Journal of Speech and Hearing Research, Volume 35, 259-273, April 1992

Articulation Testing Versus

Conversational Speech Sampling

Judith A. Morrison
University of Redlands
n s a _
Kedlands, A
A
I
Lawrence D. Shriberg
University of Wisconsin-Madison

Detailed speech analyses were performed on data from 61 speech-delayed children assessed
by both a standard articulation test and a conversational speech sample. Statistically significant
differences between the articulation accuracy profiles obtained from the two sampling modes
were observed at all linguistic levels examined, including overall accuracy, phonological
processes, individual phonemes, manner features, error-type, word position, and allophones.
Established sounds were often produced more accurately in conversational speech, whereas
emerging sounds were often produced more accurately in response to articulation test stimuli.
Error patterns involving word-to-word transitions were available only inthe context of continuous
speech. A pass-fail analysis indicated that the average subject would receive similar clinical
decisions from articulation testing and conversational speech sampling on an average of 71% of
consonant sounds. Analyses of demographic, language, and speech variables did not yield any
subject characteristics that were significantly associated with concordance rates in the two
sampling modes. Discussion considers sources of variance for differences between sampling
modes, including processes associated with both the speaker and the transcriber. In comparison
to the validity of conversational speech samples for integrated speech, language, and prosodic
analyses, articulation tests appear to yield neither typical nor optimal measures of speech
performance.
KEY WORDS: phonology, assessment, speech sampling, continuous speech, articula-
tion tests

The term communicative competence refers to performance in everyday social

interaction (Hymes, 1971). It is only in such contexts that we can observe the variety
of human discourse and the complex speech-language interactions that occur when
one is talking to be understood. Despite agreement on the face validity of such
contexts for assessing what might be termed typical or customary communicative
performance (e.g., Miller, 1981; Locke, 1983; Wren, 1985), clinicians and researchers
continue to justify the use of alternative assessment procedures on the basis of
efficiency or utility considerations.
Nowhere in the communicative disorders literature are efficiency issues more
evident than in the selection of sampling procedures for phonological analysis.
Particularly in child phonology, single word articulation tests continue to be widely
used, despite over four decades of research documenting differences that occur when
children give single-word responses, termed citing, compared to their speech when
spontaneously talking (e.g., Andrews & Fey, 1986; Dubois & Bernthal, 1978; Dyson
& Robinson, 1987; Faircloth & Faircloth, 1970; Gallagher & Shriner, 1975; Harris &
Cottam, 1985; Healy & Madison, 1987; Johnson, Winney, & Pederson, 1980; Jordan,
1960; Kenney, Prather, Mooney, & Jeruzal, 1984; Klein, 1984; Klein & Spector, 1985;
Orr, Blodgett, & Miller, 1983; Paden & Moss, 1985; Paynter & Sims, 1979; Schmitt,
Howard, & Schmitt, 1983; Siegel, Winitz, & Conkey, 1963; Simmons, Blodgett, &
Miller, 1983; Van Demark, 1964). Unlike case studies or small group studies of normal
speech development that have used continuous speech samples collected in natural

© 1992, American Speech-Language-Hearing Association 259 002-4685/923502-0259$01 .00/0

