A short primer on stuttering with a view toward ICT applications

Contenido de la comunicación

1. Introduction

This communication aims at drawing attention of researchers and developers to a field which, although not well-known, concentrates enough scientific and social interest. What is stuttering? How can we measure it? How can we treat it? These interrelated sound questions do not have a complete answer, yet. Nevertheless, stuttering has been systematically studied for about a century: no doubt there is something strange about it. The reason might be in part that stuttering behaviours have their roots in subtle, not yet well understood neural phenomena involving the way we generate speech and language, even before passing to the motor system to produce audible speech. Another important reason may be the complex relationships between stuttering behaviours and the person's social, linguistic, and affective environment.

The study of stuttering has to face the change to a new techonological paradigm as a refreshing opportunity to conceive new strategies to solve long-standing open questions which until now have resisted enthusiastic research efforts. The aim of this communication is to justify this claim. A condensed State of the Art (SoA) in stuttering research including tips and proposals of strategies in the field of stutering relying upon ICT will be displayed.

This communication does not intend to be confined within the domains of ICT, but rather to provide a mixed piece of knowledge to everyone interested in multidisciplinar problems overlapping with stuttering research and ICT.

This paper owes its existence to a six-month documentary research conducted by the author as a technical consultant at the Centre de Recerca i Investigació de Catalunya (CRIC) for the Center for Advanced Studies in Stuttering (CEAT, http://www.ceat-es.org), a newborn center for studies in stuttering research.

2. General settings

A stuttering-like disorder is a disorder which manifests through fluctuant speech behaviour characterized by involuntary syllable blocks, repetitions and prolongations, especially during connected speech, thereby imapiring normally fluent speech. Stuttering-like disorders are traditionally split into three main types: neurogenic (onset related to brain injury), psychogenic (onset related to psyshological trauma) and developmental (onset not related to any concrete circumstances, normally during early childhood). Among these three types it is the third one which occupies the mainstream of research and therapeutic efforts and it is usually referred to as  Persistent Developmental Stuttering, in the sequel  PD-stuttering.

This disorder provides a fascinating disease model of speech production and challenges the available theories of speech and language. Most studies point at the high prevalence in the population (which are usually set about 1%), the childhood incidence (the disorder ocurs in approximately 5% of all preschool-age children), the marked gender ratio (3:1 of men:women), its most probable genetic basis, and the responsiveness to a range of environmental stimuli. There is a high rate of spontaneous recovery in children (around 80%), but stuttering that persists into adolescence or adulthood is much more resistant. A key point is that self-consciousness or congnitive overloads appear to feed the stuttering mechanisms. If the condition persists beyond the age of ten then it becomes typically a lifelong, chronic problem. It is thought that PD-stuttering is a condition that adversely affects a person's education, social life and career. Quite often, stuttering is experienced as a stigma: stigmatization could be described as the process by means of which the stutterer is socially endowed with a personality that does not correspond to his genuine personality, coming from the social prejudices labeling people who stutter as shy, insecure, unsteady, or weak.

The available research evidence point at an organic, neurologic origin of the disorder. Brain imaging studies reveal anatomical differences in stutterers, even in the scarce existing studies with children. Results for the first year of stuttering indicate that the brains of young children are already beginning to wire up differently. Even basic motor and language areas show differences. Children appear to be born with a set of genetic characteristics that make them more likely to stutter. However the relationship between stuttering and genetics is neither simple nor clear: identical twins will not necessarily both stutter if one of them does.

In any case, it is known stuttering is not a speech periphery disorder. In general, there is nothing wrong in the phonation and articulation processes in people who stutter: almost every stutterer is able to speak fluently. Depending on each case fluent speech will occur provided perceptual and cognitive conditions which may vary. A common situation in which fluent speech is likely to happen in stutterers is given by the so-called choral effect, which is the effect by means of which speech fluency is induced when simultaneously producing the same speech together with another person. Many experimental designs obtain fluent speech by people who stutter by using this phenomenon. Delayed Auditory Feedback (DAF) devices aim at exploiting this phenomenon by letting the speaker hear its own speech with a very short time delay.

2.1. Manifestations

Among stuttering manifestations there should be considered not only the overt ones, such as the speech disruptions themselves, ancillary behaviours or struggle. Hidden behaviours an feelings are, for many researchers, where the majority of the problem resides, since they are not only emotionally painful and impairing, but also appear to feed the stuttering circuitry, whatever it was. In the sequel we will adopt a splitting of the set of speech-language manifestations involved in PD-stuttering, as they were stated in ¹. Overt behaviours consist of core speech disruptions (e.g., repetitions, prolongations, silent postural fixations), and ancillary behaviours (e.g., eye rolling, fist pounding, facial grimacing, jaw clenching, finger tapping, foot stomping). These are the most easily demarcated and categorised manifestations of stuttering. However research efforts towards their classification not produced conclusive outcomes. Sub-perceptual stuttering events will stand for those phenomena that, by their nature, are invisible without the use of sensitive measurement equipment. Some authors distinguish still another level, refining the sub-perceptual, which would encompass all stuttering-related behaviours that cannot be observed using any human senses or sensitive equipments that measure output from the speech periphery which include: social and linguistic avoidances, shame, and fear. These are the covert behaviours.

