2003 IFA Congress: Montreal, Canada

Cross-Linguistic Factors in the Prediction of Stuttering Across Age Groups - The Case of German

Katharina Dworzynski and Peter Howell
University College London, 26 Bedford Way, London WC1H OAP


Cross-linguistic research can establish whether stuttering patterns are consistently associated with linguistic structures irrespective of their surface form; or whether difficult motor outputs lead to stuttering independent of the linguistic unit they occur in. A dissociation can be achieved because the same motorically-difficult structures may appear in different linguistic units in different languages. Linguistic factors known to predict dysfluencies in English are investigated in German children and adults who stutter (using Brown’s four factors and J akielski’s index of phonetic complexity). Some cross-linguistic differences were observed. Children were on the whole less affected by linguistic complexity than adults. Results are discussed in light of current theories of fluency failure.

  1. Introduction
A long line of research has tried to investigate whether there is a consistent pattern to stuttered episodes and which linguistic aspects determine whether a word is more likely to be stuttered (see the review of findings by Brown, 1945 who originally inspired much of this type of research). In 1945 Brown summarized his previous research and identified four basic factors that determined whether words will be spoken disfluently by adults who stutter. The factors are: (1) word class (this has subsequently been interpreted as showing that content words are more prone to stuttering than function words); (2) word length (long words are more difficult); (3) sentence position (words that appear in early positions are more likely to be stuttered); (4) phone the word starts with (words starting with consonants are more difficult than those that start with vowels).

The study of linguistic determinants of disfluencies is not just descriptive in nature, i.e. providing a part of the facts of stuttering, but rather the findings have both scientific and theoretical implications. One important criticism that has recently been raised about such research is the fact that virtually all of these studies have used monolingual English speakers (Van Borsel et al., 2001). Cross-linguistic studies, and the analysis of patterns of disfluencies of bi- or multilingual stutterers, are ways to test the validity of models that hypothesize that certain linguistic variables lead to disfluencies (a point also stressed by Bernstein Ratner & Benitez, 1985). Even though cross- linguistic research is needed to clarify the postulated linguistic factors in stuttering, a review of the literature showed that, for instance Brown’s four factors, have only been investigated (and found to operate in) Norwegian (as cited in Bloodstein, 1995) and Kannada (a Dravidian language, Jarayam, 1983).

The appeal of Brown’s factors is that they can provide a simple and effective measure that could be used as a criterion against which other languages can be compared. In the first part of this paper a re-analysis of findings with regards to German (with a wider child age group than that reported in Dworzynski et al., 2003) for these factors is given and some cross-linguistic reasons for the differences are presented. Though Brown’s factors have the merit that they provide a benchmark for research, they are relatively crude. This means that improved techniques for the investigation of these and other factors are needed. Here a selection of findings from an analysis using Jakielski’s (1998) index of phonetic complexity (IPC) are also reported. For this scheme, words are given sums of scores along eight different characteristics (shown in Table 1):


Table 1. IPC scoring scheme

These factors have been used previously in research into dysfluencies (originally by Weiss & Jakielski, 2001) and are based on MacNeilage and Davis’ (1990) theory of children’s early phonological development. MacNeilage and Davis (1990) assume that these factors are universally associated with phonological difficulty because they originate in different stages of babbling. Weiss and Jakielski (2001) analyzed children’s speech samples and found no relation between IPC and stuttering rate. More recently IPC has been used within the EXPLAN model of fluency failure which predicts that children have a higher function word stuttering rate that is not as affected by phonological difficulty as are content words (Howell et al., submitted). Another investigation looked at the different phonetic complexity of function / content words in an English -/ Spanish comparison (Howell & Au-Yeung, submitted). IPC scores have also been used in English to show that the content words following function word disfluencies in children have, on average, higher phonetic complexity that content words following fluent function words (Howell & Ladd, in preparation).

