Sampling Context Affects MLU in the

Language of Adolescents With

Down Syndrome
RESEARCH NOTE

Sally Miles
Robin Chapman Purpose: The authors describe the procedures used to explain an unexpected finding
Heidi Sindberg that adolescents with Down syndrome (DS) had a lower mean length of utterance
Waisman Center, (MLU) than typically developing (TD) children in interviews without picture support, but
University of Wisconsin—Madison not in narratives supported by wordless picture books. They hypothesize that the
picture support of the narrative context increased the MLU for the group with DS alone.
Method: Adolescents with DS (n = 14) and TD children (n = 14) matched for receptive
syntax narrated picture storybooks and participated in interviews. Transcription
reliability, intelligibility/fluency, grammatical errors, discourse and sampling contexts,
and discourse characteristics were examined for their effects on MLU.
Results: The DS group showed a greater responsiveness to adult questions than the TD
group; an alternate MLU without yes/no responses showed the same interaction of
group and context as the original finding. An additional comparison of MLUs, obtained
from narratives present in the interview and narratives elicited using picture books,
showed that picture support in narrative increased MLUs only for the group with DS.
Conclusion: Picture support, rather than narrative context alone, increased MLUs for the
group with DS. Clinical use of narratives and picture support in assessment and
intervention with individuals with DS is discussed.
KEY WORDS: MLU, Down syndrome, language samples,
child language disorders, language assessment

D
own syndrome (DS) has been shown to have a specific behavioral
phenotype in language that includes delays in expressive
syntax, errors of grammatical morpheme omission and use,
and deficits in intelligibility (Chapman & Hesketh, 2000; Miller, Leddy,
& Leavitt, 1999). In individuals with DS, mean length of utterance
(MLU) has typically been used as a measure of expressive language
in conversational (Fowler, 1995) or narrative (Chapman, Seung,
Schwartz, & Kay-Raining Bird, 1998) speech samples, which have
been shown to differ in the opportunities for complex syntax that they
provide (Abbeduto, Benson, Short, & Dolish, 1995; Chapman et al.
1998; Leadholm & Miller, 1992; Thordardottir, Chapman, & Wagner,
2002). MLU differences have been shown to correspond to differences
in complex syntax use (Thordardottir et al., 2002). Variation among nar-
rative contexts has received less study in the DS population, although
MLU has been shown to be influenced by discourse contexts.
This research note first reports our comparison of MLUs computed
from language samples obtained in two contexts: narratives elicited using
wordless picture books and an interview that allowed varied forms of

Journal of Speech, Language, and Hearing Research  Vol. 49  325–337  April 2006  AAmerican Speech-Language-Hearing Association 325
1092-4388/06/4902-0325
extended discourse including narration but used no
picture materials to elicit any content. Two groups were
tested: adolescents with DS and a comparison group of
typically developing (TD) children matched for syntax
comprehension. Findings from previous research led us
to make two predictions: (a) that narrative MLUs, if
higher than those from the interview, would be so for both
groups (Abbeduto et al., 1995) and (b) that MLUs derived
from the same sampling context would be higher for the
syntax-comprehension-matched TD group than those of
the group with DS (Chapman et al., 1998). Instead we
found that the group with DS had an MLU in the nar-
rative context that was significantly higher than in the
interview, but the TD group’s MLUs did not differ from
one another or from the DS group’s narrative MLU. That
the two groups differed was not surprising, but that they
differed only in one context was. The unexpected inter-
action between group and context was difficult to explain
without further analyses.
We next report the series of post hoc exploratory
analyses we used to determine the source of the differ-
ences. We suspected that the picture support in the
narrative condition might be responsible for the high
narrative MLU in the group with DS for two possible
reasons. The pictures may make apparent or highlight
visual detail of characters or events and relationships
between events, or pictures may free up cognitive or
information-processing resources to be allocated to
expression. First, however, we evaluated other variables
to rule out their contribution. These variables were
participant intelligibility and fluency and grammatical
errors and adult and participant discourse character-
istics. Additional analyses evaluated the effects of alter-
nate methods of calculating MLU that would reduce the
influence of pragmatic variables and picture support in
the two language samples.
Factors Affecting MLU
MLU has been shown to be sensitive to the character-
istics of the sampling context, to the discourse character-
istics of the conversational partner, and to the criteria
used to calculate it (Abbeduto et al., 1995; Chapman,
1981; Johnston, 2001; Stalnaker & Creaghead, 1982;
Wagner, Nettelbladt, Sahlen, & Nilholm, 2000). The ef-
fects of variations in the sampling context or procedures
may be different for children with cognitive or language
impairment than for the TD children they are compared
with (see Johnston, Miller, Curtiss, & Tallal, 1993).
Discourse context. Sources of variation in the MLU
measure have been found in aspects of sampling and dis-
course contexts. Spontaneous play tends to elicit shorter
utterances (Leadholm & Miller, 1992). Unlike conversa-
tional samples, narrative samples have shown MLU to be
326 Journal of Speech, Language, and Hearing Research  Vol. 49  325–337  April 2006
a developmental measure for TD children beyond the age
of 5 years and for children with language disorders with
MLUs greater than 5.0 (Miller, Freiberg, Rolland, &
Reeves, 1992; Leadholm & Miller, 1992). Interview sam-
ples have also been used with older children with lan-
guage disorders (Evans & Craig, 1992). Here we use both
narrative and interview samples.
Discourse characteristics. Length of utterance may
reflect the pragmatic requirements of different types of
interactions rather than syntactic development (Brown,
1973). In sampling the language of children with lan-
guage and cognitive disorders, MLU may be affected by
differing levels of skill in the pragmatic domain inde-
pendent of, or in addition to, syntax; for example, higher
rates of response to questions or acknowledgments of
clarifications can cause a reduction in MLU by increasing
the proportion of one-word or constituent phrase utter-
ances (Johnston et al., 1993).
Adult discourse characteristics can affect the amount
and type of the child’s talk (Yoder, Davies, Bishop, &
Munson, 1994) and the child’s overall length of utterance
(Chapman, 1981; Johnston, 2001; Klee, 1992). In lan-
guage sampling with children with language disorders,
question asking, although a naturally occurring part of
conversational exchange, must be taken into account to
determine whether it differs from that of the control
group (e.g., because of intelligibility difficulties) or from
the sampling procedure used to create a language sam-
ple database (Johnston et al., 1993). In this study, we
designed examiner scripts to minimize question asking
in both contexts; our interest in adult discourse charac-
teristics was whether it differed for the two groups.
A greater understanding of discourse characteris-
tics in individuals with DS might assist in the design of
the sampling procedures that could effectively reduce
the influence of pragmatic variables and help determine
whether MLU would be more representative with yes/no
responses omitted. We considered pragmatic differences
from TD participants as potential contributors to the
MLU differences we found (Chapman, 1981; Johnston,
2001; Klee, 1992) and investigated using an alternate
analysis set, excluding yes/no question answers, for the
calculation of MLU.