2.2. The group difference design

So far, there has not been identified any set of abnormalities that can be found in every stutterer: this is known as heterogeneity. Hence stuttering still lacks a good operational definition.

The current popularity of the group difference design in experimental research within the field of stuttering is a consequence of the prevailing belief that this disorder is to be understood by focusing on differences between stuttering and non-stuttering speakers, especially differences related to the existence of a disfunctional central speech generator system. The primary strategy regarding this design scheme is to compare the stuttering and the non-stuttering group statistical means. However, by emphasizing the difference between group means only a limited amount of information is extracted. The strategies usually applied take advantage of supplementary indices to augment the usual test of statistical significance of the difference between means. This supplementary statistical tools may include analysis of variance (ANOVA) and power analysis.

2.3. Severity constructs

It can be asked whether the existing prevailing and most frequently used stuttering severity metrics are effective (in the sense of validity, reliablility, and efficiency) and what makes a severity metric useful. In order to answer to these questions we will examine in Section 3 how current methods of assessment work and which are its most widely proposed building blocks.

Moreover, it can be posed the question whether it can be conceived an index (as a function which takes each case of stuttering to a value within a scale or grid) which could reasonably convey stuttering severity. Such an index would ideally consist of a formula or procedure combining biomedical, behavioural or cognitive data from the stutterer, and should reflect every aspect considered to be meaningful from the point of view of a good construct of stuttering severity. However a good definition of stuttering severity might not be unique. Which should be the proper definition of stuttering severity? Which are the aspects to be included in? Until now there is no complete answer, though some promising constructs of stuttering severity have been proposed, and still many long-standing others have been discarded due to research efforts. Are there any valid, reliable, effective methods to assess stuttering severity? There is a only a collection of proposed strategies to cope important but partial aspects according to the prevailing notions of stuttering severity. Which is the knowledge and technological gap in order to implement an effective and integrated method? The basic criteria in order to answer to this question would be given by the interest of tackling the problem, namely: to check effectiveness of stuttering treatments; to decide how likely a given child would develop PD-stuttering; and to determine the degree of disability implied in each case of stutterers.

2.4. ICT and stuttering

Many of the greatest recent achievements in the field of stuttering owe their existence to the application of ITC. At this point we may highlight the trascendental importance of a family of imaging algorithms known as reconstruction algorithms, due to the weight of brain imaging in present stuttering research. This class of algorithms are useful to enhance resolution of static images, but also to compensate movement effects in functional studies, whenever a series of frames are recorded time-locked to the performance of some task or presentation of stimuli which may involve movement of the target. In the case of severe stutterers, this is a common situation, for struggle and ancillary behaviours are strong among them while performing speech tasks. Another important area related to stuttering research which has been enhanced a lot by the introduction and improvement of computers is signal analysis. Until today, a big deal of research tasks has involved the analysis of large data sets which may be stored as signals or function graphs, then operated, filtered, and conveniently transformed. Muscular electrical activity (EMG), brain electrical activity (EEG, ERP, MMN, MEG), speech production and auditive acoustic studies, openess of tracts and vocal fold vibration, autonomic correlates such as skin conductance or pulse volume, articulation stability via lip movement, are examples of sources of signal data sets which have been typically analysed using computer-based tools of signal analysis. In Section 4 an account of these parameters will be presented.

It is not necessary to underline the role of the Internet in spreading knowledge, exchange, and remote collaboration, in particular in scientific research and publishing mechanisms. But perhaps more importantly from the point of view of socially impaired people, the Internet has allowed stuttering people to enhance their communicative and social experience, thus organizing and gathering into stuttering supportive groups and associations. The versatile non-verbal communication feature, including a wide range of available, web-based, collaborative publication technologies and the improvement of web search engines, has allowed many stuttering people to come out of the closet. The shift of human relationships toward the internet-based paradigm has represented to the stuttering community an important step forward.

3. Methods of assessment

Several methods of assessment of stuttering severity have been already implemented. We do not intend to list every existing method, but rather to have a comprehensive record of the building blocks which can be repeatedly found across them. Remark there can be distinguished two main types of assessment actions: those relying upon measuring overt stuttering behaviours and those trying to extract experiential aspects of stuttering in the domain of covert behaviours.

In the first class there could be exhibited several examples among which Riley’s Stuttering Severity Instrument, a method that conceives severity as a matter of the degree of disfluency and speech struggle. The total score for the SSI is obtained from three components which have been recurrent in stuttering evaluation: stuttering rate as the frequency of disfluency events in reading aloud and speaking, assessed as the percentages of syllables stuttered (%SS); duration of stuttering events; and physical concomitants, assessed as clinician's recording regarding head and extremities movements, distracting sounds, or facial grimaces.

These dimensions admit a wide range of variants. In studies conducted by the stuttering research team of Roger J. Ingham, Peter T. Fox and collaborators from the University of Texas at San Antonio, stuttering rate is measured as the number of certain time intervals judged to contain a stuttering event; syllable production is computed as the total number of syllables spoken in the scanning period; and speech naturalness is rated on a 9-point scale, according to ².  In a study conducted by Anne-Lise Giraud, and others ³, they use also a laterality quotient (LQ) and speech naturalness subjective rating.

3.1. Overall Assessment of Speaker's Experience of Stuttering (OASES)

This assessment is based on a questionnaire which covers three main components of the experience of stuttering, plus an additional general information questionnaire:

    • Speaker's reaction (affective, behavioural, and cognitive).

