This paper reflects a number of thoughts that are still very much in progress. It is submitted to the congress participants for their comments, critique and suggestions. In developing these thoughts, I will be using information obtained from a wide variety of sources. Because this forum can be considered the electronic equivalent of a congress presentation, I will use a "congress strategy" for referencing these sources. That is, although I will name sources when the information is very specific, most will remain unnamed in order not to overwhelm the reader with a long list of references. Similarly, when referring to some of the neuroimaging data obtained in our lab that have helped shape these thoughts, I will refrain from detailed methodological explanations, and concentrate on the main observations and implications, as I see them. I count on the discussion to provide space and time for further details, if needed.

Many authors have argued that stuttering can best be seen as a multidimensional disorder of speech fluency. By that, these authors usually mean that it is difficult, or even impossible, to characterize stuttering along a single dimension, such as a psychological, physiological, or environmental disorder. It is generally implied that such a multidimensional approach allows a much better understanding of the complexity of stuttering. I will argue in the present paper that a detailed discussion of the merits of multidimensional models can not be separated from a careful consideration of the level of observation at which the data in support of the models have been obtained. Furthermore, I will argue that even if multidimensional models of stuttering are most appropriate to reflect the complexity of the disorder, this in itself does not necessarily imply that the cause of stuttering also needs to be complex (in the sense of multifactorial). Finally, I will bring in my own bias even more strongly by trying to argue that a true understanding of the causal variable(s) of stuttering will only be gained through the adoption of a neurophysiological perspective.

I think that few will disagree with the statement that the ultimate goal of a multidimensional model of stuttering is to provide a better understanding of the nature and complexity of the speech disorder. However, it is important to understand that this can occur at a number of different levels. At one level, one can try to characterize a disorder that affects humans by the observable "symptoms" and their consequences. You can call this a descriptive level of understanding. Such models of stuttering can incorporate descriptions of 1) observable behaviours (behavioural dimension), including speech disfluencies, various oral and non-oral concomittant behaviours, slower reaction times, and articulatory discoordination patterns, among others; 2) social-psychological behaviours (socio-psychological dimension) including avoidance behaviour, assertiveness, personality characteristics, locus of control characteristics, and speaker-listener interaction patterns, among others; and 3) cognitive variables (cognitive dimension), including psycholinguistic, academic, and anticipatory variables, among others.

The information gained by characterizing stuttering at the descriptive level is very important in a number of aspects. First of all, it allows researchers and clinicians to identify the disorder and to differentiate it from other disorders. Secondly, careful observation of the symptoms, and interpretation of these symptoms in light of what we know about, for instance, human physiology or psychology, may provide important clues as to what may cause the disorder. Thirdly, and importantly, a careful consideration of the various dimensions along which the disorder affects the person is crucial in planning comprehensive clinical intervention. However, it is important to note that, at this level, most if not all long-term disorders that afflict humans can be characterized as multidimensional. For instance, diseases such as Parkinson's, amyotropic lateral sclerosis, diabetes, and asthma not only have physical symptoms, but also will have considerable effects on the person's social-psychological and cognitive characteristics. Few researchers or clinicians, however, will claim that all of these dimensions necessarily need to be included in defining the cause of these disorders.

In my opinion, people working in the area of stuttering, myself included, have not always made a clear distinction between this descriptive level of characterizing stuttering and the variables that necessarily need to be considered when speculating about the cause of the disorder. An example of this is the Demands-Capacity Model (DCM) as proposed by Starkweather and his colleagues (Starkweather, Gottwald, & Halfond, 1990). In this model, stuttering is said to develop as a result of an imbalance between a child's capacities for fluent speech and the demands placed on these capacities, mainly from the child's environment. Although Starkweather has argued on a number of occasions that his DCM model is not a theory of stuttering, but rather an organizational framework for clinical intervention, it has been very tempting to consider the discrepancy, in itself, as the cause of stuttering (for instance, see (Adams, 1990). For instance, within the model stuttering apparently can result from discrepant capacities-demands, even when the child has age-appropriate capacities for fluency.

I think that one of the difficulties with models like the DCM has been the failure to make a clear distinction between trigger variables and causal variables. A large number of events may serve as triggers for the development of stuttering. Many examples of such triggers can be found described in the literature, such as advanced language use by significant people in the child's environment, academic pressures, fearful situations, traumatic experiences, sudden environmental changes, demanding parenting styles, etc. However, I, as do many other people, do not believe that such triggers are sufficient to cause stuttering. Children do not start stuttering simply because their parents use complex language. Many children who do not stutter grow up in similar linguistically demanding environments. In my opinion, many of the demands discussed by Starkweather fall into this "trigger" category, and, thus, I agree with him that the DCM says "little or nothing about the cause of stuttering" (p. 263 - italics added). Something else must be present for the trigger to work on. I strongly believe that that "something else" will be found in the neurological make-up of those who stutter, but more on that later.

