Language and Disfluency in Nonstuttering Children
Language and Disfluency in Nonstuttering Children
Language and Disfluency in Nonstuttering Children
NONSTUTTERING CHILDREN’S
CONVERSATIONAL SPEECH
J. SCOTT YARUSS
University of Pittsburgh, Pittsburgh, Pennsylvania
ROBYN M. NEWMAN
Northwestern University, Evanston, Illinois
TRACY FLORA
Tacoma School District, Tacoma, Washington
Several recent theories have suggested that the production of speech disfluencies is related,
in part, to certain aspects of language formulation. One common characteristic of these the-
ories is the proposal that the same basic mechanism is responsible for disfluency in people
who stutter as well as in people who do not stutter. To learn more about the connection be-
tween language formulation and fluency in nonstuttering speakers, this study examined rela-
tionships between syntactic complexity, utterance length, and disfluency in the spontaneous
speech of 12 normally fluent children (six girls and six boys, age 44–64 months), who pro-
duced 50-utterance spontaneous speech samples during conversations with a speech–lan-
guage pathologist. On average, disfluent utterances were longer and more syntactically
complex than fluent utterances. Discriminant analyses indicated that utterance length in
clausal constituents was the most important factor (among those assessed in this study) for
predicting the likelihood that an utterance would be disfluent. Analysis of subjects’ individ-
ual speech samples indicated that the selected aspects of linguistic complexity could only
classify fluent and disfluent utterances at better than chance levels for a subgroup of sub-
jects. Findings provide one means of evaluating theories that explain speech disfluencies
based on normal language formulation process. Results also emphasize the value of consid-
ering individual patterns in children’s speech fluency, and highlight the need to consider
factors such as the timing of utterances or other contextual factors, in addition to linguistic
complexity, in future studies of children’s speech fluency. ©1999 Elsevier Science Inc.
In recent years, there has been growing interest in studying the relationships
between various processes of language formulation and the production of
speech disfluencies (see review in Bernstein Ratner, 1997). In addition to the
Address correspondence to J. Scott Yaruss, Ph.D., Communication Science and Disorders, University
of Pittsburgh, 4033 Forbes Tower, Pittsburgh, PA 15260, USA. e-mail: jsyaruss@csd.upmc.edu
who stutter (Kadi-Hanifi & Howell, 1992; Gaines, Runyan, & Meyers, 1991;
Logan & Conture, 1995, 1997; Wall, Starkweather, & Cairns, 1981; Weiss &
Zebrowski, 1992; Wilkenfeld & Curlee, 1997; Yaruss, 1999). The majority of
these studies have indicated that longer, more complicated utterances are
more likely to be stuttered than shorter, simple utterances. Although some of
these studies of children who stutter did include a control group of children
who do not stutter (Kadi-Hanifi & Howell, 1992; Logan & Conture, 1997),
none reported direct comparisons of the fluent and disfluent conversational ut-
terances of normally fluent children. In fact, Kadi-Hanifi & Howell (1992) ap-
pear to have assessed only stuttered disfluencies. Accordingly, they reported
that their normally fluent subjects did not produce any “dysfluent syllables”
(i.e., stuttered disfluencies) in their conversational speech samples (p. 161).
Thus, further research will be necessary to determine whether linguistic fac-
tors that influence fluency in stuttering children’s conversational speech also
affect normally fluent children, and whether it is appropriate to utilize the
same linguistic theoretical frameworks for considering both normal and stut-
tered disfluencies.
Additional data regarding the relationship between linguistic factors and
speech fluency in normally fluent children’s conversational speech will also
help evaluate whether utterance length and syntactic complexity can be
viewed as “demands” that might affect children’s speech fluency when
viewed in a “demands and capacities” framework (e.g., Adams, 1990; Stark-
weather & Gottwald, 1990). Specifically, knowing more about how normally
fluent children respond to such language demands will help evaluate whether
increased utterance length and syntactic complexity might play a role in the
development of stuttering (e.g., Gaines et al., 1991; Gordon, 1991; Logan &
Conture, 1995). Although studies utilizing imitation and modeling procedures
revealed the expected result (i.e., that longer, more complicated utterances are
more likely to contain disfluencies), it is still important to assess these rela-
tionships in spontaneous conversational speech.