    • Functional communication.

    • Quality of life.

The OASES can be used alongside widely used measures such as the SSI or other real-time disfluency frequency counting procedures to provide a more complete account of the speaker's overall experience of the stuttering disorder.

This kind of methods capture information through a questionnaire, hence it is addressed to adolescents and adults, for a minimum cognitive maturity is required. A main fact about this method resides in that it was conceived according to definitions and criteria about aspects of human functioning and disability as posed by the World Health Organisation in the ICF classification, which is a good source in order to define health severity constructs. Refer to ⁴.

3.2. Anxiety inventories

Anxiety inventories constitute an important set of assessments with regard to stuttering severity. Anxiety is a construct closely related to stuttering, since it may capture aspects of an important component of stuttering: anticipatory fear.

Several inventories have been proposed across the literature regarding the hypothetical link between stuttering and anxiety, a construct which is often object of misunderstanding and confussion. One could say that for each inventory, a new different construct of anxiety is implicitly proposed. Refer to Section 4.1.1 for more references on cognitive anxiety.

Although not strictly a measure of attitude or anxiety, the Teachers Assessment of Student Communicative Competence (TASCC, ⁵) is useful in assessing child's communicative functioning in the classroom. One of its subscales purports to measure approach-avoidance in the classroom based on questions about the child's class participation and volunteering to talk.

3.3. Assessment of preschool children

It has been suggested that preschool child's sensitivity or reactivity to new situations is an important consideration in therapy and possibly predictive of chronicity (refer to ⁶). The provides some information on this dimension. Research indicates that the questionnaires such as the Behavioural Style Questionnaire (BSQ) may be able to identify those children who have a more inhibited, sensitive temperament.

As part of the operant treatment for preschool children known as the Lidcombe Program (refer to ⁷ and ⁸) it was developed the Lidcombe Program's Severity Rating Scale. It is a 1-10 scale that parents use to make daily ratings of their child's stuttering. At the beginning of the treatment parents are trained to accurately rate their child's severity using videotaped samples and observations of the child's speech in the clinic. Research of this severity rating scale has shown it to be a reliable tool for conveniently obtaining information on a child's stuttering outside of the treatment environment.

3.4. Speech naturalness

While we could say roughly that naturalness is a magnitude coping the ease of effort of speech and the closeness to a normalcy pattern, attempts to define speech naturalness univocally tackle the difficulty of establishing a normalcy speech pattern.

It has been proposed the following paradox: rather than motoric manipulations or fluency skills themselves being responsible for the reduction of overt stuttering events, it is the perceived unnatural speech quality that inhibits stuttering and allows speech to be produced without overt repetitions or prolongations, thus as a more natural speech quality is imposed on post-therapeutic speech, stuttering events that were likely absent in the highly unnatrual speech product are likely to resurface (refer to ⁹). Then, hypothetically, when a person who stutters imitates a therapeutic speech model, mirror neurons (refer to ¹⁰ and ¹¹) are engaged and speech gestures are fluently reproduced as stuttering is inhibited (refer to ¹²).

Several authors have suggested naturalness to be estimated operationally, by means of a collection of ratings given by naive judges or raters: listeners whose expertise or experience area is supposed to be unrelated to speech and hearing pathology. These judges are presented a sample of speech and express their opinion about its naturalness within a scale (see ¹³ and ¹⁴). Statistical considerations, as well as experimentation, may lead to satisfactory definitions of speech naturalness from the point of view of reliability and consistency.

3.5. Rating networks

The operational collaborative estimation of naturalness brings us to an idea which has not been implemented yet: a network with vertices clinicians and raters aiming at assessing subjetive aspects regarding stuttering evaluation. This idea fits very well with new trends on swarm-like, collaborative tools. The idea consists in allowing the clinician upload samples, in audio or video formats, which are then displayed by the system to a random set of registered naive raters. These raters would return their own ratings to the clinician. The resulting data set from this process could then be manipulated.

Such a system should provide with a series of functionalities including random matchings, homogeneous workload, statistical significance data filters, and computations of several sample statistics. The system should also provide with a way to learn about the raters’ behaviour by using artificial intelligence agents. These agents could check whether an individual rater is usually close or not to overall resulting ratings, assigning historical values or weights.

This idea could also be applied beyond speech naturalness purposes, for rating purposes with regard to other parameters involving fuzzy aspects, even out of the stuttering domain. It would be also of interest to check whether any aspect of stuttering severity would admit a reliable adaptation to this swarm-like methodology, such as evaluating disfluencies, or assessing speech effort or struggle. An interesting questions is which the obstructions would be in either case.

3.6. Self-reports

PD-stuttering is a complete syndrome of continuous compensatory behaviours: a central, experiential sense of loss of control that manifests itself across a continuum of compensatory behaviours from the central nervous system outwards to the speech periphery. The stuttering community acknowledges the need to improve the therapeutic techniques: every type of therapy conceived by the human mind cannot be effective in ameliorating stuttering. However the same community fails to recognize that measurement tools have failed research and therapy, providing a falsely inflated picture of success. Refer to [12, 77] for an account of follow-ups after successful treatments.