260 Journal of Speech and Hearing Research 35 259-273 April 1992

or experimental settings (e.g., Leopold, 1947; Stemberger, than imitative responses, with 10 different articulation tests
1988; Stoel-Gammon, 1987; Vihman & Greenlee, 1987), represented within just this sample of the literature. Thus a
large-scale studies of both normal and speech-delayed chil- major constraint on generalizations from the citation-form
dren most often have been based on diverse citation-form data across studies is that they reflect unique (i.e., unrepli-
tasks and protocols, most frequently in the form of pictured- cated) permutations of modes of evocation with specific
word articulation tests. pictured-word stimuli. Even the size and color of materials
A review of over 50 unpublished and published studies used to evoke the same pictured words have been studied as
during the past 40 years indicates a diversity of descriptive potential sources of variance (e.g., Bernthal, Grossman, &
methods to assess the clinical consequences of different Goll, 1989). Of three types of contexts used to sample
forms of citation-form testing or to compare citation-form continuous speech-spontaneous, retelling, and imitation-
testing with continuous-speech sampling. Dependent vari- most studies have used spontaneous conversational speech.
ables in this literature parallel the chronology of theoretical However, within the studies using continuous speech sam-
paradigms in developmental speech disorders. More recent ples, considerable differences in method are apparent, in-
studies have focused on the potential effects of sampling cluding differences in the child's purpose for talking (e.g.,
mode on the frequency of occurrence of errors classified by naming, repeating, informing), the level of propositional effort
phonological processes. Table 1 provides a descriptive sum- (e.g., labelling, telling, retelling, describing), the availability of
mary of 20 representative studies reported during the past situational support for talking (e.g., familiar or observable
two decades. These studies were selected to illustrate the objects, pictures, or events), the level of spontaneity reflected
methodologic diversity from which generalizations about the in the sample (e.g., spontaneous, evoked, imitative), the
influence of type and mode of sampling have been based. length and complexity of the utterance, and differences in the
Sample sizes range from 1 to 240, with children's ages semantic, syntactic, and phonetic content of the comparison
ranging from under 3 years to over 12 years. Most studies articulation test.
using citation forms have been based on spontaneous rather This diversity of methodological approaches notwithstand-

TABLE 1. A representative sample of 20 studies (1970-1990) Illustrating the methodological diversity In research comparing
phonological performance In citation forms and continuous speech.

Mode of evocation and sampling context

Subjects Citation forms Continuous speech
Age Sponta- Sponta-
Study n (Yrs:Mos) Speech Status neous Imitated neous Retelling Imitated
Faircloth & Faircloth (1970) 1 11:0 Speech-delayed xa X
Chapman & Ting (1971) 40 Preschool- Normal xa xa Xa
1st Grade
DuBois & Bernthal (1978) 18 4:3-6:2 Speech-delayed xb XC X x
Paynter & Sims (1979) 4 4:11-5:11 Language-delayed xd xC X
Johnson, Winney, & 35 3:7-9:5 Speech-delayed xe xe
Pederson (1980)
Shriberg & Kwiatkowski 10 3:0-6:0 Speech-delayed x' x' x
(1980)
9 h
Dunn (1982) 1 4:6 Speech-delayed x xh x
Bankson & Bernthal (1982) 18 4:0-4:11 Speech-delayed x x'
Schmidt, Howard, & 240 3:0-7:0 Normal x' x
Schmidt (1983)
Simmons, Blodgett, & 8 3:9-4:6 Speech-delayed x' x
Miller (1983)
Klein (1984) 10 4:1-6:1 Speech-delayed x' x
Klein & Spector (1985) 8 5:2-6:11 Speech-delayed xbg x
Paden & Moss (1985) 3 4:11-7:6 Speech-delayed x x
Andrews & Fey (1986) 14 2:8-6:1 Speech-delayed x9 x
Blodgett & Miller (1986) 15 3:1-4:9 Speech-delayed xa x Xa
Healy & Madison (1987) 20 5:4-12:8 Speech-delayed xk9 XC
Dyson & Robinson (1987) 5 3:5-6:5 Speech-delayed x x
Watson (1989) 8 3:1-7:4 Speech-delayed x' xm x
Elbert, Dinnsen, 10 3:7-5:9 Speech-delayed xb x
Swartzlander, & Chin
(1990)
Smit (1990) 21 3:0-39:9 Normal x"
Note. a Same words taken from spontaneous conversational speech; b Authors' word list; c Examiner elicited the desired response if not
produced spontaneously; d Templin-Darley Test of Articulation (Templin & Darley, 1969); e Goldman-Fristoe Test of Articulation (Goldman &
Fristoe, 1972); ' Photo Articulation Test (Pendergast, Dickey, Selmar, & Soder, 1969); gAssessment of Phonological Processes (Hodson, 1980);
i
h Phonological Process Analysis (Weiner, 1979); i Arizona Articulation Proficiency Scale (Fudala, 1974); Goldman-Fristoe Test of Articulation
(Goldman & Fristoe, 1969); k Weiss Comprehensive Articulation Test (Weiss, 1978); ' Compton-Hutton Phonological Assessment (Compton &
Hutton, 1978); mAssessment Link Between Phonology and Articulation (Lowe, 1986); nTape-recorded utterances with rate variations.
 Morrison & Shriberg: Articulation Testing Versus Conversational Speech Sampling 261

