2003 IFA Congress: Montreal, Canada

A Temporal Analysis of Disfluencies in Children Who Stutter Close to Onset and Controls

Patricia Sandrieder, Ulrich Natke, Melanie van Ark, Reinhard Pietrowsky and Karl Theodor Kalveram
Heinrich-Heine-University Diisseldotf, Institute of Experimental Psychology, Universitiitsstrasse I, 40225 Diisseldorf, Germany


The purpose of this study was to investigate the temporal proportion of normal disfluencies in children who stutter (CWS) and those who do not (CWNS). The participants of the study were 12 children who do stutter and were close to onset of stuttering, and a control group. The frequency and duration of stuttering-like disfluencies (SLD) and other disfluencies (OD), including pauses were analyzed from audio- and video-taped speech samples consisting of at least 1000 syllables. No significant differences in the proportion of OD were found for CWS and CWNS or for CWS with or without remission. OD are shown to be highly significant dependent on age.

  1. Introduction
Interruptions in the flow of speech can be differentiated into stuttering-like disfluencies (SLD) and the so-called other disfluencies (OD). The frequency of SLD has been regarded as the most important feature in diagnosing stuttering and determining severity of stuttering (Van Riper, 1971; Ambrose & Yairi, 1999). Their duration is considered an important parameter in evaluating the severity of stuttering (Ambrose & Yairi, 1999). SLD are less frequently observed in speakers who are not regarded as exhibiting stuttering (Ambrose & Yairi, 1999).

OD are found in the speech of all speakers. They may have a functional value: Starkweather assumes that they give the speaker the opportunity to gain time for the speech-planning process (Starkweather. 1987; Starkweather & Givens-Ackerman, 1997). The assumption is that older children show a larger proportion of OD in their speech than younger ones because their speech becomes more complex (Kowal, O’Conell & Sabin, 1975; Starkweather, 1987). In German- speaking children, data do not exist covering the occurrence of SLD or OD during the development of speech and language in different age groups. For English-speaking children a broad catabase exists (Yairi 1981, Yairi, 1982; Wexler, 1982; Wexler & Mysak, 1982; Yairi, 1984; Yairi & Ambrose, 1999; Pellowski & Conture, 2002).

The occurrence of OD depends on the structure of the utterance. It is known that speech parameters such as utterance length or speech rate influence [16 occurrence of OD (Yairi, 1981). For example, the longer the utterances are, the more likely pauses and revisions are to occur (Deloy & Gregory 1985; Hartmann, Schlicksupp & Jehle, 1989). For children who do stutter (CWS), a relationship was found between these speech parameters and the frequency âof SLD (Haynes & Hood, l978;Yairi, Ambrose & Niermann, 1993; Zebrowski 1995; Zackheim & Conture, 2003). However, an analogue relationship between SLD and OD depending on these speech parameters has not yet been investigated. The utilization of OD to gain time coulc. be advantageous for CWS with respect to the moments of stuttering. It can be hypothesized that CWS having a larger proportion of OD as a percentage of speech time exhibit less SLD than CWS with a smaller proportion of OD. There are indications that the proportion of OD is smaller in the speech of CWS than in children who do not stutter (CWNS) (Sandrieser, 1999). This explorative finding must be confirmed through a group-comparison to gain information about a possible relationship between the development of OD and the development of SLD. One assumption is that in children predisposed to stuttering, a reduced proportion of OD may be one factor in triggering the onset of stuttering. Another explanation might be that a large proportion of OD helps CWS to cope with their stuttering through slowing down the speech rate and facilitating repair-processes (see Postma, Kolk & Povel, 1990). In that way a high proportion of OD could support a remission. This would mean that the proportion of OD might be one factor in predicting stuttering chronicity. To gain more information about the role of OD in the origin and development of stuttering it is necessary to investigate SLD and OD in children shortly after onset of stuttering, and to follow up and check whether the children have had a remission or not.

For clinical and research purposes, SLD and OD are usually measured by counting their frequency. The number of OD in the speech of CWS does not differ significantly from CWNS (Ambrose &Yairi, 1999). To gain information about the functional value of OD, it is necessary to determine their duration. Pauses, as an additional subgroup of OD, also provide time for speech planning. But in studies that do not measure the duration of disfluencies, pauses are usually excluded as a category of OD because of difficulties in reliable identification (Althaus et al. 1993). To receive information about the benefit of OD it is necessary to analyze not only the frequency of OD but also their duration. The analysis of pauses seems to be necessary in gaining information as to whether or not they bear further information about the function of OD.

