«1.Magnitude Estimation of disfluency by stutterers and nonstutterers. Melanie Russell,1 Martin Corley,2 and Robin Lickley1 Speech and Language ...»
1.Magnitude Estimation of disfluency by stutterers and
Melanie Russell,1 Martin Corley,2 and Robin Lickley1
Speech and Language Sciences
Queen Margaret University College
School of Philosophy, Psychology and Language Sciences
University of Edinburgh
The order of the second and third authors is arbitrary. Correspondence concerning this chapter should be addressed to either Martin Corley or Robin Lickley.
2.Summary Everyone produces disfluencies when they speak spontaneously. However, whereas most disfluencies pass unnoticed, the repetitions, blocks and prolongations produced by stutterers can have a severely disruptive effect on communication. The causes of stuttering have proven hard to pin down - researchers differ widely in their views on the cognitive mechanisms that underlie it. The present chapter presents initial research which supports a view (Vasic and Wijnen, this volume) that places the emphasis firmly on the self-monitoring system, suggesting that stuttering may be a consequence of over-sensitivity to the types of minor speech error that we all make.
Our study also allows us to ask whether the speech of people who stutter is perceived as qualitatively different from that of nonstutterers, when it is fluent and when it contains similar types of minor disfluencies. Our results suggest that for closely matched, naturally occurring segments of speech, listeners rate the speech of stutterers as more disfluent than that of nonstutterers.
3.Introduction Research into stuttering often seems to fall at the first hurdle: that of defining what constitutes a stutter, in contrast to the disfluent speech that everyone produces. As of yet there is no consensus on a formal definition: researchers such as Perkins (1995) emphasise the speaker's feelings of loss of control; others, such as Postma and Kolk (1993), prefer definitions in terms of the frequencies of particular types of disfluency.
However, a consensus is slowly emerging that some of the symptoms associated with stuttering can be accounted for within a model of speech developed to account for normal hesitations, speech errors, and self-corrections (e.g., Levelt, 1983).
In this chapter, we provide initial evidence that stutterers appear to be oversensitive when assessing disfluencies common to both nonstuttering and stuttering speakers.
Our research supports a view (Vasic and Wijnen, this volume) which emphasises the role of self-monitoring in the production of stuttered speech.
1.Self-monitoring in stuttering
Self-monitoring can be described as “the process of inspecting one’s own speech and taking appropriate action when errors are made” (Hartsuiker & Kolk, 2001). Levelt’s (1983, 1989) theory assumes that both overt speech and an internal speech plan are monitored. Postma (2000) summarises a number of common speech errors and identifies evidence for two types of self-monitoring: overt speech repairs (where speakers correct themselves mid-utterance) support the monitoring of external speech, whereas covert repairs (where there is no overt error, but a repair can be inferred from a hesitation in the speech output) supply evidence for the internal monitor. In fact, evidence suggests that the repair is often ready before the error is articulated (e.g., Blackmer & Mitton, 1991), and that errors can be made in the absence of articulatory activities or spoken output (for example, when imagining that one is articulating a tongue-twister: Dell & Repka, 1992). Thus the self-monitoring system would appear to have components which are distinct from the monitoring of motor systems (such as articulation) and from the auditory channel. Importantly, the speech that we produce has already been affected by self-monitoring; there is no external record of the original, possibly imperfect, speech plan.
Recent theorists have taken this view on board. For example, Postma & Kolk (1993) hypothesise that stuttering results from covert detection and correction of errors in the articulatory plan through the internal self-monitor. Covert self-correction would prevent the speech error from becoming overt, but would, as a side-effect, compromise the fluency of speech. Evidence for this Covert Repair Hypothesis is inconclusive (for details see: Hartsuiker, Kolk and Lickley, this volume; Vasic and Wijnen, this volume), but still supported by current studies (e.g. Melnicke, Conture and Ohde, this volume, who suggest that not only phonological encoding, but syntactic and semantic processes may be impaired in the formulation of speech by children who stutter).
