecker, Reference Caramazza and Badecker1989).
In studies by Kappes et al., phonetic adaptation was examined in two contrastive cases of aphasia after anterior versus posterior lesions (Kappes et al., Reference Kappes, Baumgaertner, Peschke and Ziegler2009) and in a case series of individuals with left- versus right-hemisphere strokes (Kappes et al., Reference Kappes, Baumgaertner, Peschke, Goldenberg and Ziegler2010). In a word repetition task, participants overtly repeated pseudo-noun phrases, that is, disyllabic pseudowords preceded by the plural article “die” (e.g., /di:.’dai.gəl/), with stress on the first syllable of the pseudoword and a final schwa-syllable. Two phonetic parameters were varied along a continuum, that is, (i) the degree of pitch elevation on the stress-carrying syllable and (ii) the degree of hyper- or hypo-articulation of the schwa-syllable. The two paradigms represent naturally occurring variations of word stress and phonetic reduction in German. The two patients examined in Kappes et al. (Reference Kappes, Baumgaertner, Peschke and Ziegler2009) showed a clear dissociation in their propensity to adapt in word repetition: A patient with lesions in the anterior opercular, precentral, and anterior insular cortex clearly imitated the gradual phonetic variation in both parameters, whereas a second patient with lesions in inferior-parietal cortex and superior and middle temporal gyrus did not align with the model in either of the two parameters. In the case series of Kappes et al. (Reference Kappes, Baumgaertner, Peschke, Goldenberg and Ziegler2010), individuals with aphasia after left-hemisphere lesions showed a significantly lower degree of imitation in both phonetic paradigms as compared to neurotypical controls and individuals with right-hemisphere lesions. A voxel-wise analysis of the influence of lesion location on imitation in the left-hemisphere group revealed two closely neighboring regions centered in Heschl’s gyrus that were negatively associated with the degree of covert imitation of both parameters. On the contrary, lesions to anterior language areas did not influence imitation behavior. Taken together, these results suggest a role of auditory mechanisms in phonetic adaptation and are not supportive of the motor hypothesis of conversational alignment put forward by Castellucci et al. (Reference Castellucci, Kovach, Howard, Greenlee and Long2022) and Scott et al. (Reference Scott, McGettigan and Eisner2009). Note, however, that the motor hypothesis originally relates to interactions involving larger stretches of connected speech and prediction mechanisms based on the rate and rhythm of incoming speech signals (Park et al., Reference Park, Ince, Schyns, Thut and Gross2015), whereas the adaptation paradigms reported above used isolated single-word utterances that did not provide any space for rhythmic entrainment.
Temporal alignment in connected speech was addressed more specifically in a sequential synchronization experiment by Aichert et al. (Reference Aichert, Lehner and Falk2021). In this study, 12 patients with AOS and 12 patients with phonological impairment participated in a spoken sentence-completion task. AOS is a speech motor-planning disorder that is usually associated with aphasia due to lesions of anterior language areas, whereas lexical or post-lexical phonological impairment is associated with lesions located more posteriorly along the dorsal stream (Schwartz et al., Reference Schwartz, Faseyitan, Kim and Coslett2012). Participants in Aichert et al. (Reference Aichert, Lehner and Falk2021) were required to complete sentence fragments (the “primes”) “as smoothly as possible” by pre-specified, semantically compatible target words. Each prime sentence consisted of four disyllabic words (e.g., “le.na|pflanz.te|da.mals|die.se _”; English (literally) “Lena planted then this _”; periods indicate syllable boundaries, vertical strokes word boundaries; stressed syllables in bold) and had to be completed by a given disyllabic noun (e.g., “tul.pe”; English “tulip”). Half of the primes had a regular trochaic rhythm (as in the cited example) and half were metrically irregular, with alternations of trochaic and iambic words and a stress clash within the sentence. The target words were trochees (e.g., tul.pe) or iambs (e.g., te.nor) and were orthogonally arranged with the prime sentences into rhythmically compatible and incompatible prime-target pairs (for trochaic versus iambic meters, see Chapter 32). Response latency, that is, the time interval between the onset of the model speaker’s last word and the onset of the participant’s response, was used as a measure of entrainment. In neurotypical participants, metrically regular response words complementing metrically regular prime sentences were initiated almost precisely “in time with the beat”; that is, response latencies matched the mean duration of the metrical feet of the corresponding prime sentence. This suggests that in the rhythmically regular condition, participants entrained perfectly with the metrical pace of the model speaker’s utterances. After prime sentences with irregular rhythms, trochaic target words were initiated with somewhat shorter delays, suggesting that while hearing a rhythmically irregular prime sentence, the participants may have been unable to create a rhythmical framework for their response and therefore tried to produce the target word as quickly as possible (Aichert et al., Reference Aichert, Lehner and Falk2021). When the target words had an iambic meter, response latencies were consistently increased by ca. 30 ms in both prime conditions, consistent with the assumption that the much less frequent iambic pattern of lexical stress comes with a production disadvantage relative to the considerably more frequent trochaic pattern (Aichert et al., Reference Aichert, Späth and Ziegler2016).