With respect to Brown’s factors for German adults, it is predicted that due to cross-linguistic differences in syllable structure the stuttering rate of words starting with vowels would be high in German compared with English. The difference in syllable structure between English and German Concerns syllable onsets. In German syllable onsets are obligatory and words starting with vowels (syllable nuclei) have to have an onset added. In these cases a glottal stop is inserted as an onset to that syllable (Rogers, 2000) .Early sentence positions would have a lower level of stuttering than in English, because of differences in sentence structure between the two languages. This is mainly due to the fact that the finite verb and the past participle have rigid positions in German sentences. All finite verbs must be in second position in independent clauses, and non-finite ones in final position (with the exception of cases of extraposition of adjuncts or arguments). In dependent (subordinate) clauses, all verbs are in the final position. This inflexibility can, however, be contrasted with a much greater flexibility with regards to the position of the subject, direct object, and indirect object in German. The reason for this is that in German the case endings will always indicate how the constituents fit together syntactically. This means that arguably more planning is involved in later sentence position for German. The content word, and word length, factors are expected to operate in German in a similar way to English. With respect to German, there is a high frequency of compound nouns which might result in a comparatively higher stuttering rate for German content words. Children were predicted to be less affected by linguistic factors than adults with comparatively more stuttering on function than content words.

Language comparisons of IPC scores should clarify some of the observations obtained with Brown’s factors for German. IPC can help clarify whether the compound nouns have an effect on the overall average phonetic complexity of German content words. It is thus predicted that content word on the whole are on average phonetically more complex and that this would then also become apparent in higher stuttering rates for content words in German compared to English, at least for the adult age group. Furthermore, an index such as this allows a look at the frequency of individual factors in each language. It is expected that factors that occur less frequently within a language will be more difficult and as such will have a higher impact on stuttering for speakers of that language.

  1. Method

For the German speakers there were 26 children (mean age of 9 years with a standard deviation of 2 years) and the adult age group consisted of 15 speakers with a mean age of 29 years and 3 months, standard deviation of 10 years and 9 months. In the English speaking group 16 children were aged 6 to ll (mean age of 8.0 standard deviation of l) and ten adults classified as 18 years and older (mean age 26 years 9 month , standard deviation of 6.2 years). All of the speakers had been diagnosed as a person who stutters by two independent sources (a referring general health practitioner and a speech therapist).

Speech material and transcription

Spontaneous speech samples, usually in conversation with a speech therapist, were recorded. All samples were of a minimum of 2 minutes duration but could be as long as 25 minutes depending on how talkative the speaker was. Transcriptions were carried out using speech filing system (SFS) software. This software was used to allow accurate location of events on an oscillographic display of the waveform. Thus pauses and prolongations can be accurately determined. Data were transcribed using a broad phonetic transcription in fluent regions and a narrow system in regions of disfluencies. Disfluencies that were marked were part-word and monosyllabic whole word repetitions, as well as prolongations and blocks. All words were also classified as function or content words in type.


All words were coded according to the four factors Brown had investigated. A word was scored for every factor Brown had identified as being associated with a higher rate of stuttering. To do this, each factor was examined separately and a word was given a score of 0 or 1 for each factor, according to the following contingencies: O was given for the respective factor when a word started with a vowel, when it was a function word, when the word was shorter than five phonemes, when the word occupied a position beyond the first three in an utterance. l was given for the respective factor when a word starred with a consonant, when it was a content word, when the word was longer than five phonemes, and when the word occupied one of the -first three positions in an utterance. Words were also coded as produced fluently or stuttered.

Additionally all words were given an IPC score: Table 1 gives a breakdown of the eight IPC factors. In respect to factor 1 ‘consonant by place’ every dorsal consonant in a word was given one point whereas other consonantal articulation places received no points. For the second IPC factor ‘consonant by manner’ every fricative, affricate and liquid received one point no point to any other consonants. For the third factor ‘singleton consonant by place’ inter-syllabic relationships are taken into consideration. A point is given to a consonant pair only when in a ‘..VC-CV..’ structure (V and C stand for vowel and consonant slots) the syllable coda consonant has a different place classification to the onset consonant of the following syllable. In factor 4 only rhotic vowels receive a point. Whether the word ends in a consonant or vowel is indicated by factor 5 (‘word shape’). A word is considered long when it has three or more syllables (factor 6). ‘Contiguous consonants’ means that each consonant cluster would be given one point irrespective of the syllable or word position it occupies. Finally for factor 8, ‘cluster by place’, one additional point is given if individual consonants in a cluster are from different places of articulation. Again each word is also classified as function or content word and fluent or stuttered.