Blackmer and Mitton (1991) also ascribe a role to monitoring. According to these authors, rapid subsyllabic repetitions, a key symptom of stuttering, occur when the monitor detects a lack of input, and consequently ‘restarts’ previous articulatory movements.
More recently, Wijnen (2000; Vasic & Wijnen, this volume) has placed the emphasis entirely on the self-monitoring system, by proposing that stuttering is the direct result of an overvigilant monitor. Paradoxically, the repairs made often introduce disfluencies rather than prevent them: “stutterers stutter because they try to avoid it” (Wijnen, 2000). Such a view can be easily extended to account for aspects of stuttering such as context-dependency and linguistic distribution.
These proposals have in common the assumption that stuttering is related to selfmonitoring; they also share, to a greater or lesser degree, the entailment that there is a continuity between stuttered and normal disfluencies (in contrast to, e.g., Perkins, 1995). Arguably, the most parsimonious view is that of Vasic and Wijnen (this volume); since there are no differences in planning processes (Postma & Kolk, 1993) or timings (Blackmer & Mitton, 1991) between stutterers and nonstutterers, all differences between the two groups must be attributed to the self-monitor. Given an appropriate experimental paradigm, we should be able to find direct evidence for the self-monitor’s sensitivity in those who stutter. By a similar process of inference, we
would expect there to be continuity between the speech of stutterers and nonstutterers:
it is not errors in planned speech, but how many repairs are initiated, which differentiates the two groups.
2.Sensitivity of the self-monitor
According to Vasic and Wijnen, there are three specific ways in which the speech monitor may be ‘over-sensitive’ to (potential) speech errors. Firstly, too much cognitive effort may be invested in monitoring. Secondly, the focus (as distinct from effort) of the monitoring system may be rigid and unadaptive. Thirdly, the threshold of the monitor may be too low: a ‘hypersensitivity’ to minor speech distortions that non-stutterers would tolerate (or in other words regard as within the bounds of ‘normal’ speech) increases the likelihood of stuttering. The first two assertions are addressed in Vasic and Wijnen’s chapter; in this chapter we focus on the third.
There are three basic proposals for the nature of the self-monitoring system. The first (Levelt, 1983, 1989) supposes that the mechanisms (at the conceptual, phonetic, and auditory levels) which understand language produced by others are shared with the self-monitoring system. The second (Laver, 1973, 1980) assumes multiple monitoring devices attuned specifically to production, including the potential to monitor the articulatory motor processes themselves. A third view (MacKay, 1987, 1992a,b) suggests that error awareness arises from the prolonged activation of otherwise uncomitted nodes in the system for speech production. In an extensive review, Postma (2000) concludes that current evidence largely favours the view of Levelt (1983, 1989) in which the systems responsible for language perception and for self-monitoring are shared. If we accept this view, then people who stutter should show increased sensitivity to disfluencies in others’, as well as their own, speech. In the simplest case, this sensitivity would be manifest whatever the provenance of the disfluent speech -- i.e., whether it is uttered by a stutterer or a nonstutterer.
The current study addresses this issue by eliciting, from a group of stutterers and a comparison group of non-stutterers, ratings of the ‘severity of disfluency’ of recorded speech fragments. The fragments are excerpted from recordings made of dialogues between pairs of stutterers, and between matched pairs of nonstutterers. This allows us to simultaneously address the second, continuity, assumption of many singlemodel accounts. Few studies have directly assessed the sensitivity of people who stutter to dysfluency in the speech of others. Postma and Kolk (1992) come close, by comparing the abilities of people who stutter and fluent subjects to detect errors (rather than disfluencies) in sequences of CV and VC syllables produced by another speaker. Their finding was that people who stutter were less successful than controls in detecting errors under these conditions. In addition, they found that the two groups did not differ in their ability to detect their own errors in the production of CV and VC sequences. The results are taken as evidence that self-monitoring via auditory feedback is not impaired in people who stutter. In our study, we ask listeners to rate severity of dysfluency, rather than error, in samples of spontaneous speech, rather than non-word strings.