Our objective was to obtain data from long speech samples about the temporal aspects of normal speech disfluencies in very early childhood stuttering. In addition to providing more insight into the normal development of OD, questions regarding possible changes over time in non- stuttering preschool children were addressed.

The main questions were:

  • Do children who stutter have a smaller proportion of OD in their speech than their non-stuttering peers?
  • Do CWS with a small proportion of SLD in their speech have a large proportion of OD?
  • Do CWS who outgrow their stuttering have a larger proportion of OD in their speech near onset of stuttering than CWS with persisting stuttering?
  • Does the proportion of OD in the speech of CWNS increase with age?
  1. Method
The participants are part of a larger study including 26 children who stutter and a control group. The children in this study were selected only if they were near onset of stuttering. The experimental group consisted of 12 children who exhibited stuttering. The control group included 12 normally fluent children. All children in the stuttering group (CWS) met the following criteria: aged 60 month or under, regarded by both parents as having a stuttering problem, regarded by the examiner as having a stuttering problem, onset of stuttering less than 6 months, German as mother-tongue, and no obvious neurological problems or hearing impairment. Since there is a lack of normative data concerning the frequency of SLD in early stuttering for the German language, the well- established criterion in Anglo-American literature of at least three percent SLD per 100 syllables (Yairi & Ambrose, 1999) was not used. The children received therapy in the two-year period if their parents wished it. 2 CWS received therapy due to articulation or phonological problems, 6 children received therapy due to their stuttering. Therapy enclosed sole counseling of the parents (1), indirect approaches (3) and direct approaches (2). In subsequent examinations “remission” was stated, if the child was regarded by both parents and the examiner as not exhibiting stuttering any more.

Children participating in the control group (CWNS) were matched with the experimental group regarding sex and age. They were reported by their parents as not having a history of stuttering and were regarded by the examiner as not exhibiting stuttering. They spoke German as their mother tongue and had no reported history of neurological problems or hearing impairment. Children with concomitant speech-language difficulties were not excluded. 4 CWS and 5 CWNS showed additional speech-language difficulties

The group of CWS had a mean age of 3.4 years (2.1 - 5.0 years, SD = 0.8) and consisted of 7 boys and 5 girls, the control group had a mean age of 3.5 years (2.2 - 5.2 years, SD = 0.8). All CWS were examined within 6 months of stuttering onset (0.8 - 5.2 months, mean 3.3), with the girls as a group being examined sooner than the boys (girls mean: 2.7 months, boys mean: 3.4 months).

  1. Speech Samples
Two conversational speech samples from each child were audio- and videotaped during a play session between the child and the investigator. The audio signal was digitally recorded (sampling frequency: 22050 kHz, sampling resolution: 16 bits; video: Panasonic, S-VHS,NV-SX30EG; microphone: MKE 2-1053, Sennheiser, Wedemark, Germany). The aim of every session was to collect at least 500 syllables of representative speech through talking about the ongoing play and asking several open-ended questions.

11 of the 12 CWS were examined three additional times within a 2-year period, while one dropped out. The time between the examinations was 8 months. In these sessions, at least 500 syllables were again collected.

  1. Speech Analysis
Speech samples were orthographically transcribed and analyzed using CHILDES and CLAN (McWhinney, 1991), extended with a special coding system for disfluencies. Unintelligible utterances and isolated affirmatives and negatives were excluded. The first 100 syllables were excluded from analysis. The size of the analyzed speech samples varied from 999 to 1036 (mean 1027) syllables. SLD and OD were identified by repeatedly listening to the recording, using the visual presentation of spoken language presented by CLAN, and by watching the videotape in cases of uncertainty. In complex disfluencies each category was coded separately.