Transcription reliability. In addition to discourse
characteristics and contexts, the accuracy and reliabil-
ity of transcription have far-reaching consequences
(Johnson, 2000; Klee, 1992) for MLU and other mea-
sures of spontaneous expressive language. Reliability,
in turn, may differ as a consequence of speaker char-
acteristics such as intelligibility. Here we examined the
reliability of our transcriptions in detail.
Intelligibility. Individuals with DS are known to
have intelligibility problems, with some individuals
tending to be unintelligible throughout their lives (see
 Table 1. Characteristics of participants (n = 14, each group).

DS TD

Characteristic M SD M SD

Chronological agea 16.36 2.14 4.06 0.75

Syntax comprehensionb 5.29 1.11 5.23 1.01
Maternal education levelc 15.21 2.26 16.93 1.49
Hearing statusd n=5 n = 13
Gender 6 female, 8 male 8 female, 6 male

Note. DS = Down syndrome; TD = typically developing.

a
Age is in years. bAge-equivalent score from the Test for Auditory Comprehension of Language—Third
Edition; mean of Subtests II, Grammatical Morphemes, and III, Elaborated Sentences. cMother’s years of
education. dPassed hearing screening at 20 dB bilaterally at 500, 1000, 2000, and 4000 Hz. Minimum
criterion for participation: e40 dB for at least one ear.

the review by Stoel-Gammon, 1997). In addition to prob-
lems with accurate articulation of specific sounds and
a greater range and variation of substitution types, chil-
dren with DS display atypical prosodic patterns at the
suprasegmental level in phrasing, speech rate, and place-
ment of sentence stress. For some individuals with Down
syndrome, dysfluency is also a problem (Kumin, 1994)
and could affect intelligibility.
These intelligibility difficulties may affect the calcu-
lation of MLU in a number of ways. First, when a word
is unintelligible, the probability of understanding other
words in the utterance decreases; longer utterances are
thereby differentially excluded from the MLU calculation.
Intelligibility problems affect the utterances excluded
from the analysis set (the complete and intelligible set of
utterances used to calculate MLU), reducing both the
total number of utterances and the number of longer
utterances available for analysis. Second, in discourse
with the examiner, intelligibility problems may result
in frequent adult questions to clarify or confirm content
of the child’s utterances, skewing the number of single-
word utterances in the analysis set. Third, the amount of
shared knowledge between child and examiner will affect
the number of communication breakdowns in the sample.
Of the two contexts examined in this study, the context
in which no shared knowledge was provided by pictures
would be expected to have a larger number of adult clar-
ification requests, resulting in a disproportionately larger
number of shorter child utterances than the context with
picture support. For these reasons, we compared intelli-
gibility rates in the two groups.
Grammatical errors. Adolescents with DS show a
greater number of bound morpheme omissions than TD
children matched for MLU (Chapman et al., 1998). There
may be contexts (e.g., narrative samples, picture-sup-
ported samples) that support improved grammatical
production in groups with language impairment. We
wanted to know if higher rates of omission, or other types
of errors, occurred more frequently in one context and
contributed to MLU differences in this study. Never-
theless, it was difficult to predict which context might be
associated with higher omissions; the expression of more
narrative content supported by pictures might lead to
more omitted grammatical morphemes, but less support
might lead to more omissions in the interview.
For this research note, we investigated the roles of
multiple factors as they affected MLU measures obtained
from interview and narrative contexts in adolescents with
DS compared with TD children matched for syntax com-
prehension. We evaluated the effect of discourse context
(narrative vs. interview), examiner questions, participant
responsiveness to questions, transcription reliability, par-
ticipant intelligibility, and participant grammatical errors
on the calculation of MLU. We evaluated the effect of
picture support by developing an alternate analysis set
asking whether it was narrative context alone or narra-
tive context plus picture support that led to differential
increases in MLU in the DS group.
Method
Participants
Twenty-eight children, adolescents, and young
adults participated in this study: 14 individuals with
DS age 12 years, 10 months, to 21 years and 14 TD
children, age 2 years, 11 months, to 5 years, 8 months (see
Table 1 for participant characteristics). Originally, the
study had 30 participants, but 1 of the TD participants
was unable to participate in one of the language-
sampling procedures used in this study and so was
excluded, as was her DS match.
All participants were part of a seven-session study of
narrative development and word learning in adolescents
and young adults with DS. The TD participants were
recruited from the Madison, WI, area; the group with DS

Miles et al.: Sampling Context Affects MLU 327

was recruited from the state of Wisconsin. Potential
participants were screened initially by telephone to
determine that (a) individuals with DS were Trisomy
21 and had no other developmental disorder or history
of severe behavior, attentional problem, or psychiatric
disorder; (b) TD children had no history of behavioral,
attentional, or psychiatric disorder; and (c) oral language
was the primary means of communication (no partic-
ipants were augmentative and alternative communica-
tion or sign language users); and (d) no participants were
bilingual speakers.
All participants met the following additional cri-
teria for participation in the study, determined on the
1st day of testing: (a) MLU was at least 2.0, calculated
from the narrative language sample, elicited using
wordless picture storybooks. This MLU level allowed
us to capture the beginnings of the developmental
course of narrative skills (Nelson, 1996) as well as the
range of expressive language skills in individuals with
DS, both in the initial narrative samples and in the
longitudinal sessions. (b) Hearing loss, if present, was
not greater than a mild loss, as indicated by a bilateral
hearing screening at 20 dB, performed at 500, 1000,
2000, and 4000 Hz. Threshold testing was conducted
for any frequency a participant did not respond to at
20 dB. Participants proceeded with the language and
cognitive assessment protocol only if the screening re-
vealed a hearing loss of 40 dB pure-tone average (500,
1000, 2000 Hz) or less. This criterion allowed inclu-
sion of a participant sample that is representative of
the DS population (Brooks, Wooley, & Kanjilal, 1972;
Dahle & McCollister, 1986).
TD children were matched for syntax comprehension
on Subtests II and III of the Test for Auditory Compre-
hension of Language—Third Edition (TACL–3; Carrow-
Woolfolk, 1999; see Table 1). The entire TACL–3 was
administered (three subtests: I—Vocabulary, II—Gram-
matical Morphemes, and III—Elaborated Sentences and
Phrases). However, the score used for matching was
based only on the mean of the age-equivalent scores from
Subtests II and III. Subtest I, composed of single-word
items, was not used in the calculation of syntax compre-
hension for matching purposes. This decision was made
because of the concern that syntax rather than recep-
tive vocabulary be matched. The syntax comprehension
match group was developed for the study as a whole, in
which we expected different levels of expressive syntax in
the two groups in the narratives (Chapman et al., 1998)
but similar levels of narrative content (i.e., plot line and
theme; Miles & Chapman, 2002).
Three individuals with Down syndrome had chal-
lenging behaviors that precluded completing individual
tasks or the full protocol. Three of the younger TD
children refused to participate during the narrative
328 Journal of Speech, Language, and Hearing Research  Vol. 49  325–337  April 2006
elicitation. These participants were excluded. All families
were paid for their participation in the study.