While multidimensional models are excellent tools to explain the great behavioural diversity seen among individuals who stutter, it falls short explaining why the various interactions among the assumed dimensions and their subcomponents should result in that one phenomenon that characterizes all those who stutter: a motor disruption of speech fluency. In other words, why would the hundred and one ways in which demands can exceed capacities all result in highly similar speech disfluencies in different children? And why do not all children who are experiencing similar discrepancies stutter? In order to try to answer these questions, one has to build models at an etiological level. Ideally, at this level one should only include variables for which cause and effect relationship with stuttering are reasonably well established. Of course, this not being an ideal world, and human behaviour being famous for its complexity, this is not always possible and a certain amount of speculation will be necessary. However, such speculation (or "dreams" as Cairns-Smith (1996) calls them) should be driven by at least two motivations. First of all, the fewer variables that need to be included to explain stuttering, the better. In Isaac Newton's words when paraphrasing "Occam's razor": "We are to admit no more causes of natural things than such as are both true and sufficient to explain their appearances". Secondly, the role of variables included in the model should be congruent with what we currently know about brain-behaviour interaction.

I assume that very few will have an argument with my call to keep the explanation of stuttering as simple as possible. We obviously still have a long way to go. To the best of my knowledge we have not been able to identify a single variable that can be considered to be either necessary or sufficient as cause of stuttering (including discrepant capacities-demands).

More disagreement may come from the second criterion that places the focus on the neural system as an explanatory framework. For sure, the idea that the neural system should have a central role in any explanation of stuttering is not new. In outlining a theoretical framework for stuttering research, Zimmerman and colleagues (Zimmermann, Smith, & Hanley, 1981) already suggested that any explanation of stuttering should clarify how variables hypothesized to cause stuttering influence the "final common pathway" (i.e., motor neurons). I would like to go a step further and suggest that because the neural system is the central driving force behind everything we think, feel, and do, true understanding of the etiology of stuttering will only emerge if we understand the neural bases of human behaviour, and more importantly, if we recognize its central role by placing it at the very centre of our theoretical model. This is not a trivial task. Despite the tremendous gains in our ability to study how the neural system regulates human behaviour, these advances in many instances have at most only allowed us to appreciate the beauty and complexity of the system, rather than lead to any new and proven insights into the causal mechanisms that underlie specific aspects of the behaviour under scrutiny. Therefore, I will not even attempt to suggest a new theory of stuttering in this paper. But I hope that the current thoughts will help in pointing out the right direction for the development of such a theory.

In another place (De Nil, 1997) I have argued that one of the reasons why we observe interindividual differences among people who stutter, in their behaviour, and how they handle and react to certain situations, can best be understood by contemplating the role of the brain as a "filter" of incoming information. I've presented a model which stressed that any input received by the speaker (whether consciously perceived or not) in most cases will only affect the output (behaviour, in its most broad interpretation) after being filtered and processed by the brain. How the brain filters and processes the incoming information will depend on its innate and/or acquired anatomical and functional characteristics, which may (and in many instances will) differ from person to person. These characteristics can be relatively stable (as for instance in certain psychiatric conditions), or temporary (for instance when under the influence of psychoactive drugs).

Seen in this light one can understand, for instance, why an environment in which fast speaking rate is the norm may in one child result in stuttering, while another child in a similar environment may herself develop advanced speaking skills. If one tries to argue that the fast speaking rate, as an environmental demand, is the "cause" of stuttering, it would be hard to explain why stuttering develops in one child and not in the other one. Moreover, one is bound to be confronted with cases where children seem to develop stuttering without being bombarded by fast language. It is tempting then to look for other variables (for instance advanced language levels), which may pose similar demands on the child. And pretty soon, we are talking about a multitude of variables that may "cause" stuttering. However, if one shifts his focus to the neurological apparatus that sits between the input and output, it becomes evident that the nature of the input is less important than how the central nervous system is able to process that input. Consequently, studying the "demand" (= input) side of the equation will not reveal the true cause of stuttering (= output). If the demands are not a good place to look for the "cause" of stuttering (although they certainly may help us to formulate some decent hypotheses), a logical conclusion seems to be that one should take a close look at the "capacities". I would agree as long as it is clear what one understands under "capacities". I have some difficulties with the view that the onset of stuttering depends on a variety of capacities (cognitive, linguistic, motor, etc.) which may lead to stuttering, either separately or in combination. Given the relative uniformity with which stuttering manifests itself in children and even adults (after all, although it may be quite difficult to identify individual moments of stuttering, there generally is considerable agreement on identifying an individual as someone who stutters), it would seem to me that there has to be a single (and, I would argue, neural) factor that lies at the root of the problem. Let me state right away that the term "single factor" does not imply that the factor necessarily is a simple one. Indeed, there is ample evidence in neuroscience that human functions, for instance memory or language, are not subserved by an anatomically localized and identifiable single structure, but rather depend on complex neural systems which are distributed throughout the brain. For instance, it would be an enormous oversimplification to say that memory is located in the hippocampus, or that language is located in Broca's area, or in Wernicke's area, or even in the left hemisphere. When using the term "single factor" in talking about the cause of stuttering, I'm referring to something akin to the "fluency generating system" suggested by Watson & Freeman (1997). Indeed, only such a system's approach to stuttering, in my opinion, can explain why neurogenic stuttering, which can be remarkably similar to developmental stuttering in its manifestation, has been observed following trauma to such a wide variety of neural structures.