Due to the unavailability of another native German transcriber, rather than inter judge reliability a measure of consistency was calculated. For this, 20% of samples were randomly selected for a second transcription and then an alpha measure was obtained. Cronbach’s alpha is a measure of consis;ency. A higher alpha coefficient signifies better consistency. Nunnaly (1978) indicated 0.7 to be an acceptable reliability/consistency coefficient, though lower thresholds are used in some of the literature. For the re-transcribed data alpha values for both fluency judgments as well as content/function word classification ranged from 0.80 to 0.98 which indicated a high level of consistency. The values for the IPC and Brown scoring ranged from 0.82 to 0.95 also indicating good internal consistency.

  1. Results
Individual stuttering rates were calculated, by dividing the number of stuttered words by the total number of words (i.e. stuttered and non-stuttered). Stuttering rates (percentages) for Brown’s individual word factors were calculated in the same way. Figure 1 shows individual characteristics for the two age groups in German:


Figure 1. Along the abscissa individual word characteristics with regards to Brown’s factors are given. Bars represent mean stuttering rate. Difierently shaded bars represent the diflerent age groups (lighter children - darker adults, see legend). The asterisk indicate significant difierences (p<0.05).

An overall ANCOVA (individual stuttering rate was treated as a covariate) indicated a three way interaction between age group, factor, and factor level (F(3, lll)=4.49, p<0.0l). The figure indicates clearly that for the children the factors do not have as much of an impact as for the adults. As predicted the factors vowel / consonant and sentence position were not significant for German.

Both the factors word length and grammatical class are interlinked since content words are the words that are also longer.

To analyze this pattern in more detail and to possibly account for the dramatic difference between function and content word stuttering in German adults, the IPC measure was used. Here the two languages are compared for the same age groups. Each word receives scores as indicated inthe above table and is also marked as being either fluent or stuttered as well as being a function or content word. Mean scores (represented in Figure 2) were calculated from the values on individual words:


Figure 2. Line graph representing mean IPC scores according to age group and fluency (as indicated along the x-axis) separated into grammatical word class and language (as described in the legend). Unfilled markers and solid lines always refer to German cases whereas dotted lines and filled markers refer to English cases.

There is a great deal of information in this figure and the attention will be drawn in this analysis to a few individual details. First the figure shows quite clearly that for both languages content words are phonetically more complex than function words (F(1, 6l)=506.90, p<0.00l). Moreover, stuttered words are associated with higher IPC score than fluent words (F(l, 61)=l8.20, p<0.00l). The interaction that links the left and right parts of the figure together is the word type by language interaction (F(l, 6l)=27.42, p<0.00l). In both. languages function words have lower IPC scores than content words, however the gap is considerably larger in German compared to English. Another significant two-way interaction was between word type and fluency (E(l, 61)=l9.90, p<0.00l). This refers to the fact that stuttered words have a higher IPC score only when the word is a content word in type.

  1. Discussion
Two different methods have been used to assess linguistic difficulty and its relation to stuttering rate in different age groups as well as different languages. The Brown analysis showed that subtle cross linguistic differences in syllable structure and word order could have an effect on stuttering rates. As such it was shown that both the factor vowel /consonant and sentence position did not have a significant impact. This should be treated with caution since it is an observation based on a null result. However, the data do show a similar trend to the English analyses, albeit less pronounced - both words starting with vowels and positioned earlier in sentences were associated with a higher stuttering rate. This means that the hypotheses concerning articulatory tension in vowel onsets and syntactical sentence position (set out in the introduction to the present study) cannot be definitely ruled out. Furthermore a change from children to adults was observed in respect to word type. Function word stuttering decreased and content word stuttering increased with age.

From the current analysis it can be seen that the IPC metric (once divided into function and content words) is a useful tool to further investigate the original factors proposed by Brown (1945). Higher IPC scores were found to be associated with higher stuttering rates, particularly in the adult age group. Moreover the results brought up interesting mediating factors: Function words in both languages had lower IPC values (and more or less identical in both languages) whereas content words had both higher IPC scores, with a particularly large difference between function and content IPC scores for German. A common feature of the German language is the frequent use of compound nouns which is one of the possible reasons for the considerably higher mean IPC score associated with German content words compared to English.

It should be noted that a number of the characteristics in this scheme correspond to the late emerging consonants, multiple syllable and consonant clusters investigated by Throneburg and colleagues (1994) which has also been found to be a useful method to investigate disfluencies (used by Howell et al., 2000, to analyze internal structure of English content words). One of the criticisms that could be raised in respect of the IPC schema is that the factors are not sufficiently independent of one another. There are many factors concerned with consonantal difficulties which are not mutually exclusive. This makes the interpretation of results more difficult since the impact of some of the factors would not be felt as highly because of close correlations with other factors.