Finally, one additional issue that deserves some consideration is whether
the relationships between utterance length, syntactic complexity, and speech
fluency that have been found in group-averaged data can also be found on an
utterance-by-utterance basis (Yaruss, 1999). As noted above, previous studies
on the conversational speech of children who stutter have shown that stuttered
utterances are, on average, likely to be longer and more syntactically complex
than fluent utterances. Still, one recent evaluation of individual subjects’ data
(Yaruss, 1999), which attempted to relate the length and complexity of a par-
ticular utterance to the likelihood that that same utterance would contain a dis-
fluency, found that such a direct relationship existed for only a small subgroup
of subjects. A similar discrepancy between group and individual findings was
presented by Gaines et al. (1991). This is notable since many current theories
of stuttering, such as the ones highlighted above, appear to suggest that rela-
LANGUAGE AND DISFLUENCY 189
METHOD
Subjects
Subjects were 12 normally fluent children (six girls and six boys), aged 44–64
months (mean 5 54.7 months; SD 5 5.7 months). These 12 subjects were se-
lected from a larger pool of 24 normally fluent subjects who had been re-
cruited from a local preschool as part of a larger study of children’s speech/
language production skills. Five potential subjects were excluded because
they did not produce 50 utterances in their speech samples that met the criteria
for the Developmental Sentence Scoring procedures (DSS; Lee, 1974), and
two older potential subjects were excluded to allow for closer age-matching
between male and female subjects (6 3 months) and to reduce the overall av-
erage age of subjects to 4 years, 6 months. In addition, five potential subjects
were excluded from the present study because they did not meet a minimum
criterion of five disfluent utterances in their 50-utterance samples (10% disflu-
190 J. S. YARUSS, R. M. NEWMAN, and T. FLORA
ent utterances), selected so that there would be at least a nominal of level dis-
fluency data for statistical comparisons in this study. Although this minimum
criterion might seem somewhat high for normally fluent children, it is impor-
tant to remember that all disfluency types, including normal disfluencies such
as revisions or interjections, were coded in this study. Overall, subjects’ fre-
quency of all types of disfluency in these 50-utterance conversational speech
samples ranged from 3.02% to 7.19% (i.e., between three and seven speech
disfluencies per 100 words of conversational speech), values that fall well
within the normal range of disfluencies for children in this age range. Because
of the relatively small number of so-called “less typical” types of disfluencies
produced by these children, and because of the statistical restrictions that
would be required to separate out disfluencies by type; however, no attempt
was made to determine how the different types of disfluencies might be re-
lated to the linguistic variables examined in this study.
As a part of their participation in the present study, each child’s speech and
language abilities were screened by a licensed and ASHA-certified speech–
language pathologist to verify that none of these 12 subjects in this study ex-
hibited any significant speech, language, or hearing concerns. DSS results
(see below) indicated that the degree of syntactic complexity exhibited by
these subjects during their conversational speech samples was within normal
limits for each child’s chronological age.
Data Collection
Each child was audio/videotape recorded during a 30-min spontaneous, con-
versational interaction with the first or third author. Spontaneous speech sam-
ples were elicited through parallel play using a toy farm and by creating sto-
ries about daily activities of the characters on the farm. All recordings were
made in a quiet room at the child’s preschool away from the child’s regular
classroom.
Video recordings were made using a camcorder (Panasonic AG-455)
linked to a portable S-VHS videotape recorder (Panasonic AG-5700) via a
portable timecode generator (Horita TG-50), which placed a visually apparent
timecode (hours:minutes:seconds:videoframes) on the screen above the child’s
image to facilitate data analysis. Audio recordings were made using a portable
wireless microphone system (Samson MR-1) retrofitted with a directional la-
valiere microphone (Audio Technica AT-831B).