Compared with this response pattern, patients with aphasia, especially those with a concomitant motor speech impairment, had substantially longer response latencies, which may have been due to an unspecific effect of the brain lesion or the speech-language disorder. However, more importantly, in both aphasia groups, response latencies were not modulated to any significant extent by the rhythmic pattern of the prime sentences; that is, the patients with aphasia did not entrain with the regular speech rhythm of the model speaker in terms of the timing of their responses. In the speech-apraxic patients, this outcome appears consistent with the motor hypothesis of interactive alignment advocated by Castellucci et al. (Reference Castellucci, Kovach, Howard, Greenlee and Long2022) or Scott et al. (Reference Scott, McGettigan and Eisner2009), whereas in the patients with intact motor speech, disruption of auditory aspects of the perception–action mechanism involved in interactive alignment, as suggested by Kappes et al. (Reference Kappes, Baumgaertner, Peschke, Goldenberg and Ziegler2010), may be more plausible. However, speech error analyses of the target-word responses revealed that individuals with aphasia with or without AOS made fewer errors in responses to regular than to irregular primes (Aichert et al., Reference Aichert, Lehner, Falk, Späth and Ziegler2019). This seems to suggest that they implicitly did profit from the rhythmical regularity of another speaker’s utterances, if not in terms of temporal alignment then at least in terms of the accuracy of their responses.
Speech entrainment methods have long been recognized as an efficient intervention strategy in individuals with AOS, for example through synchronous production of training items, where the clinician instructs the patient “to attend carefully to the auditory and especially to the visual cues of correct production as they say the utterance together” (e.g., Rosenbek et al., Reference Rosenbek, Lemme, Ahern, Harris and Wertz1973, p. 464). Since Rosenbek’s joint-speech intervention in speech apraxia did not necessarily involve continuous speech at the level of phrases or even texts, its effect was not easily explainable by rhythmic entrainment mechanisms. In contrast, the synchronous or choral speaking methods described and discussed by Cummins (Reference Cummins2003, Reference Cummins2009) and applied, for example, in stuttering therapy, involve synchronous production of longer stretches of continuous speech and are explicitly considered to rely on rhythmic entrainment (for a discussion, see Bradshaw and McGettigan, Reference Bradshaw and McGettigan2021; see also Chapters 45 and 46). Fridriksson et al. (Reference Fridriksson, Hubbard and Hudspeth2012) implemented a computerized version of audiovisual entrainment in which patients mimic, in real time, a videotaped speaker producing short scripts, and found a significant fluency-enhancing effect in patients with Broca’s aphasia. Johnson et al. (Reference Johnson, Yourganov and Basilakos2022) hypothesized that the improvements achieved by this method are ascribable to a synchronization of activations in anterior and posterior language areas, potentially through rhythmic entrainment mechanisms as described by Assaneo and Poeppel (Reference Assaneo and Poeppel2018) and others (see Section 44.3).
44.4.2 Parkinson’s Disease (PD)
It has repeatedly been shown that auditory rhythmic stimulation through the beat of music or of a metronome enhances movement control in individuals with PD (see Chapter 45). Most of the studies of rhythmic entrainment by such nonlinguistic auditory cues have addressed gait problems and reported greater stride length and higher walking speed (Nombela et al., Reference Nombela, Hughes, Owen and Grahn2013), while only few studies reported on therapeutic applications addressing speech impairments (e.g., Thaut et al., Reference Thaut, Mcintosh, McIntosh and Hoemberg2001). The beneficial effects of auditory–motor synchronization were explained within a theoretical framework of self-paced movements in which akinesia in PD is ascribed to a breakdown of a basal ganglia-thalamocortical circuit supporting attention-dependent motor timing and initiation mechanisms (Schwartze et al., Reference Schwartze, Keller, Patel and Kotz2011). In rhythmic auditory–motor synchronization, this dysfunctional mechanism is thought to be compensated by a cerebellar-thalamocortical circuit that supports the matching of movements to an external rhythmic template (for an outline, see Dalla Bella et al., Reference Dalla Bella, Benoit, Farrugia, Schwartze and Kotz2015; see also Chapter 45).