SLD were defined as part-word-repetitions, monosyllabic-word-repetitions, and prolongations of sounds and blocks before or within words. OD were defined as interjections, revisions, multisyllabic-word repetitions and phrase repetitions. Pauses were also assigned to the category OD, using Goldman-Eislerâs scheme (Goldman-Eisler, 1961): filled pauses, unfilled conventional pauses and unfilled unconventional pauses. Pauses were analyzed if they exceeded a minimum duration of 0.3 seconds. The frequencies of each type of disfluency were obtained in terms of the number of disfluencies per 100 syllables. In order to test interjudge reliability, 18 speech samples were randomly selected and analyzed by two judges (first and second authors). Wilcoxon-signed- rank~tests revealed nonsignificant differences between mean frequencies of SLD (Z = -0.568; n =18; p = 0.570) and OD (Z = -0.540; n = 18; p = 0.589). Pearson correlation coefficient between the scores of both judges was 0.99 for SLD and 0.93 for OD. Both judges checked all speech samples. Differences were resolved by repeatedly listening to the sample until an agreement between the judges was reached.

In the second step of analysis, the duration of the different types of disfluencies were determined using CLAN. This was done by listening to the audio-signal repeatedly and watching the visualized audio-signal (Sound Forge 4.5). The onset and ending of the disfluency was determined by means of clear visible and audible changes in the waveform. The duration of a disfluency was defined as the “extra-time” (for example: the duration of the syllable repetition “ga-ga-garage” starts with the beginning of the first “ga” and ends with the beginning of the third “ga” which leads to the word “ga-rage”). 7 speech samples were randomly selected and the duration of disfluencies analyzed by two judges (first and third author). The interrater reliability showed a mean difference of 0.0395 seconds.

The proportion of OD without pauses, OD with pauses (OD + P) and SLD was calculated for each participant. This was obtained by summing the disfluency durations and dividing by the participant’s total speaking time. The total speaking time was obtained by summing the durations of each utterance. Speech-rate was calculated by dividing the total speech time through spoken syllables and articulatory rate was calculated by subtracting the duration of all pauses from the total speech-time and dividing by the spoken syllables. Group-differences were tested for significance at the 0.05 level, using one-sided t-test. Correlations were calculated and tested for significance using a one-sided Spearman-test.

  1. Results
All categories of disfluencies occur in both groups, however the frequency of SLD is higher in the group of CWS.

Mean speech rate for CWS was 2.4 syllables per second (1.90 - 3.31syl/sec; SD = 0.40) and for CWNS 2.87 (2.19 - 3.46syl/sec; SD = 0.37). There was no significant correlation between speech rate and age (r = 0.274; p = 0.098). Mean articulatory rate for CWS was 2.61 (2.00 - 3.45; SD 0.41) and for CWNS 3.12 (2.42 - 3.63; SD 0.37). The correlation of age and articulatory rate was not significant (r = 0.333; p = 0.056).


Table 1. Number of disflaertcies per 1 00 syllables for children who starter ( C WS ) and children who do not stutter ( C WNS ) including stuttering-like disflaencies (SLD), other disfluencies (OD) and pauses (P).

The frequency of SLD, OD and pauses per 100 syllables is shown in Table l and Figure l. The data confirm completely the threshold of 3% SLD for CWS and less than 3% SLD for CWNS. In every case the parentsâ opinion about their children’s speech was confirmed by the opinion of the examiner and by the analysis of their speech sample with regard to the criterion of 3% SLD for diagnosing stuttering. The number of OD and P did not differ significantly between the groups (OD: T(22) = 1.001; p = 0.328; P: T(22) = -1.26; p = 0.221). The correlation between age and OD was not significant, neither for the CWS nor for the CWNS (CWS: r = 0.365; p = 0.122; CWNS: r = 0.194; p = 0.273).

The proportion of time disfluent is shown in Table 2 and Figure 2:


Figure 1. Mean number of stutter-like disfluencies (SLD), other disfluencies (OD) and pauses in 100 syllables in the speech of children who do stutter (CWS) and children who do not stutter ( C WNS ).


Table 2. Proportion of disfluencies as a percentage of speech time for children who stutter ( C WS ) and children who do not stutter ( CWNS ). Stuttering-like disfluencies (SLD), other disfluencies (OD) and pauses (P).

No significant differences were found for the two groups of children concerning the proportion of OD (T(22) = -0.196; p = 0.846).

No significant correlation was found between SLD and OD in the group of CWS: children with small proportions of SLD did not show large proportions of OD (r = 0.021; p = 0.474).

No significant differences concerning the proportion of OD were found within the group of CWS: children who had had a remission after two years did not show significantly higher proportions of OD near the onset of stuttering (T(9) = -0.859; p = 0.412)

The comparison of SLD near the onset of stuttering between the two groups of CWS also did not show a significant correlation (r = -0.089; p = 0.391).