Language Samples
Narrative Samples
Materials. The first author elicited two narratives
using the wordless picture stories, One Frog Too Many
(Mayer, 1975) and April Fools (Krahn, 1974). In One Frog
Too Many, a boy’s pet frog is unhappy about the arrival
of a new, smaller pet frog and attempts to get rid of it.
The plot line of April Fools follows two boys’ adventures
as they surprise and scare people with a dragon they
construct. The two stories were modified so they were
equivalent in story structure, total number of pages,
story plot line (Berman, 1988), and number and length
of episodes.
Elicitation procedure. The narratives were counter-
balanced for order of presentation and for previewing
condition. (The previewing conditions, i.e., one story was
previewed before telling and the other was not, were
established for a separate study to evaluate the effect of
previewing on narrative content and linguistic complexity.)
During the narrative elicitation, the examiner pre-
sented each story picture, stated the page number, and
waited for the participant’s spontaneous utterance about
the picture. Nonspecific prompts were provided to elicit
story content only when the participant produced no
spontaneous utterances about the picture (e.g., ‘‘What’s
happening?’’ If this prompt was not effective, ‘‘Tell me
about the picture ’’ was used). The examiner corrected
mislabeling or incorrect interpretation of pictures. She
repeated participants’ utterances for each page; the
repetitions were produced to aid transcription but were
incorporated into the adult–child interaction as inter-
ested encouragements (Peterson & McCabe, 1994) or as
statements with rising intonation that sought confir-
mation of the child’s story content. No time limit was
imposed on narration; total time elapsed in producing
narratives ranged between 3 and 13 min.
Interview Language Sample
An interview format was used to elicit spontaneous
language samples (adapted from Evans & Craig, 1992) by
introducing topics of personal experience. This format
was preferable to sampling in free-play interactions
because it was more appropriate for adolescents and
because free-play samples tend to elicit shorter MLUs
(Leadholm & Miller, 1992). Depending on an individual
participant’s skill level, interview topics permit narration
or extended discourse of varying elaboration.
Elicitation procedure. The interview was timed to
last approximately 10 min, allotting about 2 to 3 min per
topic. No props or pictures were used to elicit this sample.
Participants were informed that they would be talking to
the examiner, the first author, who wanted to get to know
more about them. The examiner introduced a sequence of
four topics: family, school (preschool or other regularly
attended activities were substituted as necessary), activi-
ties done outside of school (or its substitute), and favorite
movies, videos, or TV shows. Topics were introduced in
the same order for each participant and were initiated
with the phrases ‘‘ Let’s talk aboutI ’’ or ‘‘ Tell me
aboutI ’’ Examiner prompts were formulated similarly
(e.g., ‘‘That’s great; tell me more about that’’; ‘‘Tell me
about a movie you saw’’). Topic-extending prompts were
provided to extend the participant’s own content within
the adult-initiated topic (‘‘Tell me more about ___’’). More
specific prompts (e.g., ‘‘Tell me about who lives with you’’
or ‘‘Tell me about your mom’’) within the more general
topic (e.g., family) were used when a participant showed
difficulty in getting started on the topic. The examiner
repeated participants’ utterances to aid transcription
only when they were particularly difficult to understand.
Although this made the two contexts slightly different in
this respect, this decision was made because it was not
possible to incorporate repetition of all utterances into
the interaction without being disruptive. A pause time
was imposed on the first utterance of two topics to inves-
tigate, in a separate study, its effect on productive lan-
guage (Ellis Weismer & Schraeder, 1992; Evans, Viele, &
Kass, 1997). The pause procedure was presented to the
participant as a brief time to think about what she or he
wanted to say. Pause times were counterbalanced across
the topics.
Audiotapes of the narratives and interviews were
made on a Marantz portable cassette recorder PMD222,
using a Crown PZM-185 microphone located on the
table between the participant and the experimenter.
Videotapes were made with a Panasonic WV-3260 video-
camera, located in a corner of the room, with a Crown
PZM-185 microphone located on the table between par-
ticipant and experimenter. Videotapes were recorded
and replayed on a Panasonic videocassette recorder
AG-1970.
Procedure
The language samples were obtained on the initial
day of testing, the first of seven sessions, which lasted
approximately 3 hr. Each participant was given a break
approximately midway through each session; other
breaks were inserted as needed. Most participants
separated from parents; a parent was present in the
testing room, as needed by some of the younger partic-
ipants. The initial session included additional language
and cognitive measures and experimental tasks (non-
verbal cognition, auditory and visual memory, working
memory, phonological memory, visual problem solving,
fast mapping), with counterbalancing, selected for other
aspects of the larger study. Because the protocol for
the initial session was both long and challenging, it was
broken into two sessions for the youngest TD partici-
pants (n = 2), taking into consideration developmental
limitations in attention span and stamina. The possi-
ble confound favoring these individuals’ performance
was preferable to the loss of data or the participants’
discomfort.
Transcription
The narratives and interviews were transcribed
and entered into the Systematic Analysis of Lan-
guage Transcripts; SALT; Miller & Chapman, 2003).
Transcribers, who were undergraduate and graduate
students in communicative disorders, typed the tran-
scripts from audiotapes, occasionally from video when
mechanical failure occurred in audiotaping. Tapes were
replayed on Sanyo Memo-scriber TRC-8080 transcrip-
tion machines with foot pedal and speed control, us-
ing Radio Shack headphones. Transcribers had access
to the SALT manual (Miller, 1997), a list of SALT
spelling conventions, the decision rules for intelligi-
bility and segmentation, and the picture storybooks
used in the narrative language samples. They listened
to an utterance up to three times, using speed control
as necessary.
SALT conventions for segmentation are based on
capturing thought completion. Two factors are used to
determine the completion of a thought, intonational
contour, and the presence of a pause. However, children
may produce more than two independent clauses joined
by conjunctions without pausing or changing intonation
contour. When this occurs, utterances are segmented af-
ter the second conjoined clause so that run-on clauses
do not unduly inflate the MLU measure. For example, an
utterance without pauses, such as ‘‘And then the frog
jumped and he fell on the boat/and the boat is sinking’’
would be segmented at the slash mark preceding the
third independent clause. (All SALT transcription con-
ventions were based on Brown, 1973, with the exception
of the segmentation convention; also see Leadholm &
Miller, 1992.)
Reliability for segmentation and morpheme tran-
scription. All of the transcripts were submitted to
reliability checks of segmentation and morpheme tran-
scription (100% of the data) to assure quality control
across numerous transcribers and throughout the dura-
tion of data collection (3 years). For the narratives, mean
percentage of agreement in morpheme transcription was
98% for the group with DS and 99% for the TD group;
agreement on segmentation and maze identification was
98% and 99% respectively. For the interviews, mean
Miles et al.: Sampling Context Affects MLU 329
percentage of agreement in morpheme transcription and
segmentation was 99% for both groups.
Scoring for the Initial Analysis
MLU was computed from complete and intelligible
(C & I) utterances in the SALT program. To compare
MLUs for the two groups in the two contexts, we
computed narrative and interview MLUs, as follows.