Speech samples for the present study consisted of 50 consecutive com-
pletely intelligible conversational utterances that met the inclusion criteria for
the DSS procedure (see below). In keeping with prior research on this topic,
an utterance was defined as a string of words that (1) communicated an idea,
(2) was set apart by pauses, and (3) was bounded by a single intonational con-
tour (Kelly & Conture, 1992; Meyers & Freeman, 1985; Logan & Conture,
LANGUAGE AND DISFLUENCY 191
1995; Yaruss, 1997b, 1999; Yaruss & Conture, 1996). Following the audio-
video recording session, each child’s speech sample was transcribed into a
customized computer data base developed for this project by the first author,
and specific information pertaining to the fluency and linguistic complexity of
each utterance was determined, as follows:
Speech fluency. Each utterance was analyzed for the presence of speech
disfluencies, based on the broad continuum of “more typical” and “less typi-
cal” disfluencies described by Gregory and colleagues (Gregory, 1986; see
also Yaruss, 1997a). More typical disfluencies are those types that are most
likely to be produced by normally fluent speakers, including repetitions of
phrases (e.g., “I want- I want that”), revisions (e.g., “I want- I need that”), and
interjections (e.g., “um,” “er”). Less typical disfluencies are those types that
are most likely to be produced by individuals who stutter, including repeti-
tions of words, sounds, or syllables (e.g., “l-l-look”), prolongations (e.g.,
“lllook”), or blocks (e.g., “l—ook”). Given that the present subjects were nor-
mally fluent speakers, the vast majority of the disfluencies examined in this
study were of the more typical variety; however, less typical disfluencies were
also coded in order to provide a more complete picture of the relationship be-
tween language complexity and speech fluency in children who do not stutter.
Because the study was designed to examine potential relationships between
fluency and language at the utterance level, and because of the relatively small
number of utterances containing multiple disfluencies, no distinction was
made between utterances that contained only a single disfluency (148 of the
600 utterances) and utterances that contained more than one disfluency (34 of
those 148 utterances).
Utterance length. Based on the findings of Brundage and Bernstein Rat-
ner (1989; see also Yaruss, submitted) that different measures of utterance
length may be related to speech fluency in different ways, utterance length
was measured via a number of different indices. Specifically, length was mea-
sured in terms of the number of words, morphemes, and syllables per utter-
ance. In addition, length was also measured in terms of the number of clausal
constituents per utterance, based on a procedure described by Blake, Quartaro,
and Onorati (1993). According to this procedure, clausal constituents are de-
fined as subjects, verbs, objects, adverbials, and complements (Crystal, Gar-
man, & Fletcher, 1989). This measure was utilized in order to provide an indi-
cation of utterance length that is presumably related to syntactic complexity
(see also Logan & Conture, 1997; Yaruss, 1999).
Syntactic complexity. In keeping with previous studies on syntactic
complexity in the conversational speech of children who stutter (Gaines et al.,
1991; Logan & Conture, 1995; Weiss & Zebrowski, 1992), syntactic complex-
ity was measured using the DSS (Lee, 1974). The DSS assigns a point-value to
eight different components of children’s utterances (indefinite pronouns, per-
sonal pronouns, main verbs, secondary verbs, negatives, conjunctions, inter-
192 J. S. YARUSS, R. M. NEWMAN, and T. FLORA
Data Analysis
Several different analyses were employed to evaluate the relationship between
utterance length, syntactic complexity, and speech fluency in these 12 nor-
mally fluent children’s conversational speech. First, Spearman rank-order cor-
relations were calculated for both group and individual data to assess relation-
ships between the various measures of length and complexity. This nonparametric
correlation procedure was selected for the group analyses because of the rela-
tively small number of data points in the group average data (n 5 12), and
also for the individual analyses because of the unequal sample sizes between
fluent and disfluent utterances. Second, the length and syntactic complexity of
fluent and disfluent utterances were compared using Mann-Whitney U tests
and Wilcoxon Signed-Ranks tests to determine whether disfluent utterances
were likely to be longer or syntactically more complex. Again, nonparametric
tests were selected because of unequal sample sizes and also because of an ap-
parent lack of heterogeneity of the variance between fluent and disfluent utter-
ances. Finally, 12 separate discriminant function analyses were used to deter-
mine whether the utterance length and syntactic complexity of a given
conversational utterance, when considered together, could be used to predict
the likelihood that an utterance would contain a normal disfluency. These
analyses were conducted for each subject’s 50-utterance speech sample indi-
vidually, and significance levels were Bonferroni-adjusted to maintain an
overall a 5 .05 (individual significance level for each of the 12 comparisons
was set at a 5 .004).