Figure 2. Proportion of stutter-like disfluencies (SLD), other disfluencies (OD) and pauses in the speech of children who do stutter (CWS ) and children who do not stutter (CWNS ).

A significant correlation (r = 0.755; p = 0.002) was found between age and proportion of OD in the group of CWNS. A similar correlation was found for the CWS (r = 0.720, p = 0.004) and the whole group (r = 0.755 ; p = 0.002).

Pauses were analyzed and gave the following information:
  • According to the other categories of OD we found that CWNS do not have significant differences in the proportion of pauses than CWS (T(22) = 0.098; p = 0.923).
  • CWS with remission did not show larger proportions of pauses near the onset of stuttering: (T(9) = -0.851; p = 0.417).
Only the group of CWS shows a correlation between age and proportion of pauses (CWS: r = 0.594; p = 0.021; CWNS: r = 0.000; p = 0.500).

The clear correlation of age and OD changes when pauses are included in the OD: we found a weaker correlation between age and OD+P for the whole group (r = 0.604; p = 0.001) and for the CWS (r = 0.734; p = 0.003) but no significant correlation for the CWNS (r = 0.434, p = 0.80).

  1. Discussion
The study confirmed the threshold of 3 percent SLD per 100 syllables as an adequate number in diagnosing stuttering in children (Ambrose & Yairi, 1999, Pellowski & Conture 2002). The maximum number of 2.34 SLD for a CWNS support the necessity to analyze large enough speech samples and to collect further information beside the number of disfluencies in a diagnostic setting. Fluctuation in the occurrence of SLD might otherwise lead to wrong conclusions.

The findings about the frequency of disfluencies are Very similar to those of other studies (Mysak 1982; Yairi, 1982; Yairi & Ambrose, 1999; Pellowski & Conture, 2002). Therefore it seems appropriate to assume that the occurrence of disfluencies does not have language-related differences, at least for English and German.

The finding that parents are able to identify stuttering in their child’s speech was also confirmed (Yairi and Ambrose, 1999).

The assumptions concerning the dependency of SLD on the proportion of OD could not be confirmed with this group of children. The CWS did not have a smaller proportion of OD in their speech. A look at the ranges (Tab. 2) shows quite similar values for both groups with respect to the category of OD as well as pauses. The big variability concerning the proportion of OD indicates the dependence of OD on other factors above all linguistic parameters. In view of a more complex event it seems necessary to examine a larger group of children and to select subgroups with similar linguistic profiles. In this study we did not exclude children with difficulties in their speech-language development. The other possibility is that the development of OD and SLD are not connected, and that other factors have greater influence than the functional value of OD for the speech-planning process. This might explain the absence of correlation between remission and proportion of OD near the onset of stuttering. If the occurrences of OD are explained as part of the speech-planning process, it is necessary to take a closer look at the different levels of speech planning. A methodological problem was the duration of this study: Yairi and Ambrose (1999) were able to show that recovery is a slow and gradual process and that children had remissions in the third and fourth year of their study. Another weak point is the possible influence of therapeutic intervention on the development of OD and SLD. Ethical reasons made it impossible to hold therapy back from children if their parents wished it.

Nevertheless, we were able to show that the proportion of SLD does not correlate significantly with the chance of remission, a finding that was first reported by Yairi and Ambrose (1999).

The hypothesis that the proportion of OD increase with age was a clear result of this study and strengthens the idea of a functional value of OD in the speech of children. It indicates that there is a development of OD and that the findings of other studies that investigated the number of disfluencies (see for example Kowal et al., 1975) can be supported by this finding. The fact that the group of CWS shows the same correlation weakens the idea of a stuttering-specific development of OD. Only the analysis of duration was able to show the increase of OD with time - this finding was not visible after the analysis of the number of OD.

The role of pauses as one category of OD is not unequivocal: pauses corroborate the missing correlation of a smaller proportion of OD in CWS and a larger proportion of OD in CWS with remission and they confirm the idea of an increasing proportion of OD with age for the whole group as well as the subgroups. The fact that the statistically significant correlation between age and OD+P becomes weaker for the whole group including the CWS, and disappears for the CWNS cannot yet be explained. The results of a bigger sample-size (n = 52) will show whether this effect is stabil or not. If the proportion of pauses again weakens the correlation between OD and age, then we will need further studies to examine whether pauses are a category of OD or whether they have to be treated separately. If the effect it not stabil, the result concerning the pauses might be due to the small number of children in this study. This would confirm the functional character of OD, including the pauses.