The analysis set for MLU–Narrative–All consisted of
the full set of participant utterances in the two
narrative transcripts, which included yes/no responses
to adult questions and to adult utterances with rising
intonation and off-task conversation about topics other
than the narrative or storytelling procedure. MLU–
Interview–All was calculated in the SALT program
from all participant utterances in the interview tran-
script, which included yes/no responses to adult ques-
tions and to adult utterances with rising intonation
and participant-initiated topics that differed from the
four initiated by the examiner.
Results
MLUs in the Two Groups and Contexts
This analysis was conducted to determine the
effects of group and context on MLU. MLUs were
analyzed in a mixed-design two-way 2 (Group)  2
(Context) analysis of variance (ANOVA), with repeated
measures on the second factor, narrative versus inter-
view context. A Type I error rate of .05 was used for
statistical tests.
The ANOVA calculated for MLU–All (all C & I
utterances; see Table 2 for means and standard devia-
tions) revealed a main effect of context, F(1, 26) = 10.11,
p G .01, hp2 = .28, and a significant Group  Context
interaction, F(1, 26) = 5.14, p = .03, hp2 = .17. There was
no significant effect of group. For the group with DS,
MLU–Narrative was significantly higher than the
MLU–Interview and did not differ significantly from
Table 2. MLU and number of utterances, both contexts: All C & I
utterances.
DS
Context Measure M SD M SD
Narrative MLU 5.69 1.89 6.19
Number of utterancesa 68.2 33.0 48.1
Interview MLU 4.38 1.56 5.98
Number of utterances 128.4 34.4 88.5
Note. MLU = mean length of utterance; C & I = complete and intelligible.
a we computed percentage omitted bound morphemes,
Mean number of utterances per narrative.
330 Journal of Speech, Language, and Hearing Research  Vol. 49  325–337  April 2006
the TD group’s MLU–Narrative. The TD children’s
MLUs did not differ in the two contexts.
Discussion
The finding of an interaction between group and
context led us to formulate hypotheses about the possible
source of this difference and to perform a series of
exploratory analyses to identify it. We speculated that
the picture support provided in the narrative task’s
wordless storybooks might have aided the group with
DS in producing longer utterances expressing complex
relations shown in the pictures (Abbeduto et al., 1995).
However, the source of the MLU differences might be
found in any one of a number of other factors, described in
the introduction, or might be the result of a combination
of factors. In the first phase of our follow-up analyses,
(a) we ruled out interactions between group and context
in transcription reliability and participant intelligibility/
fluency; (b) taking into consideration the characteristic
expressive grammatical deficits in conversational lan-
guage samples of individuals with DS, we asked whether
the narrative context contributed to a higher MLU by
reducing errors for that group; and (c) we examined adult
and participant discourse characteristics to determine
whether they differed by group or context, thus differ-
entially affecting the length of participant utterances.
Follow-Up Analysis 1: Measures of
Factors Affecting MLU
In addition to transcription reliability, the following
variables were computed, using the SALT program, for
participants’ C & I utterances, unless noted. These vari-
ables include measures of the participants’ intelligibility/
fluency and grammatical errors and adult discourse char-
acteristics (the extent to which the experimenter asked
questions) and participant discourse characteristics (re-
sponsiveness to questions), to be considered as contribu-
tors to MLU differences.
Intelligibility and fluency. To evaluate possible effects
of intelligibility and fluency on the calculation of MLU, we
computed percentage intelligibility and percentage mazed
words. Percentage intelligible utterances was the percent-
TD age of the participant’s total utterances that were complete
and intelligible. Percentage mazed words was the percent-
age of the participant’s total words that were in mazes.
1.79 This proportional measure of fluency was chosen over
22.7 others available in the SALT program because it controls
1.58 for the number of opportunities to maze at the word level.
18.1 Grammatical errors. To evaluate the effect of gram-
matical errors on MLU in the two contexts and groups,
percentage omitted words, percentage errors at the word
level, and percentage errors at the utterance level.
Percentage omitted bound morphemes was of interest
because in a previous study, adolescents with DS showed
a greater number of bound morpheme omissions than
TD children matched for MLU (Chapman et al., 1998).
We wanted to know if higher rates of omission in one
context contributed to MLU differences in this study.
This variable was calculated by dividing the number of
omitted bound morphemes by the total of omitted and
produced bound morphemes. Percentage omitted words
was calculated by dividing the total number of omitted
words by the sum of omitted words and words produced.
Total omitted words was the total word tokens omitted
in the main body of the utterance; words with omitted
bound morphemes and those within mazes were not
counted as omitted words. Percentage of errors at the
word level was calculated by dividing the number of
words coded as errors by the total number of words.
Errors included overgeneralization errors (e.g., goed/
went), pronoun errors (e.g., him/he), and semantic errors
(He does/causes lots of trouble). Percentage of errors at
the utterance level was calculated by dividing the
number of errors by the total number of utterances. This
type of error occurred when there were errors in word
order, when several words were omitted, or when the
utterance did not make sense and it was not possible to
determine which words were in error.
Experimenter discourse characteristics. To evaluate
this variable, we computed the proportion of experimenter
questions and the proportion of experimenter requests for
information as a function of the number of participant
utterances. Experimenter questions were defined by
SALT as utterances ending with a question mark. These
included all experimenter utterances with rising intona-
tion: requests for clarification, confirmation and repeti-
tion, tag questions and other forms of yes/no questions,
and wh-questions. Proportion of experimenter questions
was calculated by dividing the total number of experi-
menter questions by the total number of participant
utterances. Experimenter requests for information were
coded as a subset of experimenter questions; these were
defined as wh-questions asking for new information, not
to confirm or clarify information already stated. Experi-
menter scripts were designed to eliminate or reduce this
type of question in the sampling procedure; we expected
to find few of these. Proportion of experimenter requests
for information was calculated by dividing the total
number of requests by the total number of participant
utterances.
Participant discourse characteristics. For this vari-
able, we computed the percentage of response to exam-
iner questions. This measure was calculated by
comparing the number of responses with the number
of opportunities to respond. It is a measurement of the
participant’s responsiveness to questions. The SALT
program computes the number of answers to wh- and
yes/no questions, including their variants, such as yeah,
um-hum, and uh-uh.
Results
The dependent variables that might affect the
calculation of MLU were each analyzed in a mixed-design
2 (group)  2 (context) ANOVA, with repeated measures
on the second factor of narrative versus interview con-
text. A Type I error rate of .05 was used for all statistical
tests. All calculations were on C & I utterances except
for the calculation of percentage intelligibility, propor-
tions of examiner questions, and percentage participant
response to questions. Arcsine transformations were per-
formed on proportional data.
Transcription reliability. We analyzed transcription
reliabilities to be certain that reliabilities were not poorer
for the narrative transcripts in the group with DS. The
ANOVAs for morpheme transcription reliability and
segmentation reliability revealed no significant main
effects or interaction.
Intelligibility. The ANOVA calculated for percentage
intelligibility revealed no significant main effects or inter-
actions. Intelligibility was high for both groups (96%–98%),
and the groups were not significantly different. This was
surprising given the phonological characteristics of the
population with DS in general (Stoel-Gammon, 1997) and
the characteristics of this particular cohort. Two factors
may have contributed to the high intelligibility found
in this study. First, the content of the two sampling con-
texts was constrained, the narrative by the wordless
picture storybooks and the interview by a set number of
topics presented in the same order for all participants.