RESULTS
Summary Data
Table 1 summarizes the average utterance length (in words, syllables, mor-
phemes, clausal constituents), average DSS scores, and the total number of
disfluent utterances for each of the 12 subjects. Preliminary comparison re-
vealed so significant differences in the results of these measures between boys
Table 1. Average utterance length (in words, syllables, morphemes, and clausal
constituents), DSS scores, and number of disfluent utterances for each subject
and girls (paired t-tests; p-values ranged from .43 to .66), so results from boys
and girls were combined for the following analyses.
Figure 1. Differences between average utterance length (in words, syllables, mor-
phemes, and clausal constituents) and syntactic complexity (DSS) scores for fluent and
disfluent utterances. All differences were statistically significant (Wilcoxon Signed
Ranks Tests; p , .05).
pears that utterance length and syntactic complexity were related to the occur-
rence of normal disfluencies for only a small subgroup of these normally flu-
ent children.
ings) for the variables in the model (number of clausal constituents, number of
morphemes, and DSS score), (4) the percent of utterances correctly classified
by the discriminant function based on a jackknifed classification matrix, and
(5) a Chi-square test evaluating whether the model could classify utterances as
better than a chance level. For 10 out of 12 subjects (83.3%), the number of
clausal constituents in the utterance received the highest loading in the dis-
criminant functions. Utterance length in morphemes received the highest
loading for the other two subjects (16.7%). Syntactic complexity, represented
by DSS scores, appeared to add little to the discriminant functions for any of
these 12 subjects. Multivariate significance tests suggested that the discrimi-
nant functions could reliably classify the data for six out of 12 subjects (50%);
however, these tests reached the strict Bonferroni-corrected significant level
(a 5 .004) for only one out of 12 subjects (8.3%). Finally, discriminant func-
tions were able to correctly categorize between 52% and 78% of subjects’ ut-
terances as fluent or disfluent in a classification matrix. Several of these clas-
sifications appeared to be above chance levels; however, only one reached
Bonferroni-adjusted significance levels (for Subject 6; x2 5 8.13; p 5 .004).
DISCUSSION
The purpose of this study was to determine whether normally fluent children
produce more speech disfluencies on longer, more syntactically complex ut-
terances in spontaneous conversational speech. This analysis was designed to
test some of the basic tenets of current theories of stuttering that suggest, di-
rectly or indirectly, that certain aspects of language formulation play a pri-
mary role as a source factor for the occurrence of stuttering within a given ut-
terance. Previous research has revealed significant relationships between
utterance length, syntactic complexity, and stuttering in the conversational
speech of children who do stutter, though there have been no similar analyses
of conversational speech in children who do not stutter. Although the extent to
which results obtained with normally fluent children’s more typical speech
disfluencies can be extended to explain the less typical speech disfluencies ex-
hibited by children who stutter is not entirely clear, current language-based
theories do postulate a similar mechanism for explaining disfluencies in the
two populations. Thus, a greater understanding of the relationship between
language factors and normally fluent children’s production of disfluencies
should provide one means for evaluating the basic tenets of current theories.
This study was designed to supplement earlier work on the imitated and mod-
eled utterances of children who do not stutter by examining whether normally
fluent children exhibit more disfluencies on longer or more complex conver-
sational utterances.
The present finding that disfluent conversational utterances had a higher
average length and complexity than fluent conversational utterances is quite
Table 3. Summary of the Discriminant Analyses for classifying whether an utterance was fluent or disfluent based on the utterance length in
199
(.02) (.01)
200 J. S. YARUSS, R. M. NEWMAN, and T. FLORA
speech disfluencies (e.g., Karniol, 1995; Perkins et al., 1991; Postma & Kolk,
1997; Wingate, 1988). In general, these theories focus on some sort of deficit
in the phonological encoding process as the primary factor leading to fluency
breakdown. This factor was not assessed in this study of conversational
speech; however, the theories also suggest, directly or indirectly, that aspects
of linguistic formulation such as syntactic complexity or utterance length
should play a role in determining the likelihood that the phonological encod-
ing problem will ultimately lead to the production of a speech disfluency. For
example, the neuropsycholinguistic theory (Perkins et al., 1991) appears to
suggest that so-called “linguistic disfluency” is more likely when demands on
the language formulation system are increased. Although the specific nature
of these demands is not fully understood, it can be argued that longer or more
syntactically complex utterances result in increased sentence processing de-
mands (e.g., Levelt, 1989; see also Yaruss, 1999), perhaps through increased
working memory requirements (Bosshardt, 1995). The present findings con-
firm that for some children, increased utterance length or syntactic complexity
do appear to be associated with an increased likelihood of disfluency. Further-
more, this relationship does hold for the planning of individual utterances (i.e.,
on an utterance-by-utterance basis). Still, the fact that the prediction levels
were not 100% (one would not expect them to be), combined with the fact that
some subjects did not exhibit such a relationship, suggest that other factors
also play a role in determining the likelihood of disfluencies and that a com-
plete theory should encompass other aspects of speech/language production in
addition to linguistic complexity.