It has been concluded any efficient and effective means of evaluating intervention methods over the long-term should include a form of self-report as a primary tool, as it best accesses the experiential sense of loss of control and other covert behaviours (refer to ¹⁵). Among the aforementioned assessments, OASES belongs to the category of self-administered methods of assessment. Each of its items tackles a different important aspect of the experience, demanding situations, social context, and personal impact of stuttering. However, altough practical, these kind of methods take the emotional link of the indvidual with its experience for granted, whereas this process is clearly non-trivial for many people.

In order to enhance the validity of assessments conceived as self-reports, it can be proposed whether there could be developed a system which we could refer to as a multimedia self-administered protocol of evaluation. These protocol would strongly rely upon existing methods and schemes, but enhancing the emotional, cognitive link through the reproduction of audiovisual contents related to the varying aspects the clinician may want to assess. The protocol could focus on particular aspects whenever needed, introducing new questions to refine the patient’s answers based on significance or reliability estimates, and it could also allow the patient to interact with the scenes in order to invoking real emotional, cognitive, or social responses. This protocol could be implemented over an expert, AI-based system which would drive each evaluation session according to the clinician’s preferences and would be able to decide at each step whether the gathered data is significative, consistent or reliable enough.

3.7. Agent-driven disfluency check

Those existing methods which cope overt aspects of stuttering tackle two main difficulties: reliability and time cost. There is a wide consensus in including a fluency check in any valid method of assessment. Even if the definition of disfluency appears to be objective enough, experiments addressed to check the reliability of counting disfluencies from speech samples reveal there are differences depending on the judge. There have been proposed methods in order to maximize this reliability, but the resulting procedures cannot avoid the problem.

There is a clear need to implement ITC assisted protocols in order to capture disruptions and overt behaviours by analysing objectively complex image or sound samples. While there is no general pattern of disfluency, an AI-driven system would be able to learn to recognize individual disfluency patterns, tics or even ancillary behaviours through image, acoustic data or even combined data derived from position transducers which could keep track of lip or eye movement, or face musculature activation. The spectrum of available sources of data to play with is wide. Several studies already point at the consistency of individual disfluency patterns via the measurement of certain simple biomedical parameters related to motor speech production (see Section 4.9), autonomic nervous system arousal, laryngeal and respiratory neuromuscular activity and brain activity. Indeed, there are constant references to intuitive individual patterning identification by researchers in experiments across the stuttering scientific literature. In many cases, whenever experimenters failed at identifying a trait or parameter that separated stutterers from non-stutterers in general, they were able, at least, to identify individual patterns in the meantime. Today it is possible for a computer program to learn to reliably recognise simple patterns, let them be geometrical (position data, image analysis), numerical or relational.  But, could it be developed an agent-driven expert system aiming at reliably recognizing individual disfluency patterns and affordable?

3.8. Toward a stuttering index

The existing instruments for assessing stuttering can be distinguished depending on the aspect or dimension they are linked to. The interest of the existing assessments relies on the fact that any of them involves aspects which are acknowledged as meaningful from the point of view of assessing the degree of severity in stuttering. Is there any way to combine or somehow unify the existing methods for assessing stuttering? We might distinguish two postures with regard to this question. The first one would be to assume we have attained the essential building blocks, then we have to look for a procedure or formula to capture a good notion of stuttering severity. The second one would be to conclude the question makes no sense with regard to the State of the Art, assuming every aspect covered by existing assessments is not conclusive or, at least, has to be refined, so that the scientific community has to keep on analysing new parameter candidates.

The next section tackles the analysis of some promising parameters which could be useful in order to improve stuttering assessment and characterisation.

4. Stuttering parameters

In the sequel the term stuttering parameter will denote any magnitude which allow us to define conditions correlated to the hypothetical stuttering condition. Note the notion of correlation used here is to be understood in a weak, non-formal sense.

The search for new stuttering parameters is justified provided they can be used in order to construct more effective conditions which allow us to characterise stuttering. From the theoretical point of view, the more stuttering parameters are available, the more accurate operational characterisations can be conceived. Moreover, the more powerful computing machines and tools we get, the more sustainable this assertion is. However, the overabundance of studied parameters appears to be linked to the lack of truly valid ones.

In the following sections we will summarize describe potentially useful parameters in order to characterize the stuttering disorder, as they merged and were selected in a recent bibliographical study of the State of the Art.

4.1. Anxiety

When dealing with terminology such as fear, anxiety, and concern a great deal of misunderstanding could raise. Significantly, it is thought anticipatory fear feeds the stuttering circuitry, in the sense of producing psychological suffering and a higher rate of speech accidents.

Cognitive anxiety

Measures of anxiety by means of self-reports or clinician-assisted reports which involve cognitive constructs related to fear and anxiety. Each metric may measure the anxiety manifested in a specific situation (state anxiety produced in speech demanding contexts, social anxiety) or the set of stable personal traits (which is known as trait anxiety).

Nonsurprisingly state anxiety in speech demanding situations is higher in stutterers (refer to ¹⁶). It is suggested a quantitative approach to the somewhat diffuse notion of anticipatory fear relying on well known standard methods of assessment of cognitive anxiety.

Assessing social anxiety may be meaningful when trying to quantify anxiety associated to stuttering. In ¹⁷ the authors use the Liebowitz Social Anxiety Scale (LSAS) to assess social anxiety in stutterers compared to normally fluent speakers, suggesting anxiety is not an essential factor, but an important element in the assessment of adults who stutter.