Compared with a metronome sound or a piece of music with a prominent beat, the speech signal of an interacting interlocutor presumably has much less rhythmic salience. One may therefore ask if individuals with PD entrain with the rhythm of others’ speech to a similar extent as they entrain with music or a metronome. This question was addressed in the sentence-completion experiment already outlined in Section 44.4.1 (Aichert et al., Reference Aichert, Lehner and Falk2021), which, along with the two aphasia groups described above, also included a group of individuals with PD. In this study it turned out that the speakers with PD showed the same response pattern as the neurotypical participants; that is, they completed the model speaker’s metrically regular prime sentences with a delay that was adjusted to the perceived speech rhythm. Thus, speakers with PD, like healthy speakers, appeared to anticipate the point in time when their response fitted into the rhythmic pattern of the model speaker’s utterance (Aichert et al., Reference Aichert, Lehner and Falk2021).
In a later study, Späth et al. (Reference Späth, Aichert and Timmann2022) examined if basal ganglia dysfunction in PD not only preserves the ability to temporally align with others’ speech but also the propensity to covertly imitate another speaker’s phonetic idiosyncrasies. Two semi-interactive paradigms were used in this study to elicit sentence utterances, that is, a sentence repetition paradigm where participants were instructed to simply repeat a model speaker’s prerecorded sentence, and a pseudo-dialogue paradigm where participants answered to the model speaker’s sentence by producing a scripted response sentence, such as in a short dyadic exchange. In a control condition, participants read the test sentences aloud. Speech rate in the model sentences was varied experimentally between 2.9 and 4.0 syllables per second. The study included 15 individuals with PD, along with 12 individuals with spinocerebellar ataxia, type 6 (SCA6; see Section 44.4.3), and 27 neurotypical controls. The research question was whether patients with PD would covertly imitate the individual rate and rhythm of the model sentences. Regarding speech rate, a linear regression model revealed that the PD group followed the item-to-item changes in the model speaker’s speech rate in the sentence repetition paradigm by a proportion of more than 30%. In the pseudo-dialogues, the degree of adaptation was smaller (ca. 20%), but still significant. There was no difference between the healthy and PD participants. On an individual basis, significant rate adaptation was found in 80% of both neurotypical and PD individuals.
To assess rhythm adaptation, the individual rhythm of each spoken sentence was represented by a vector of nine inter-beat intervals, that is, the intervals between the p-centers of the 10 successive syllables of a sentence (for p-centers, see Chapter 11). The closeness of a participant’s sentence rhythm to the associated model sentence was determined as the Euclidean distance between the two corresponding vectors of inter-beat intervals. Adaptation occurred when the Euclidean distance between the participant’s and the model speaker’s sentence became smaller in the semi-interactive as compared to the noninteractive (i.e., reading) condition. Using this measure, the PD group demonstrated significant rhythm adaptation in both the repetition and the pseudo-dialogue paradigms. Rhythm adaptation was even stronger in the PD than in the control group, with 80% of the individuals with PD versus 52% of the neurotypical participants showing significant rhythm adaptation across all sentences. This finding fits with observations of a high responsiveness of PD patients to external rhythmical stimulation and the hypothesis that they resort to cerebellar mechanisms of matching an external rhythm (Dalla Bella et al., Reference Dalla Bella, Benoit, Farrugia, Schwartze and Kotz2015). Hence, the basal ganglia dysfunction underlying PD obviously did not prevent participants from covertly adapting to a model speaker’s rate and rhythm, at least in experimental paradigms that promote attention to external stimuli, such as sentence repetition (Späth et al., Reference Späth, Aichert and Timmann2022; for a similar result, see Späth et al., Reference Späth, Aichert and Ceballos-Baumann2016).
44.4.3 Cerebellar Degeneration
Compared with basal ganglia disorders, only little is known about the role of the cerebellum in speech entrainment. Therefore, clinical studies of phonetic alignment involving cerebellar pathologies are particularly relevant. As far as temporal alignment and rhythmic entrainment in interactive language use is concerned, the critical involvement of cerebellar-thalamocortical circuits in motor and perceptual timing in the sub-second range (Buhusi and Meck, Reference Buhusi and Meck2005; Konoike and Nakamura, Reference Konoike and Nakamura2020) and in rhythmic auditory–motor synchronization (Thaut et al., Reference Thaut, Stephan and Wunderlich2009) suggest that cerebellar dysfunction would presumably interfere with smooth turn-taking in conversations or with synchronized speaking. By contrast to this assumption, however, Breska and Ivry (Reference Breska and Ivry2018) put the cerebellum’s contribution to perceptual timing into perspective by showing that temporal prediction based on rhythmic (visual) cues was preserved in patients with spinocerebellar ataxia. Similarly, Breska and Ivry (Reference Breska and Ivry2016) suggested that motor timing based on rhythms emerging implicitly from motor control parameters, such as in the repetitive drawing of circles, is unimpaired in patients with cerebellar degeneration. However, it remains open whether the conclusions drawn from such paradigms can be extrapolated to phonetic alignment in interactive speech.