In sum, results from the present study confirm prior reports that disfluent
utterances are, on average, likely to be longer and more complicated than flu-
ent utterances in normally fluent children’s conversational speech. Present
findings extend previous research on normally fluent children’s speech by
highlighting the importance of individual differences between subjects and
provide a means for evaluating the suggestion in current theories that the
mechanism for disfluent speech is similar in children who stutter and children
who do not stutter. Of course, additional research will be necessary to assess
the degree to which such relationships are apparent in the conversational
speech of children who stutter. In addition, present findings highlight the im-
portance of considering differences on an utterance-by-utterance basis, rather
than only comparing averages across subjects’ utterances. Finally, because the
measures of linguistic complexity examined in this study could not adequately
explain the occurrence of disfluencies in children’s conversational speech, it
seems clear that factors other than linguistic complexity, such as various as-
pects of utterance timing (e.g., speaking rate, response latency; e.g., Logan &
Conture, 1995; Yaruss, 1997b) or other aspects of the conversational interac-
tion, such as pragmatics (e.g., Weiss & Zebrowski, 1992) are also likely to
play a role in determining the likelihood that a given utterance will contain a
204 J. S. YARUSS, R. M. NEWMAN, and T. FLORA
speech disfluency. Future research could further our understanding of the pa-
rameters influencing the occurrence of disfluencies by considering various
combinations of these different factors and extending the methodology to ex-
amine relationships in both normally fluent and stuttering children’s elicited
and conversational speech.
The authors are grateful to: Nina Capone and Susan T. Mulhern for their assis-
tance with DSS measures; Nan Bernstein Ratner, Ken Logan, Karla McGre-
gor, and Cynthia Thompson for discussions about the analysis of syntactic
complexity in children’s conversational speech; and Steve Zecker for discus-
sions about discriminant analysis. Portions of this analysis were completed as
part of the third author’s undergraduate honors thesis at Northwestern Univer-
sity (NU) and portions of the manuscript were prepared while the first author
was on the faculty at NU.
REFERENCES
Adams, M. (1990). The demands and capacities model I: Theoretical elabora-
tions. Journal of Fluency Disorders 15, 135–141.
Bernstein Ratner, N., & Sih, C.C. (1987). Effects of gradual increases in sen-
tence length and complexity on children’s dysfluency. Journal of Speech
and Hearing Disorders 52, 278–287.
Bernstein Ratner, N. (1997). Stuttering: A psycholinguistic perspective. In R.
Curlee & G. Siegel (Eds.), Nature and treatment of stuttering: New direc-
tions (2nd Ed., pp. 99–127). Needham Heights, MA: Allyn & Bacon.
Blake, J., Quartaro, G., & Onorati, S. (1993). Evaluating quantitative mea-
sures of grammatical complexity in spontaneous speech samples. Journal of
Child Language 20, 139–152.
Bosshardt, H.-G. (1995). Syntactic complexity, short-term memory, and stut-
tering. Paper presented at the Annual Convention of the American Speech-
Language-Hearing Association, Orlando, FL.
Brundage, S.B., & Bernstein Ratner, N. (1989). Measurement of stuttering
frequency in children’s speech. Journal of Fluency Disorders 14, 351–358.
Crystal, D., Garman, M., & Fletcher, P. (1989). Grammatical Analysis of Lan-
guage Disability. London: Whurr Publisher’s Ltd.
Cohen, J. (1960). A coefficient of agreement for nominal scales. Educational
and Psychological Measurement, 20, 37–46.
Gaines, N.D., Runyan, C.M., & Meyers, S.C. (1991). A comparison of young
stutterers’ fluent versus stuttered utterances on measures of length and com-
plexity. Journal of Speech and Hearing Research 34, 37–42.
LANGUAGE AND DISFLUENCY 205