Transcribers were thus able to anticipate, to some ex-
tent, the content of the participants’ utterances in both
contexts. Second, the transcription practice established
in our lab allowed transcribers to listen to an utterance
three times, and the examiner aided transcription by
glossing all narrative utterances and by glossing difficult
utterances in the interview.
Fluency. The ANOVA calculated for percentage of
maze words revealed a significant main effect of context,
F(1, 26) = 7.85, p G .01, hp2 = .23, but no significant
interaction between context and group. There was not a
significant main effect of group. Both groups produced a
slightly higher percentage of mazed words in the inter-
view context (TD: 11% interview, 9% narrative; DS: 9%,
6%, respectively). The picture support of the narrative
context may have given both groups a slight edge in
fluency by reducing demands on the formulation of
utterance content.
In summary, our transcription procedures allowed
us to capture a considerable proportion of both groups’
language for further analysis and resulted in similarly
Miles et al.: Sampling Context Affects MLU 331
high intelligibility rates, and low mazing rates, for both
groups and contexts. Differences in MLU cannot be
attributed to differing rates of intelligibility or fluency
or to differing transcription practices or reliability.
Grammatical errors. The ANOVA calculated for the
arcsine transformation of the percentage omitted bound
morphemes revealed no interactions or main effect of
context, but the main effect of group was significant,
F(1, 26) = 5.47, p = .03, hp2 = .17. The group with DS
omitted a greater percentage of bound morphemes than
the TD group in both contexts. While the TD group omit-
ted 6% in each context, the group with DS omitted
slightly more in the narrative context (18%) than in the
interview context (12%) although this difference was not
statistically significant. Despite the greater bound mor-
pheme omission in the narratives, the group with DS had
a higher MLU in the narratives than in the interview.
The current cohort with DS shows a deficit in the pro-
duction of bound grammatical morphemes consistent
with that reported in Chapman et al. (1998); the error
rates, however, do not explain the MLU difference we
found here. To achieve an MLU not significantly differ-
ent from the TD children who make fewer errors, the
group with DS may be adding greater length by pro-
ducing more content words.
We also computed three other error types: percent-
age omitted words (where omissions were not optional, as
in ellipsis), percentage errors at the word level (incorrect
use of a word), and errors at the utterance level (typically,
this type of error meant that the utterance was incompre-
hensible). In absolute numbers, the group with DS had
more errors (2%–3% for the group with DS; 1%–2% for
the TD group), but for both groups, these error types were
so infrequent overall that subsequent analyses were not
undertaken. The small percentage of these errors in both
contexts and both groups rules them out as sources of
difference in the DS group’s MLUs in the two contexts.
Adult discourse characteristics. The mean proportion
of experimenter questions ranged between .26 and .28
for the group with DS and between .28 and .35 for the
TD group. The ANOVA calculated for the arcsine trans-
formation of this variable revealed no significant main
effects or interaction.
The proportion of experimenter questions that were
true requests for information was minimal for both groups
and contexts. The ANOVA calculated for this variable
revealed no significant main effects or interaction. We
expected this type of question to be rare, and it was (.004
to .01 for the group with DS; .01–.02 for the TD group).
This type of question was a minute proportion of the
number of questions overall, indicating that the type of
examiner question that predominated was yes/no ques-
tions. Requests for information, even if frequent, would
be likely to elicit constituent phrases and, being longer
332 Journal of Speech, Language, and Hearing Research  Vol. 49  325–337  April 2006
than single-word yes/no responses, would not have as
great an impact on MLU. In these samples, the most
common type of response to adult questions was single-
word yes/no responses; however if one group answered
more questions than the other, or in one context over
another, MLU would be disproportionately affected. Our
last variable was examined for this possibility.
Participant discourse characteristics. The ANOVA
calculated for percentage response to examiner questions
revealed a significant main effect of group, F(1, 26) = 1.26,
p G .01, hp2 = .28, and a significant interaction between
group and context, F(1, 26) = 5.47, p = .03, hp2 = .17. There
was no main effect of context. Percentage response to
experimenter questions reveals the number of times the
participant responded to questions relative to the num-
ber of opportunities to do so. The group with DS
responded more frequently than the TD group in both
contexts (DS group: 82% narrative, 77% interview; TD
group: 53%, 63%, respectively), even though no signifi-
cant differences were found in the amount of experi-
menter questioning for the two groups and contexts (see
above). The group with DS responded to examiner
questions more frequently in the narrative context than
in the interview context, whereas the TD group
responded to questions less frequently in the narrative
context than the interview context. This result, the
interaction between group and context, would lead to a
prediction that the DS group’s MLU would be lower than
that of the TD group in the narrative, the opposite of the
results of our first analysis.
This finding indicated a higher responsiveness to
adult questions in the DS group. It did not, however,
explain the DS group’s narrative MLU, although it did
indicate that the group would have a relatively higher
proportion of single-word acknowledgments and back-
channel responses in the transcripts, reducing their
MLUs disproportionately more than those of the compar-
ison group.
Summary. When we examined variables with a
possible effect on MLU, none of the analyses appeared
to explain the different MLUs for the group with DS.
However there were two interesting findings. (a) The
group with DS had a higher MLU only in the narrative
sample despite a higher frequency of omitted bound
morphemes than the TD group in both contexts. (b) The
group with DS showed a higher percentage of response to
questions, suggesting a greater degree of skill in this
aspect of discourse than the control group and the
likelihood of more frequent single-word utterances affect-
ing MLU. The second finding was a reason to consider an
alternate calculation of MLU. The use of an alternate
calculation of MLU has been suggested when there
is a high rate of adult questioning (Chapman, 1981;
Johnston, 2001). In this study, we found that our rates
of questioning did not differ between the groups, but
differences in participant discourse characteristics pro-
duced a similar result: a higher proportion of single-word
utterances in one group. On the basis of this finding, we
recalculated MLU omitting yes/no responses.
Follow-Up Analysis 2: Alternate Means
of MLU Calculation
For this analysis, in addition to eliminating the effect
of one group’s greater rate of answering questions, we
wanted to address our initial hypothesis, that the picture
support of the narrative condition was a factor in the
higher MLU for the DS group. We created new analysis
sets for alternate MLU calculations, by excluding yes/no
responses in both contexts and by omitting any utter-
ances from the narratives that were not supported by the
picture storybook. This created a narrative analysis set of
strictly picture-supported utterances to compare with the
interview. Thus, we were able to compare a narrative-
relevant MLU scaffolded with pictures to an interview
MLU without such support.
Creation of New Analysis Sets From
Subsets of the Interview and Narrative
MLU–Narrative–Subset (narrative-relevant utter-
ances only; yes/no responses excluded). The analysis set
for this calculation was a subset of all participant utter-
ances, consisting of story-relevant utterances, that is,
participant utterances referring to the story as shown in
the picture book, excluding off-task remarks, remarks
about the act of story telling, or conversation about topics
other than the narrative. (Nonrelevant utterances con-
sisted of 8% of the transcripts for the group with DS and
13% for the TD group. The difference was not statistically
significant.) Utterances in the narrative transcripts were
coded for story relevance; a reliability check for this
scoring was performed on 10% of the transcripts and was
95%. Disagreements were discussed, and changes were
made by the original coder when necessary. This MLU
calculation excluded yes/no responses to adult questions
and to adult utterances with rising intonation and yes/no
utterances acknowledging examiner statements.