Autonomic anxiety

Autonomic anxiety stands for those changes of the arousal of the Autonomic Nervous System (ANS) related to anxiety. Two main sub-systems of the ANS can be identified, the sympathetic nervous system (S-ANS) and the parasympathetic nervous system (P-ANS), which cooperate to adapt the bodily functions to different situations and demands. S-ANS arousal entails, among other organic changes, an increment of heart rate, an increment of blood flow in skeletic musculature and lungs, whereas P-ANS arousal entails a decrease of heart rate, an increase of gastro-intestinal blood flow, peristhalsis acceleration, and salivar secretion stimulation.

It does not imply that both subsystems are reciprocal or opposite. Indeed, they do co-activate under some circumstances. Autonomic co-activation in humans was shown in ¹⁸. There are two main fear or defense responses in humans from the point of view of autonomic arousal: the fight or flight response, which is achieved via S-ANS arousal, and the freezing response, which is achieved via co-activation.

A possible relation between the freezing response and stuttering was first suggested by Peters and Guitar ¹⁹. Three important contributions to this subject are due to ²⁰, ²¹, and ²². These studies' results suggest that what appeared to be a lack of or only moderately high autonomic arousal during speech production in stutterers, was in fact a co-activation linked to the freezing response. Hence higher co-activation or freezing response can be linked to stuttering, and its measurement can be a good index to assess covert dimensions of stuttering as well as to anticipate overt disfluency.

The need for monitoring autonomic co-activation faces a technological gap, since available equipment is not addressed to be useful in speech-therapy environments, but rather in more generic biofeedback contexts. Then the use of these instruments is expensive and restricted to few clinical and research environments. However there is no doubt that autonomic recordings in a wide range of speech-demanding situations would be useful in order to perform valid and reliable assessments and prognostics. New trends in transducers, wireless, low-energy sensor networks and portable interfaces may provide with new instruments which would allow retrieving real-time parameters related to autonomic arousal, such as pulse volume or skin conductance baseline and responses, allowing management of complex data and later analysis.

4.2. EMG activity

Electromyography (EMG) is a widely used technique for measuring muscular activation through the measurement of muscular electrical activity. This method can detect tiny changes of electrical potentials in the muscle fibres. Since the nature of muscular electrical activity is oscillatory, analysis of the resulting signals may serve to shed new light on phenomena linked to muscular activity.

Laryngeal EMG activity

There are two studies that are widely cited with regard to laryngeal activity via muscular activation: ²³ and ²⁴. However, there is some controversy around the conclusions of these studies. The study undertaken by ²⁵ analyses the problem of the increment of laryngeal muscular activation and concludes that the results of the former authors are not well justified; moreover, there is no agreement when reproducing the same experimental settings with different sampling groups. In conclusion they reject the hypothesis about laryngeal muscular overactivation.

Surprinsingly, what emerged from ²⁶ was that each subject tended to have an individual consistent oscillatory pattern of activity in CT (cricothyroid) or TA (thyroarytenoid) muscles during stuttering, regardless of the type of disfluency, even though there are substantial differences between subjects.

Neuromuscular physiology

In ²⁷ the authors explore the onset of speech or metabolic respiration and its relationship with speech production in stutterers and non-stutterers. By means of EMG recordings of muscles associated with respiration, the authors study a high frequency component of the signal (HFO) which is originated at the brain stem, known as High Frequency Oscillations (HFO). This signal is related to the regulation of respiratory muscular activity. There is some evidence of an atypical HFO signal in stutterers. Some studies have analysed the relationship between two central respiratory controllers: the metabolic respiratory controller (MRC) and the speech respiratory controller (SRC). The methodology used has been to check the maximum coherence (MC) of the HFO-signal, which can be linked to the activation-inhibition of the MRC system. High MC-HFO for the speech breathing may identify stuttering subjects most at risk for respiratory-related effects of emotional stress (refer to [18, 19]).

4.3. Brain Anatomy

Improved brain imaging techniques allow the study of brain anatomy. So far, brain imaging studies have revealed brain differences between groups of stutterers and non-stutterers. These results are among the most relevant since the beginning of systematic stuttering research.

In a series of studies, Foundas and colleagues identify several anatomical group differences related to stuttering: in ²⁸ the authors identify anatomical differences between groups in the morphology of the perysylvian fissure using MRI. Measurements of asymmetries revealed in ²⁹ a difference in the cortical folding of the right perisylvian region observed in stuttering subjects. According to ³⁰ an atypical prefrontal and occipital volume ratio is identified in groups of stutterers using volumetric MRI techniques, and ³¹ found increased white matter volumes were found in the right hemisphere, including the superior temporal gyrus, using voxel-based morphometry (VBM).

Some studies suggest connectivity between left cortical regions is smaller in people who stutter, hence a cortical disconnection between the frontal operculum and the ventral premotor cortex (refer to ³²). More recently ³³ investigated the differences of regional grey matter volume between adults with PD-stuttering and fluent speaking adults, using VBM. The analysis revealed that compared with the controls, the stuttering adults had significant clusters of locally gray matter volume increased and decreased.