In our own research work, my colleague Dr. Kroll, and I have concentrated primarily on attempting to elucidate the neural mechanism that may lead an individual to develop stuttering. In recent years, we have used primarily functional neuroimaging to help us in this search. This work, of course, needs to be interpreted against the background of the research work done by other investigators who have pursued the same goal, often using different methodologies. This is not the place to go into a lot of detail of this work, but I would like to discuss briefly a number of observations that, I think, are important in shaping our current thinking about the neural basis of stuttering.

In our positron emission tomography studies (De Nil, Kroll, Kapur, & Houle, 1996; De Nil, Kroll, Kapur, & Houle, submitted; Kroll & De Nil, submitted; Kroll, De Nil, Kapur, & Houle, 1997) we have used very simple language tasks, such as single word reading and verb generation, to investigate differences in how stuttering and nonstuttering individuals process such tasks. The idea is that by using such simple tasks, rather than more complex language tasks, we have better experimental control over the cognitive and sensorimotor processes used during task completion. We believe this will allow us to narrow down the number of interpretations that can give to our results, thereby helping us to formulate more focussed hypotheses for further investigations. Furthermore, it is assumed that whatever neural processes are responsible for stuttering will be active during simple as well as more complex language tasks. In an attempt to identify some of the "invariant' neural characteristics of stuttering, if any are to be found, the strategy we have used is to investigate stuttering subjects prior to treatment and immediately following treatment. The rationale behind this is that although behavioural changes certainly may result in observable changes in neural activation patterns, innate neural processes, those that are more or less hard wired into the system, do not change overnight, if ever. Our current findings have supported some of these assumptions. I will briefly review some of these findings in the following paragraphs, with the cautionary note that some of these observations are based on preliminary analysis of very recent data.

Stuttering subjects clearly show different neural activation patterns compared to nonstuttering subjects even during silent reading of single words, when no stuttering is reported. While silent reading of single words results in bilateral activation in both subject groups, activation in the nonstuttering speakers is biased toward the left hemisphere, while in the stuttering subjects, the bias clearly is toward the right. So far, it appears that the focus of activation differences between the two groups is located in the frontal cortex (rostral to the central sulcus), which makes sense given the fact that stuttering is primarily a motor disruption of speech. The fact that such between-group differences can be observed even during a task as simple as silent reading of words, suggest that the neural activation differences are not just a result of the presence of stuttering in speech, but reflect a fundamental, innate or early acquired, characteristic.

As can be expected, the differences in neural activation between stuttering and nonstuttering adults are present during oral speech as well. So far, we have been able to observe right hemisphere biased activation in stuttering speakers during oral reading of words and during self-generation of verbs following presentation of a noun. In every instance, as with silent reading, the activation is primarily focused in the frontal cortex, especially implicating cortical areas involved in the motor control of movements.

An important observation is the fact that following intensive behavioural treatment for stuttering, when the subjects are essentially fluent using their learned fluency skills, the basic pattern of right-biased activation does not seem to change dramatically. However, on top of that (possibly invariant) activation pattern, there appears to be additional neural activation that involves primarily the sensorimotor cortex in the left hemisphere.

The picture that emerges seems to be that there are fundamental differences in the way stuttering and nonstuttering subjects process language and/or produce speech. The precise relationship between these differences and the observed speech disfluencies is still unclear, although the observation that the differences primarily involve motor cortex may point to a potential causal relationship. It is tempting, of course, to think that the observed activation patterns may be part of the "fluency generating system" mentioned before. Contrary to what others have reported previously, these differences in activation patterns do not appear to become completely "normalized" following treatment. Rather, clinical treatment, such like the 'Precision Fluency Shaping Program' based approach used in our studies, seems to result in potentially compensatory activation which is overlaid on top of seemingly invariant neural processing mechanisms. This seems to be congruent with a number of (admittedly selected) well-established observations about stuttering:

1. stuttering is more easily managed in young children. Given the greater plasticity of the brain in younger children, one can see how it may be easier for compensatory mechanism to replace existing processes.

2. Stuttering in adults usually is a chronic problem that can be controlled, but seldomly disappears completely. Again, this seems to make sense given the reduced plasticity in adulthood. Atypical neural processes are there to stay, so to speak, and continue to exert their influence during speech.

3. Most adults who stutter report that following treatment they need to continue working on their speech skills in order to keep them fresh. This observation also is related to the previous one, and has to do with reduced brain plasticity and the resultant reduced ability to automatize new motor skills.

4. Regardless of the type of treatment used, some clients will relapse more or less upon completion of the treatment. In my opinion, this has to do with the observation that new techniques compensate, but do not replace existing neural patterns. Depending on the strength of these atypical patterns, they will continue to influence (invade) speech production, making prolonged use of compensatory techniques difficult or nearly impossible. The extent to which compensatory activation is able to more or less overwrite innate patterns may to a large extent depend on an individual's brain plasticity characteristics. This may help to explain why some people complain that treatment techniques that were initially successful seem to lose their effectiveness, despite continued efforts to practice them.