MLU–Interview–Subset (yes/no responses excluded).
The analysis set for this calculation was a subset of all
participant utterances, excluding yes/no responses to
adult questions or rising intonation and yes/no utter-
ances acknowledging examiner statements. Unlike the
narrative context, all utterances were considered rele-
vant in the interview because the main focus there was
the interests and activities of the participants, so no
utterances were excluded as off-topic. Examination of a
random sample of 10% of the transcripts showed that
only 6% of C & I interview utterances were off the topics
introduced by the examiner.
Results and Discussion
For the group with DS, the mean MLU–Interview–
Subset was 5.35 (SD = 1.51), and the mean MLU–
Narrative–Subset was 7.01 (SD = 2.26). The TD children’s
mean MLU–Interview–Subset was 6.64 (SD = 1.64) and
their MLU–Narrative–Subset was 6.73 (SD = 1.90). The
ANOVA calculated for MLU–Subsets revealed a main
effect of context, F(1, 26) = 12.37, p = .00, hp2 = .32, and a
significant interaction between group and context, F(1,
26) = 10.04, p = .00, hp2 = .28. There was no significant
effect of group.
For the group with DS, MLU–Interview–Subset
was significantly lower than MLU–Narrative–Subset.
The difference between MLU–Narrative and MLU–
Interview was greater when the analysis set was the
subset of all C & I utterances omitting yes/no utterances
in both contexts and using relevant-only utterances in
the narrative calculation. The TD children’s MLUs did
not differ in the two contexts, using either method of
calculation. Thus, for the group with DS, the higher
narrative MLU continued to appear using this method of
calculation, and the gap between receptive and expres-
sive language again showed only in the interview lan-
guage sample. The gap between receptive and expressive
language was greater when we used the alternate means
of calculating MLU; the difference between the two DS
MLUs was also greater.
These results however led to another and final
question. Was the increased MLU related to narrative
content (e.g., complex relations among events) alone or to
the picture support of complex relations shown in the
storybook? To answer this, we had to compare narrative
sample MLUs to MLUs derived from interview utter-
ances identified as narratives.
Follow-Up Analysis 3: Comparison
of Narratives With and Without
Picture Support
To evaluate the effect of picture support in narrative
alone, we calculated an MLU–Narrative within the in-
terview (receiving no picture support), to contrast with
the MLU–Narrative–Subset calculated from picture-
supported fictional narratives.
Narrative utterances in the interview. For both
groups, we found four types of discourse in the interview
transcripts for both groups: lists of family members and
school classes/activities, descriptions of people and
events, explanations of various situations, and narra-
tives. For each participant’s interview, we identified the
utterances expressing narratives within the interview.
We established minimal criteria for both fictional and
personal experience narratives that were developmen-
tally appropriate for the lowest age range of TD
Miles et al.: Sampling Context Affects MLU 333
participants and for the least experienced narrators in
the group with DS. We defined narratives of past event or
personal experience as a reference to a past experience, or
an anticipated future event, containing a minimum of two
nonroutine actions or events (modified after Peterson &
McCabe, 1994, 1983, and Labov & Waletsky, 1967/1997).
This definition allowed narratives of any length, includ-
ing single-utterance narratives (e.g., ‘‘I threw up at
school so my dad got me home’’); however, there was
only one instance of a single-utterance narrative. Fic-
tional narratives were scored as any attempt to tell
content recalled from TV shows and movies. Utterances
were scored as fictional narrative if they included story
action or event or a listing or description of characters.
Reliability for identifying narratives within the inter-
view. Reliability was calculated on 20% of the interview
transcripts because the proportion of narrative utter-
ances was relatively small. The mean MLU for the
narratives in the six interview transcripts used in the
reliability check was 7.84 and differed only .04 of a
morpheme from the checker’s MLU. Agreement on the
number of narrative utterances was 81%; agreement on
the number of utterances identified as nonnarrative was
95%. These percentages are not the same because agree-
ment was on the total number of utterances, not point-to-
point agreement on individual utterances.
Results
In the group with DS, 93% of the participants (13/14)
produced at least one narrative during the interview. The
range for the group was 1–6 narratives. All TD partic-
ipants produced at least one narrative, the range being
1–8 narratives. For the group with DS, 19% of the inter-
view utterances were in narratives (M = 26.3 utterances);
for the TD group, 18% (M = 17.3 utterances). A separate
Figure 1. Mean lengths of utterance in narrative and narratives in
interview. DS = Down syndrome; TD = typically developing.
334 Journal of Speech, Language, and Hearing Research  Vol. 49  325–337  April 2006
MLU was calculated for these (interview) narrative
utterances; 5.90 for the group with DS, 7.64 for the TD
group. Figure 1 compares MLUs for picture book narra-
tives and narratives in interview. A paired-samples t test
was used to compare the MLUs for each group. For the
DS group only, the difference between the MLU–Narra-
tive and the MLU–Narrative in Interview was statisti-
cally significant. Narrative increased MLU in both
contexts, but MLU was the highest in the language
samples that were narratives supported by the picture
storybook. For the TD group, differences between the two
narrative MLUs were not statistically significant.
Summary and Conclusions
The group with DS had a higher MLU in the
narrative context than the interview, and the TD group
showed no difference. Our analyses ruled out tran-
scription reliability, participant intelligibility, and
experimenter discourse characteristics as the source of
this interaction. The DS group did differ from the TD
group in higher grammatical morpheme error rates, but
this did not explain the MLU interaction. The DS group
showed a greater responsiveness to adult yes/no ques-
tions, elevating disproportionately the number of single-
word utterances in their samples and justifying the use of
an alternate calculation of MLU. The greater response to
questions may show greater sensitivity to the needs of the
listener than the TD group, or it may reflect more
experience in repairing communicative breakdown due
to intelligibility difficulties. Experimenter questions
were predominately requests for clarification and con-
firmation of the participant’s production.