3.Continuity between stuttered and normal disfluencies
To some researchers (e.g. Bloodstein, 1970), the difference between the clinical disorder of stuttering, and “normal” speech disfluency is simply a matter of degree.
Stuttering is recognised by the frequency and severity of syllable-sound repetition.
“There is no test for determining the precise point at which speech repetitions stop being ‘normal’ and become ‘stuttering’. We cannot specify where the wall of an igloo ends and the roof begins. It is not a scientific question” (Bloodstein, 1970). In order to strengthen his argument, Bloodstein (1970) describes what he calls the “Consistency Effect”: the distribution of disfluencies in the speech sequence is supposedly similar for stutterers and nonstutterers. Cross (n.d.) agrees that a categorical differentiation between stutterers and nonstutterers is both unnecessary and invalid, because the nature and degree of the problem vary from one individual to the next. He concludes that the issue is not whether the person is a stutterer or not, but whether the form or frequency of speech disruptions interferes with their ability to convey a message.
However, Perkins (1990) insists that a qualitative categorical distinction does exist between stutterers’ and nonstutterers’ speech. He suggests that there are two definitions of stuttering. The observer’s viewpoint corresponds to the continuity hypothesis, whereas the stutterer’s viewpoint corresponds to a categorical judgement.
According to this perspective, speakers know when they stutter, but listeners can only guess. So, disfluency in nonstutterers is concerned with the motor control aspects of speech, whereas disfluency in stutterers seems to involve additional psychological aspects such as loss of control and feelings of helplessness.
In order to disentangle these views, the current study obtains ratings of fluent and disfluent speech fragments recorded from dialogues between stutterers and between nonstutterers. We should be able to ascertain whether there is a general distinction to be made between stutterers’ and nonstutterers’ speech, and (based on Levelt’s, 1983, view of self-monitoring outlined above) whether stutterers perceive a discontinuity where others perceive a continuum.
4.The Present Study
The present pilot study investigates the phenomenon of stuttering perceptually, in contrast to previous work (e.g., Vasic & Wijnen, this volume) which has posited selfmonitoring accounts of stutterers’ speech production. In the experiment reported in this chapter, we asked judges who stuttered to rate the fluency or otherwise of short extracts from recordings made in naturalistic circumstances of dialogues between pairs of stuttering participants or pairs of nonstuttering controls. For each type of dialogue, half of the extracts were of fluent speech, and half were of mildly disfluent speech, where the onset of a word was repeated a single time. We would not expect either set of judges to rate extracts obtained from dialogues between stutterers as more disfluent overall than those obtained from nonstutterers’ dialogues; we expect there to be little or no qualitative difference between the speech of the two groups. However, to test Vasic and Wijnen's hypothesis directly, the ratings given by our judges were compared with those from a second group of judges without stutters. If Vasic and Wijnen are correct, the judges who stutter should be more sensitive to disfluency.
This sensitivity could manifest itself in one of two ways: if the judges who stutter detect and are sensitive to minor infelicities in the fluent speech extracts, we might expect them to rate these (as well as the disfluent samples) as worse. On the other hand, an increased sensitivity to disfluency may make people who stutter likely to differentiate more between fluent and disfluent speech.
There are two justifications for the approach taken here: firstly, we avoid prejudging whether disfluent speech should be considered as ‘normal’ or ‘stuttered’, an absolute distinction which many researchers dispute; and secondly, if we accept Levelt’s view that the processes responsible for self-monitoring are also responsible for the processing of others’ speech, we are in a position to directly compare the sensitivities of stutterers and nonstutterers to disfluencies in speech. The approach relies on using a rating system which is sensitive enough to capture small differences in listeners’ perceptions of the fluency of recorded speech. We have chosen to use Magnitude Estimation, an approach used increasingly in linguistic studies where fine judgements are required, which we outline below.