The planum temporale is a region in the brain related to the central auditory function. It is located both in the left and right hemisphere and is typically more developed in the left hemisphere. A study revealed that in a large subset of stutterers this asymmetry is rightward reversed (refer to ³⁴). Moreover, subjects with this atypical rightward asymmetry are more susceptible to Delayed Auditory Feedback (DAF): they improve their speech performances, avoiding fluency, under DAF conditions, whereas subjects without this anomaly do not experiment such changes.

Childhood brain anatomy

Because asymmetries in brain structure in adults could have resulted from functional differences during development as has been shown for handedness, language laterality, reversed laterality, bilingualism, and instrument practice, some of the brain structure differences in adults found in studies with adults may have occurred not as a cause, but as a result of stuttering.

A crucial study was undertaken by ³⁵ with sample groups of non-stuttering, stuttering and recovered 9-to-12 year-old children. This is the first imaging study with children. They used VBM to measure gray matter volume in brain regions associated to speech, showing reduced gray matter volumes in stuttering children. Diffusion Tensor Imaging was used to check the left hemispheric representations of the larynx and face, showing reduced fractional anisotropy. Interestingly, the study reveals no differences in the right hemisphere. These findings suggest a structural origin of stuttering and that atypical rightward asymmetries in adults arise as an structural adaptation or response to stuttering in the long term.

4.4. Brain Physiology

In the last decades, the neural basis of PD-stuttering has been extensively assessed using brain imaging techniques such as functional Magnetic Ressonance Imaging (fMRI), Positron Emission Tomography (PET), Magnetoencephalography (MEG), and Electroencephalography (EEG) and its derived methodologies, such as Event Related Potentials (ERP).

The importance of imaging techniques is due in part to the underlying hardware and software implementations which allow image processing and reconstruction. A whole familiy of procedures, known as reconstruction algorithms, tackle the problem of the reconstruction of structures from data collected based on transmitted or emitted radiation. The problem occurs in a wide range of areas. One of the key points is to improve efficiency. The essential role of hardware as a limiting factor has to be emphasized, which is due to the fact that the underlying processes and mathematical problems behind are complex, involving large systems of equations.

Several brain function differences have been found during speech in stuttering adults: 

    • Reduced or abnormal activity in the auditory associated areas;

    • Increased activity in the right frontal and left cerebellar regions relating to stuttering;

    • Abnormal timing relationships between premotor and primary motor regions in the left hemisphere;

    • Increased activity in the left putamen, ventral thalamus and inferior anterior cingulate related to stuttering.

Whether increases in activity in the right hemisphere speech regions represent a compensatory brain activation for stuttering or brain function differences undelying stuttering has been debated (see ³⁶, ³⁷, ³⁸). According to ³⁹ there are right-sided increases both during speech production and other tasks, suggesting that greater right hemisphere activation may be inherent in adults who stutter.

Following ⁴⁰ there are three general classes of functional neuroimaging findings that have emerged: overactivation of cortical motor areas, such as the primary motor cortex and supplementary motor area; anomalous lateralization, such that speech related brain areas that typically have left-hemisphere dominance in fluent speakers are active bilaterally or with right-hemisphere dominance in stutterers; auditory suppresion such that primary and secondary auditory areas that are normally active during speech production are not activated (see ⁴¹).

Although the frontal operculum of the left hemisphere has well established functional linkages with speech, language process, and even manual imitation, the functional role of the Right Frontal Operculum (RFO) is far more elusive. Activity in the RFO stands out as being unique in two respects. First, unlike other lateral motor areas, activity was found exclusively in the right hemisphere. Second, activity was found uniquely in stutterers. Refer to [51, 58] for more on RFO and its relationship with mirror neurons.

The published literature supports a robust auditory inhibitory effect in stutterers (refer to [25, 26, 34, 51, 70]). On the other hand, there is now considerable evidence that increased skill is associated with concomitant decrease in brain circuitry activation. In stuttering, it is likely that this effect come into play. Most of the studies show a consistent atypical lateralization of speech production functions. Activity patterns vary across subjects, but there is a consistent trend to rightward asymmetry: right hemisphere is overactive with respect to left hemisphere. Intensity of activation in differential regions does not correlate in general to conventional metrics of stuttering severity. Moreover, it seems to be a relationship between the presence of atypical brain functional organisation and stuttering compensation for the long term.

4.5. Electrophysiology

Recording electrical activity produced by the firing of neurons within the brain has been a widely used technique to understand brain functioning regarding the stuttering disorder. We shall examine two essentially different methods of recording this activity and transforming it into a signal:

    • Event related potentials (ERP). It is a derivation of the EEG technique, consisting in averaging the EEG signal time-locked to the presentation of a stimulus (either visual, sensorimotor or auditory) or task. The key feature of EEG methodology regarding ERP studies is its high temporal resolution.

    • Magnetoencephalography (MEG). It is based on the measurement of the magnetic field induced by the electrical cortical activity on the scalp. This technique is also event related, meaning that recordings are time-locked to the presentation of a stimulus or task.

 In both methods, the signal obtained after recording the electrical activity typically shows some characteristic oscillations due to the synchronised activity over networks of neurons. These characteristic oscillations, as well as the underlying networks of neurons and their corresponding brain functions are well understood only in a few cases.