A different line of research ties in with the well-established role of cortico-cerebellar circuits in the representation of forward models in motor control, that is, in the prediction of the sensory consequences of one’s own planned movements (Wolpert et al., Reference Wolpert, Miall and Kawato1998; see Chapter 6). The “predictive cerebellum” concept has in recent years been translated to the domain of action observation, suggesting that the cerebellum is not only engaged in internal sensorimotor forward modeling but also in the prediction of the consequences of motor actions observed in others (Abdelgabar et al., Reference Abdelgabar, Suttrup and Broersen2019). Even more generally, cerebro-cerebellar forward models have been considered as a mechanism to understand and predict the outcome of others’ behaviors, not only in sensorimotor terms but also in social cognition and affective processing (Sokolov et al., Reference Sokolov, Miall and Ivry2017; Van Overwalle et al., Reference Van Overwalle, Manto and Cattaneo2020). Transferred to language comprehension in conversational interactions, the posterior cerebellum was shown to be engaged in using earlier context in a perceived sentence to predict what an interlocutor is going to say in the further course of the sentence (e.g., Moberget et al., Reference Moberget, Gullesen, Andersson, Ivry and Endestad2014). This places the predictive functions of the cerebellum at the center of “active inference” (Friston and Frith, Reference Friston and Frith2015), in cognitive sequencing (Morgan et al., Reference Morgan, Slapik and Iannuzzelli2021) or in interactive alignment models (Pickering and Garrod, Reference Pickering and Garrod2013), and leads to the prediction that cerebellar pathology would interrupt the propensity of individuals to align with and adapt to others in interactive speech.
This hypothesis was tested in 12 individuals with SCA6 who were included in the rate- and rhythm adaptation experiment described in Section 44.4.2 (Späth et al., Reference Späth, Aichert and Timmann2022). Recall that in this experiment, neurotypical individuals and individuals with PD showed similarly clear tendencies to adapt to a model speaker’s speech rate and rhythm, both in sentence repetition and in dialogue-like sentence dyads. In remarkable contrast to this, the spinocerebellar group showed hardly any adaptation, neither of speech rate nor of rhythm, and none of the SCA6 individuals adapted to both rate and rhythm to a statistically significant extent. Interestingly, the proportion to which SCA6 patients aligned with the model speaker’s rate and rhythm did not depend on general motor or speech motor abilities assessed by ataxia-rating scales and a dysarthria test, and also not on auditory perceptual abilities assessed by an auditory rate discrimination test. Hence, the reduced propensity of individuals with cerebellar degenerative disorders neither resulted from purely auditory nor from purely motor dysfunctions as far as they could be assessed experimentally. Späth et al. (Reference Späth, Aichert and Timmann2022) proposed an explanation that links up with the generalized forward-modeling account of cerebellar function, suggesting that cerebellar degeneration impairs listeners’ forward modeling and prediction of their interlocutors’ speech in conversational interactions and thereby undermines the cognitive processes supporting interactive alignment and phonetic convergence (Späth et al., Reference Späth, Aichert and Timmann2022).
Summary
Neuroimaging and neurophysiological data as well as clinical evidence point to a major role of left fronto-temporal neural coupling and cortico-cerebellar circuits in interactive speaker alignment. PD as a clinical model of basal ganglia involvement appears to preserve speakers’ propensity to adapt to others’ speech rate and rhythm.
Implications
Rhythmic neural entrainment models have emphasized the role of neural oscillators in the entrainment of listeners to the quasi-rhythmic envelope of the speech signal and a coupling of auditory with motor speech areas as a platform for entrainment in interactive speech. Phonetic adaptation in the absence of rhythmic speech cues may involve other mechanisms.
Gains
Accumulating evidence for phonetic adaptation and rhythmic entrainment in interactive language use calls for a second-person approach in the modeling of speech motor control. Cognitive theories of conversational alignment through predictive coding and generalized forward-modeling mechanisms converge with modern neuroanatomical and neurophysiological perspectives on action and perception.
Open access