The higher narrative MLU in the group with DS was
indeed partially attributable to narrative, rather than
interview, context, but as we had hypothesized, an even
greater increase in MLU was seen with picture support of
the narrative. We believe that the pictures in the
wordless storybooks were a form of scaffolding for the
group with DS that allowed access to and expression of
knowledge that they have but would not express inde-
pendently. This resulted, for the DS group alone, in a
statistically significant increase in MLU over the sam-
pling context without pictures.1
MLU’s sensitivity to context and to the discourse
characteristics of conversational participants is not new
1
In a prior study, by contrast, adolescents with DS (Chapman et al., 1998)
showed a greater expressive language deficit relative to syntactic
comprehension when MLU was obtained from narrative samples. We believe
the differences in the two studies can be traced to differences in sampling
procedure. In Chapman et al. (1998), the sample was elicited using a frog story
(Mayer, 1969) and mixed sources from other topics, with and without picture
support, to create a standard length in minutes. In the current study,
two wordless storybooks were used; the resulting MLUs were averaged,
regardless of sample length.
in the literature. Because of this sensitivity, the MLU
measure’s reliability and validity have been examined in
a variety of studies. Measures of children’s language and
cognitive abilities are generally recognized to be task
sensitive (e.g., Berman’s, 1995, review). However, the
idea that features of the task may affect a language
measure such as MLU differentially in a clinical pop-
ulation and the comparison group has only recently been
explored (e.g., Johnston, 2001; Miolo, Chapman, &
Sindberg, 2005). Interactionist models of language devel-
opment (e.g., Bock, 1982; Chapman et al., 1992; Elman
et al., 1996; Snow & Pan, 1993) propose that multiple
domains contribute to the language that gets produced at
any one time and that the domains drawn on may vary
depending on the context. These models would predict
that MLU would vary with sources contributing to
language production, including variations in sampling
procedures. Contextual support may affect individuals
with deficits according to their differing strengths in
linguistic, cognitive, and social domains. Contextual
variations can have different effects on different popula-
tions, playing a greater role in one type of group than
another. In the current study, our data show that
sampling context had an influence on the group with
DS that did not affect the TD group.
Implications for Future Research
Decisions about what sampling context to use will
affect both the measure of expressive language and the
creation of a TD language comparison group matched
for expressive language. A sampling context eliciting a
higher MLU, such as the picture-supported narratives
used here, will match the DS group to TD children
with higher language skill and be a more conservative
match. A context eliciting a lower MLU, such as play
or conversation, will result in a match to TD children
functioning at a lower developmental level. Generaliza-
tions can be made to other individuals in the popula-
tion from the data, but with a cautious eye toward the
nature of the sampling context and its effects on out-
comes in the language measure and construction of the
control group.
Investigators must also choose language measures.
MLU does not capture all features of children’s expres-
sive language. In Chapman et al. (1998) and the current
study, the group with DS shows a specific deficit in the
omission of bound morphemes, a rate of omission greater
than the control group matched for MLU and than the
group matched on the TACL–3 here. Language sampling
done in a context eliciting longer utterances, as we see
here for the DS group in the picture-supported context,
cannot necessarily be interpreted as a reduction of
grammatical errors. When MLU does not statistically
differ from that of the control group, it is not safe to
assume that no other characteristics of expressive
language differ.
A limitation of our study is that the narrative
sampling contexts we compared differed in number of
utterances as well as picture support. Future research
could alter the interview procedure, for example, by using
explicit instructions, ‘‘Tell me about a time youI ’’
(Hudson, Gebelt, Haviland, & Bentivegna, 1992), or by
using Peterson and McCabe’s (1983) method of providing
a model of the desired type of experience and asking, ‘‘Did
anything like that happen to you?’’ Multiple narratives
can be elicited using these or similar procedures, permit-
ting equal numbers of narrative utterances to be obtained.
However, the interview narratives and wordless
picture book narratives may not provide equivalent
opportunities for complex syntax. We observed that our
participants reported differing types of personal experi-
ence, varying in the types of relations expressed (e.g.,
some only temporal, others with complex causal relation-
ships, some not explicitly expressed with conjunctions). A
study in which participants have common experiences to
report (experienced in experimental conditions), with
opportunity for the expression of varied relationships
among events and internal states, would be desirable.
Creating two contexts that are equivalent in complexity
is a research challenge yet to be solved.
Clinical Implications
Narrative, rather than conversational, samples for
expressive language assessment (cf. Abbeduto et al., 1995)
for individuals with DS make it possible to capture, and
determine the extent of, an individual’s skill in the use
of longer utterances to express complex relations among
events. Other measures of language characteristics, such
as omitted bound morphemes, will complete the clinical
picture of an individual’s grammatical skill. Computer-
ized language analysis programs are excellent tools for
clinicians in that respect, to save time in counts of specific
grammatical elements and to hone the analysis for any
individual client.
This study highlights the importance of careful
observation of both adult and child behavior in clinical
interactions and their bidirectional effects. Intelligibility,
for example, is a child discourse characteristic that has an
effect on adult behavior. Where there are intelligibility
problems, it is difficult to avoid questioning the child to
confirm or clarify an understanding of the child’s utter-
ances. Questioning in turn may influence the length of
utterances, creating shorter or longer utterances than
the child’s typical production. Depending on the amount
and type of adult questioning, and how responsive the
child is, the clinician may want to consider an alternate
calculation of MLU, omitting question responses. As in
Miles et al.: Sampling Context Affects MLU 335
research, sampling context decisions in clinical practice
will affect the MLU outcome.
In intervention with adolescents with DS, we believe
our results indicate that the use of pictures will support
access to knowledge not expressed without such support
and hence practice of more advanced syntax. Pictures sup-
porting narrative production may reduce the demands on
resources for sentence formulation and memory for story
events and may aid individuals in operating at the outer
bound of their zone of proximal development. Given prac-
tice at this level, the higher level of expression provided
by such support may become established in the repertoire
of individuals with DS (Vygotsky, 1978).
Acknowledgments
This research was supported by National Institutes of
Health Grant R01 HD23353 to Robin Chapman, with additional
funding from the National Down Syndrome Society. We thank
Cynthia Renk for her diligent work in transcription and
reliability.
References
Abbeduto, L., Benson, G., Short, K., & Dolish, J. (1995).
Effect of sampling context on the expressive language of
children and adolescents with mental retardation. Mental
Retardation, 33, 279–288.
Berman, R. A. (1988). On the ability to relate events in
narrative. Discourse Process, 11, 469–497.
Berman, R. A. (1995). Narrative competence and storytell-
ing performance: How children tell stories in different
contexts. Journal of Narrative and Life History, 5, 283–313.
Bock, J. K. (1982). Toward a cognitive psychology of syntax:
Information processing contributions to sentence formula-
tion. Psychological Review, 89, 1–47.
Brooks, D. N., Wooley, H., & Kanjilal, G. C. (1972).
Hearing loss and middle ear disorders in patients with
Down’s syndrome. Journal of Mental Deficiency Research,
16, 21–28.
Brown, R. (1973). A first language: The early stages.
Cambridge, MA: Harvard University Press.
Carrow-Woolfolk, E. (1999). Test for Auditory Comprehen-
sion of Language, Third Edition (TACL–3). Austin, TX:
Pro-Ed.
Chapman, R. S. (1981). Computing mean length of utter-
ance in morphemes. In J. Miller (Ed.), Assessing language
production in children (pp. 22–25). Baltimore: University
Park Press.
Chapman, R. S., & Hesketh, L. J. (2000). Behavioral
phenotype of individuals with Down syndrome. Mental
Retardation and Developmental Disability Research
Reviews, 6, 84–95.
Chapman, R., Seung, H.-K., Schwartz, S., &
Kay-Raining Bird, E. (1998). Language skills of children
and adolescents with Down syndrome: II. Production
336 Journal of Speech, Language, and Hearing Research  Vol. 49  325–337  April 2006
deficits. Journal of Speech, Language, and Hearing
Research, 41, 861–885.