Event-Related Potentials (ERPs)

An Event-Related Potential (ERP) can be defined as a characteristic component of the signal obtained from the electrical activity recording on the scalp, and time-locked to the presentation of a stimulus or performance of a task. ERPs admit a qualitative description based on their amplitude over the baseline signal, latency with respect to the stimulus onset, and polarity. ERPs, as an EEG-derived technique, are obtained by means of electrodes distributed on the scalp.

Although an ERP is completely described by giving the location and set up of the electrodes, the amplitude of the signal (usually measured in micro-volts), the latency and the polarity, an ERP response is usually denoted by a letter (P/N) for its polarity and a number its latency expressed in milliseconds: e.g. P300 would refer to an ERP response with positive polarity and latency of 300 ms.

ERP methodology can be useful to determine whether there are functional aspects of the brain which are atypical, by means of testing them under specific tasks or stimuli, and it is widely used in several clinical contexts. Let's summarize some suggestive findings obtained via ERP methodology with references. In ⁴² it was conducted an oddball paradigm tonal stimulus experiment and identified an atypical P300 ERP response in the stuttering group. In ⁴³ a lexical access task produced atypical N280, N350 and N400 ERP responses in the stuttering group. Finally, in ⁴⁴ atypical P600 ERP response is shown in the stuttering group while performing a syntactic recognition task. The Mismatch Negativity (MMN) is a particular event related potential (ERP) elicited to unexpected auditory stimuli deviating from the preceding standard sounds in any of their physical or even more complex atributes. In ⁴⁵ the authors aimed at determining whether adults with PD-stuttering had auditory perceptual deficits comparing the MMN elicited to concrete auditory stimuli in a sample of people who stutter with that recorded in a sample of paired fluent control subjects. They found the altered MMN instead of being absent or reduced was abnormally enlarged, suggesting an overexcited response of the auditory cortex to specific speech sounds.

Magnetoencephalography (MEG)

Magnetoencephalography (MEG) is a brain imaging technique based on the measurement of the magnetic field induced by cortical electrical currents. It is based on the well known fact that neuronal electrical currents induce orthogonal magnetic fields. In this case the measured neuronal potentials are small (on the order of microvolts), and the resulting magnetic field is tiny (on the order of femtoteslas, fT).

MEG has revealed abnormal temporal patterns of activation in people who stutter (see ⁴⁶ and ⁴⁷). Indeed, the activation patterns after seeing a word progressed in stutterers from a frontoparietal region encompassing the left lateral central sulcus and the dorsal premotor cortex to a left inferior frontal cortex region, whereas fluent speakers had a reversed activation sequence, suggesting that stutterers trigger speech motor programs before activation of the articulatory code.

When fluent speakers read out loud isolated words the sequence initially follows that observed in silent reading. In addition, activation is observed in the left inferior frontal cortex when preparing to speak, probably reflecting access to the phonological representation of the word for articulation, and bilaterally in the motor and premotor cortex and supplementary motor area during actual speech production. Intriguingly, these are the very areas in which the timing and strength of activation in developmental stutterers has been found to differ from that in controls, thus suggesting abnormalities in overt speech production rather than core linguistic analysis (see ⁴⁸, ⁴⁹). Furthermore, activation of these same areas was affected also when stutterers simply listened to spoken sentences that they need to repeat or transform after a short delay (see ⁵⁰).

4.6. Kinematics and motor stability

Speech kinematics studies the movement of the anatomical elements involved in speech articulation. The general purpose of studies of this type consists in producing a kinematic recording, and further analysis, from speech samples.

The study of motor stability of speech production involves a concrete analysis of the data obtained from several speech samples of the same utterance performed by the subject at a similar rate. Kinematic recordings are normalized and manipulated in a precise way, thus obtaining an index known as spatio-temporal index (STI) and may be thought as a measure of the variablility of a family of curves, in this case, signals obtained as position graphs. For details on definitions the reader should refer to ⁵¹ and ⁵².

 In ⁵³ the authors registered STI of stuttering and non-stuttering subjects by measuring upper lip movements via infrared transducers. Kinematic changes in the fluent speech of adults who do and do not stutter in response to increasing syntactic complexity and length of utterances, were observed and analyzed. Each subject performed the same series of utterances, which varied in length and syntactic complexity. The study concluded that the stuttering group showed a significatively greater STI variability than the non-stuttering group.

4.7. Time perception

Time perception has been proposed as a representative parameter in order to assess the cognitive loads associated to speech performance. Various experiments have been undertaken with coincident results, pointing to time overestimation in adult stutterers. Time estimation may be a way to measure the extent to which stutterers pass into a state of internal thinking during the moment of stuttering or, more generally, while performing speech tasks which somehow induce a cognitive overload, as was suggested by ⁵⁴. The experiences of psychological time pressure during communication is thought to be a major source of stress for the adult stutterer. Also, stress and time pressure are interrelated trait and state factors that influence cognitive-affective processes.

Research indicates that time estimation is sensitive to the presence of perceptual or cognitive demands, whereas tasks that are done automatically do not interfere with it. Thus when performed simultaneously with another task, time estimation may be used as a sensitive and practical index of mental workload (see ⁵⁵ and ⁵⁶).