Howell and Au-Yeung’s (2002) EXPLAN theory was explicitly developed to account for the increased incidence of disfluencies on function words in early development. In their model they assume that the cause of all stuttering is on the content word. Content words are phonologically more complex than function words in English as in German. Fluency failure is then viewed as a sign that planning and execution processes are out of synchrony. The pattern of children’s disfluencies, also observed in the current investigation, is then interpreted as a way of gaining time to complete the plan for the following content word. In this sense, the emphasis in EXPLAN is on a problem of timing and not, as suggested in the Covert Repair Hypothesis (Kolk & Postma, 1997; Postma & Kolk, 1993) an error-prone phonological system. Function words are repeated for the purpose of gaining more time for completing the plan of the following content word.

The second author is support by the Wellcome Trust


Bernstein Ratner, N. & Benitez, M. (1985). Linguistic analysis of a bilingual stutterer. Journal of Fluency Disorders, 10, 211-219.

Bloodstein, O. (1995). A Handbook on Stuttering. (5th ed.) Chicago: The National Easter Seal Society.

Brown, S. F. (1945). The loci of stutterings in the speech sequence. Journal of Speech Disorders, 10, 181-192.

Dworzynski, K., Howell, P., & Natke, U. (2003). Predicting stuttering from linguistic factors for German speakers in two age groups. Journal of Fluency Disorders, 28, 95-113.

Howell, P., & Au-Yeung, J . (submitted). Phonological difficulty and stuttering in Spanish. Journal of Speech, Language and Hearing Research.

Howell, P. & Au-Yeung, J. (2002). The EXPLAN theory of fluency control and the diagnosis of stuttering. In E.Fave (Ed.), Pathology and therapy of speech disorders. In Current Issues in Linguistic Theory series. (pp. 74-94). Amsterdam: John Benjamins.

Howell, P., Au-Yeung, J ., & Sackin, S. (2000). lntemal structure of content words leading to lifespan differences in phonological difficulty in stuttering. Journal of Fluency Disorders, 25, 1-20.

Howell, P., Au-Yeung, J ., Yaruss, S. & Eldridge, K. (submitted).Phono1ogical difficulty and stuttering in English. Journal of Speech, Language and Hearing Research.

Jakielski, K. J. (1998). Motor organization in the acquisition of consonant clusters. Dissertation / PhD thesis: University of Texas Austin.

Jarayam, M. (1983). Phonetic influences on stuttering in monolingual and bilingual stutterers. Journal of Communication Disorders, 16, 278-297.

Kolk, H. & Postma, A. (1997). Stuttering as a covert repair phenomenon. In R.F.Curlee & G. M. Siegel (Eds.), The nature and treatment of stuttering: New directions. (2nd ed., Boston MA: Allyn & Bacon.

MacNeilage, P. & Davis, B. (1990). Acquisition of speech production: Frames, then content. In M.Jeannerod (Ed.), Attention and performance XIII: Motor representation and control. (Hillsdale: Lawrence Erlbaum.

Nunnaly, J. C. (1978). Psychometric theory. New York: McGraw-Hill.

Postma, A. & Kolk, H. (1993). The covert repair hypothesis: Prearticulatory repair processes in normal and stuttered disfluencies. Journal of Speech and Hearing Research., Vol 36, 472-487.

Rogers, J. (2000). The phonatory correlates of juncture in German. Proceedings of the 5th Seminar on Speech Production: Models and Data: May 1st to 4th, 2000, Kloster Seeon, Bavaria, Germany.

Throneburg, R. N., Yairi, E., & Paden, E. P. (1994). Relation between phonologic difficulty and the occurrence of disfluencies in the early stage of stuttering. Journal of Speech and Hearing Research, Vol 3 7, 504-509.

Van Borsel, J ., Maes, E., & Foulon, S. (2001). Stuttering and bilingualism: A review. Journal of Fluency Disorders, 26, 179-205. ,

Weiss, A. L. & Jakielski, K. L. (2001). Phonetic complexity measurement and prediction of children’s disfluencies:A preliminary study. In Proceedings of the 4th International Speech Motor Conference (pp. 278-281). June 13th-16th: The 4th Nijmegen Conferences on Speech Motor Control and Stuttering.


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