Chapman, R. S., Streim, N. W., Crais, E. R., Salmon, D.,
Strand, E. A., & Negri, N. A. (1992). Child talk: Assump-
tions of a developmental model for early language learning.
In R. S. Chapman (Ed.), Processes in language acquisition
and disorders (pp. 3–19). Chicago: Mosby/Yearbook.
Dahle, A. J., & McCollister, F. P. (1986). Hearing and
otologic disorders in children with Down syndrome.
American Journal of Mental Deficiency, 79, 636–642.
Ellis Weismer, S., & Schraeder, T. (1992). Discourse
characteristics and verbal reasoning: Wait time effects on
the performance of children with language learning dis-
abilities. Exceptionality Education Canada, 3, 71–92.
Elman, J. L., Bates, E. A., Johnson, M. H., Karmiloff-
Smith, A., Parisi, D., & Plunkett, K. (1996). Rethinking
innateness: A connectionist perspective on development.
Cambridge, MA: MIT Press.
Evans, J. L., & Craig, H. K. (1992). Language sample
collection and analysis: Interview compared to free play
assessment contexts. Journal of Speech and Hearing
Research, 35, 343–353.
Evans, J., Viele, K., & Kass, R. (1997). Response latency
and verbal complexity: Stochastic models of individual
differences in children with specific language impairments.
Journal of Speech, Language, and Hearing Research, 40,
754–764.
Fowler, A. (1995). Linguistic variability in persons with
Down syndrome: Research and implications. In L. Nadel, &
D. Rosenthal (Eds.), Down syndrome: Living and learning
in the community (pp. 121–131). New York: Wiley-Liss.
Hudson, J., Gebelt, J., Haviland, J., & Bentivegna, C.
(1992). Emotion and narrative structure in young child-
ren’s personal accounts. Journal of Narrative and Life
History, 2, 129–150.
Johnson, C. E. (2000). What you see is what you get: The
importance of transcription for interpreting children’s
morphosyntactic development. In L. Menn & N. Bernstein
Ratner (Eds.), Methods for studying language production
(pp. 181–204). Mahwah, NJ: Erlbaum.
Johnston, J. R. (2001). An alternate MLU calculation:
Magnitude and variability of effects. Journal of Speech,
Language, and Hearing Research, 44, 156–164.
Johnston, J. R., Miller, J. F., Curtiss, S., & Tallal, P.
(1993). Conversations with children who are language
impaired: Asking questions. Journal of Speech and Hearing
Research, 36, 973–978.
Klee, T. (1992). Measuring children’s conversational lan-
guage. In S. F. Warren, & J. E. Reichle (Eds.), Communi-
cation and language intervention: Vol. 1. Causes and
effects in communication and language intervention
(pp. 315–330). Baltimore: Brookes.
Krahn, F. (1974). April fools. New York: Dutton.
Kumin, L. (1994). Intelligibility of speech in children with
Down syndrome in natural settings: Parents’ perspective.
Perception and Motor Skills, 78, 307–316.
Labov, W., & Waletzky, J. (1997). Narrative analysis: Oral
versions of personal experience. Journal of Narrative and
Life History, 7, 3–38. (Reprinted from Essays on the Verbal
 and Visual Arts: Proceedings of the 1966 Annual Spring
Meeting of the American Ethnological Society, pp. 12–44,
by J. Helm, Ed., 1967, Seattle: University of Washington
Press.)
Leadholm, B., & Miller, J. (1992). Language sample
analysis: The Wisconsin guide. Madison: Wisconsin
Department of Public Instruction.
Mayer, M. (1969). Frog, where are you? New York: Dial.
Mayer, M. (1975). One frog too many. New York: Dial.
Miles, S., & Chapman, R. S. (2002). Narrative content as
described by individuals with Down syndrome and typically
developing children. Journal of Speech, Language, and
Hearing Research, 45, 175–189.
Miller, J. F. (1997). SALT database programs. Madison:
Language Analysis Laboratory, Waisman Center, Univer-
sity of Wisconsin.
Miller, J. F., & Chapman, R. S. (2003). Systematic Analysis
of Language Transcripts [Computer software]. Madison:
University of Wisconsin—Madison, Waisman Center,
Language Analysis Laboratory.
Miller, J. F., Freiberg, C., Rolland, M., & Reeves, M. A.
(1992). Implementing computerized language sample
analysis in the public school. Topics in Language Disorders,
12, 69–82.
Miller, J. F., Leddy, M., & Leavitt, L. A. (1999). Improving
the communication of people with Down syndrome (pp. 241–
262). Baltimore: Brookes.
Miolo, G., Chapman, R. S., & Sindberg, H. (2005).
Sentence comprehension in adolescents with Down syn-
drome and typically developing children: Role of sentence
voice, visual context, and auditory–verbal short-term
memory. Journal of Speech, Language, and Hearing, 48,
172–188.
Nelson, K. (1996). Language in cognitive development: The
emergence of the mediated mind. Cambridge, United
Kingdom: Cambridge University Press.
Peterson, C., & McCabe, A. (1983). Developmental psy-
cholinguistics: Three ways of looking at a child’s narrative.
New York: Plenum.
Peterson, C., & McCabe, A. (1994). A social interactionist
account of developing decontextualized narrative skill.
Developmental Psychology, 30, 937–948.
Snow, C. E., & Pan, B. A. (1993). Ways of analyzing the
spontaneous speech of children with mental retardation:
The value of cross-domain analyses. International Review
of Research in Mental Retardation, 19, 163–192.
Stalnaker, L. D., & Creaghead, N. A. (1982). An exami-
nation of language samples obtained under three exper-
imental conditions. Language, Speech, and Hearing
Services in Schools, 13, 121–128.
Stoel-Gammon, C. (1997). Phonological development in
Down syndrome. Mental Retardation and Development
Disabilities Research Reviews, 3, 300–306.
Thordardottir, E., Chapman, R., & Wagner, L. (2002).
Complex sentence production by adolescents with Down
syndrome. Applied Psycholinguistics, 24, 163–183.
Vygotsky, L. S. (1978). Mind in society: The development of
higher psychological processes (M. Cole, V. John-Steiner,
S. Scribner, & E. Souberman, Eds.). Cambridge, MA:
Harvard University Press.
Wagner, C. R., Nettlebladt, U., Sahlen, B., & Nilholm, C.
(2000). Conversation versus narration in pre-school chil-
dren with language impairment. International Journal of
Language and Communication Disorders, 35, 83–93.
Yoder, P. J., Davies, B., Bishop, K., & Munson, L. (1994).
Effect of adult continuing wh-questions on conversational
participation in children with developmental disabilities.
Journal of Speech and Hearing Research, 37, 193–204.
Received September 30, 2003
Revision received March 19, 2004
Accepted September 6, 2005
DOI: 10.1044/1092-4388(2006/026)
Contact author: Sally Miles, 540 Waisman Center, University
of Wisconsin—Madison, Madison, WI 53705.
E-mail: miles@wisc.edu
Miles et al.: Sampling Context Affects MLU 337