4.8. Instrumented assessment

Some of the aforementioned parameters appear to be very specific of PD-stuttering, whereas others point at more generic organic behaviours which can be somehow perturbed and hardwired since the onset of the stuttering disorder takes place. In any case, if one assumes enough computing power, one may expect that considering measures of as many of these parameters as possible, clinicians and raters in general would be able to refine their judgements about stuttering. The point is how to manage large sets of data, in the light of the State of the Art and new advances in ITC in order to make them meaninful and profitable.

While the group difference design has been the prevailing paradigm in stuttering research for the last decades (often examining a single parameter at a time), studies have been undertaken which examine a given set of parameters on a group of stuttering subjects, looking for individual patterns or invariants which could lead to therapeutic improvements. A big deal of effort has been addressed toward research on non-invasive instrumented assessment methods combining different sub-perceptual parameters (e.g. physiologic, kinematic, aerodymanic, acoustic). Recall, for instance, that⁵⁷ concluded that even in the abscence of perceivable stuttering, the speech of stutterers contained numerous abnormalities, often individually shaped.

A special interest have received those instrumented assessment methods conveying a complex signal which may be integrated in a biofeedback process: the aim of this process has been to raise the patient's awareness of the elicited bodily functions and therefore the possibility of conscious control; the point is to transform bodily functions information into stimuli which could be interpreted by the subject. This has been usually done via visual, real-time data presentation.

Potentially useful instrumental procedures for stuttering assessment were already outlined by ⁵⁸. These procedures include monitoring several aspects, such as changes in vocal fold contract area through electroglottography (EGG), a technique which estimates geometric changes of the larynx by measuring changes in conductance; respiratory movement through inductive plethysmography, which estimates the volume of the lungs by means of an inductive transducer; articulatory-laryngeal-respiratory coupling effects on airflow, with a pneumotachograph; and acoustic signal produced when speaking, by itself, or in conjunction with kinematic, airflow or EGG data.

Several studies prove the usefulness of non-invasive instrumentation based on different parameters for assessment and visual biofeedback-based therapy. In ⁵⁹ the authors aimed at investigating the feasibility of applying non-invasive instrumentation in order to examine speech physiology in assessment and biofeedback-based therapeutic of stuttering. The results of their experiments suggest parameters which represent respiratory, phonatory, and articulatory physiologic events (speech motor events) may be not only theoretically useful but clinically practical for evaluating and altering perceivable disfluency behaviours through a monitoring process. The important fact is that these kind of procedures allow perturbing deeply rooted behaviours related to speech production. The question may be posed whether not only visual information, but acoustic (sound, music stimulation) or other kind of sensory information could be introduced in order to improve or help parallel biofeedback, allowing the involvement of more parameters and enriching the feedback awareness of the underlying organic events or phenomena. Whether cybernetics must be oriented not only to enhance interaction with the environment (outward interaction in a wide sense), but also to enhance interaction with the own bodily functions and processes (inward interaction) is a very intriguing emerging source of questions and future research, in the way of assessment and therapy in stuttering, as well as in other fields of life and health.

The key point would be to address the recognition of individual biomedical patterns in order to partially, but objectively, support the assessment of stuttering severity, thus reinforcing the information provided by other means, such as self-reports, distributed judgements and, above all, the expertise, experience and sensibility of the therapist.

4.9. Advances in experimental design

Among the parameters described there are some which involve signals. Roughly, a signal is a discrete or continuous collections of coordinates in a euclidean reference system. This is the case of almost every biomedical recording, including blood pressure, skin conductance, brain and muscular electrical potentials, openess measures of vocal folds, relative positions of speech articulation organs, acoustic signals and imaging. Until now, when comparing aspects involving the capture of any of these resulting signals, for instance, in a group difference design experiment, the vast majority of authors tended to collapse entire signals into simple magnitudes, rather arbitrarily or without much justification, with the corresponding information loss (an example could be the very common process of transforming a whole sample statistical distribution into its mean and variance, two numbers which represent certain traits of the starting distribution, but could not surrogate it). Then, quite surprisingly, many of these experimenters implicitly or explicitly hope to find out connections or rules through manipulation of these collapsed data, getting rid of the initial, raw data. At the end of the process the picture is almost invariably the same: non-conclusive results or positive but modest confirmation of the starting hypothesis. These methodologies may be justified in a context where computing conditons are weak. However, as time goes on and a new technological paradigm consolidates, the scientific community should avoid gradually these kind of practices.

Rather that making arbitrary choices in order to drastically reduce data dimensions and simplify analysis, emerging techniques allow us to perform complex data analyses. The point would be to address the recognition of individual biomedical patterns in order to partially, but objectively, support the assessment of stuttering severity, thus reinforcing the information provided by other means, such as self-reports, distributed judgements and, above all, the expertise, experience and sensibility of the therapist.

 I may draw the reader's attention to graphical datamining methodologies in order to skip abusive arbitrary simplifications and to attack problems from another point of view, which we could roughly enounce as follows: do not force yourself to guess any arbitrary simplifications, but rather let the computer device do it for you. In any case, the digital storage, format and transfer allows more extensive and intensive data analysis as never was seen before. Hypothetically, experimental designs would gradually shift from the classical statistical estimation analysis to the complexity analysis of the procedures behind the computer-assisted data prospecting conceived by the experimenter.

Bibliografía/Referencias

NOTES:

Reconocimiento - Nocomercial - SinObraDerivada (CC-by-nc-nd 3.0 o cualquier otra posterior)