Supported by grant Ka 417/28-2 from Deutsche Forschungsgemeinschaft (DFG).


Althaus, M, Goorhuis-Brouwer, S.M., Kloosterman, H., Vink, H & Minderaa, R.B. (1993) Verlasslichkeit der Beurteilung horbaren Stotterverhaltens. F olia Phoniatrica, 45, 261-267
Ambrose, N. G. & Yairi, E. (1999) Normative Disfluency Data for Early Childhood Stuttering. Journal of Speech, Language and Hearing Research, 42, 859-909
DeJoy D.A. & Gregory H.H. (1985) The relationship between age and frequency of disfluency in preschool children. Journal of Fluency Disorders, 10, 123-135
Goldman-Eisler, (1961) The continuity of speech utterances: its determinants and its significants. Language and Speech, 4, 220-231
Hartmann, S, Schlicksupp, P. & Jehle, P. (1989) Der Zusammenhang zwischen dem Alter und der Sprechflijssigkeit von fliissig sprechenden Vorschulkindern. Sprache Stimme Gehoâr, 13, 26-31
Haynes, W.O. & Hood, S.B. (1978) Language and disfluency in normal speaking children from discrete chronological age groups. Journal of Fluency Disorders, 11, 57-74
Kowal, S., OâConell, D. & Sabin, E. (1975) Development of temporal patterns and vocal hesitations in spontaneous narratives. Journal of Psycholinguistic Research, 4, 195-207
MacWhinney, B (1991) The CHILDES Project: Tools for Analyzing Talk. Hillsdale, NJ: Erlbaum.
Pellowski, M.W. & E.G. Conture (2002) Characteristics of speech disfluency and stuttering behaviors in 3- and 4-year-old children. Journal of Speech, Language and Hearing Research, 45, 20-34
Postma, A., h. Kolk & D.-J. Povel (1990) Speech planning and execution in stutterers. Journal of Fluency Disorders, 15, 49-59
Sandrieser, P. (1999) Unfliissigkeiten im Sprechen von stottemden und nichtstottemden Kindern. Forum Logopiidie, 13, 13-16
Starkweather, C.W. (1987) Fluency and Stuttering. Englewood Cliffs, N.J.: Prentice Hall
Starkweather, C.W. & Givens-Ackerman (1997) Stuttering. ProEd, Austin
Van Riper, C. (1971) The nature of stuttering. Englewood Cliffs, N J : Prentice Hall
Wexler, K. B. (1982) Developmental disfluencies in 2-, 4-, and 6-year-old boys in neutral and stress situations. Journal of Speech and Hearing Research, 25, 229-234
Wexler, K.B. & E.D. Mysak (1982) Disfluency characteristics of 2-", 4-, and 6-year-old males. Journal of Fluency Disorders, 15: 37-46
Yairi, E. (1981) Disfluencies of normally speaking two-year-old children. Journal of Speech and Hearing Research, 24, 490-495
Yairi, E (1982) Longitudinal study of disfluencies in two-year-old children. Journal of Speech and Hearing Research, 25, 155-160
Yairi, E. (1983) Disfluencies at the onset of stuttering. Journal of Speech and Hearing Research, 27, 154-159
Yairi, E., Ambrose, N.G. & Niermann, R. (1993) The early month of stuttering: A developmental study. Journal of Speech and Hearing Research, 36, 521-538
Yairi, E. & Ambrose, N.G. (1999) Early childhood stuttering 1: Persistency and recovery rates. Journal of Speech, Language and Hearing Research, 42, 521-538
Yairi, E. & Lewis, B. (1984) Disfluency at the onset of stuttering. Journal of Speech and Hearing Research, 27, 145-154.
Zackheim, C.T. & E.G. Conture (2003) Childhood stuttering and speech disfluencies in relation to children’s mean length of utterance: a preliminary study. Journal of Fluency Disorders, 28, 115-142
Zebrowski, P.M. (1995) The topography of beginning stuttering. Journal of Communication Disorders, 28, 75-91


In preparation for the 2018 World Congress the IFA is implementing Japanese translations of some pages on the site. Choosing Japanese below to see these translations.

Not all pages are translated, but you can use Google translate to see a machine translation using the switch below

Google Translate

Follow the Joint World Congress