ing, some common trends in the literature on the effects of deletions of consonant singletons, consonant clusters, and
speech sampling mode on speech data are apparent. Table unstressed syllables. More generally, the pattern of errors
2 is a summary of effects that have received some support obtained from a speech sample might be a reflection of the
across several studies. In general, more frequent and varied number, complexity, and saliency of multisyllabic forms inthe
errors occur in talking compared to citing, but group and corpus (e.g., Klein & Spector, 1985; Shriberg & Kwiatkowski,
individual-level findings also include instances in which 1980; Simmons, Blodgett, & Miller, 1983). The acquisition of
speech errors are more frequent in citation modes. The more continuous speech rules at the phonological level involves
recent studies using the phonological process as the unit of complex interactions among semantic, syntactic, morpho-
articulatory analysis have the disadvantage of concealing logic, pragmatic, and prosodic tiers (Levelt, 1989; Selkirk,
specific descriptive information on the interaction of error 1984). Such structural and phonological complexities are
sounds by error types by word positions (cf., Shriberg, 1990). presumed to invoke resource allocation processes, as ob-
Most generally, continuous speech appears to be associated served in both younger normal-speaking children (e.g., Dunn
with more deletion errors, especially of consonants in word- & Davis, 1983) and adults with other communicative disor-
final position, and with increased errors involving clusters ders (e.g., Kohn, 1988).
and unstressed syllables. In recognition of these frequently A second potential source of explanation for group- and
observed yet unpredictable differences, most current assess- individual-level differences in error sound patterns in the two
ment guidelines recommend using narrowly transcribed, sampling modes is the contextual support available to the
whole-word analyses of both citation-form and continuous speaker in talking compared to citing. Although resource
speech sampling. Sampling in multiple contexts presumably allocation demands presumably are greater in talking, con-
ensures the broadest data base from which to make judg- versational discourse may offer a different form of cognitive-
ments about what Stoel-Gammon (1988) has termed chil- linguistic support than available during articulatory testing.
dren's relational and independent phonological competence Snyder's (1984) discussion of children's communicative
(Bernthal & Bankson, 1988). competence reflects such a perspective, including observa-
Explanations for the findings of more articulation errors in tions on how such information might contribute to estimates
either talking or citing (see Table 2) have focused on three of a child's typical communicative ability as well as predict a
potential sources of variance. Most salient are the conse- child's potential for communicative growth:
quences of differences in the speech sample's linguistic
structure and content, with its associated demands on pho- ... if we can develop a model of communicative competence
nological and speech-motor processing. Linguistically, the that can accommodate the diversity of communicative output,
its processing, and the competitive and collaborative use of
canonical complexity of stimulus words in articulation tests resources within the system, we might make more realistic
has been viewed as increasing the probability of phoneme assumptions and predictions about children's communicative

TABLE 2. Some literature findings comparing phonological performance In citation forms (citing) and continuous speech (talking).

Construct Findings References

Severity of involvement More articulatory errors occur intalking than citing, but Andrews & Fey, 1986; DuBois & Bernthal,
severity ratings may be equivalent or poorer when 1978; Faircloth & Faircloth, 1970; Healy &
based on citation forms. Madison, 1987; Johnson, Winney, &
Pederson, 1980; Klein, 1984; Simmons,
Blodgett, & Miller, 1983
Error types The majority of error types using phonological process DuBois & Bernthal, 1978; Kenney, Prather,
typologies have been observed inboth talking and Mooney, & Jeruzal, 1984; Klein, 1984;
citing: Assimilation; Cluster Reduction; Gliding: Paden & Moss, 1985; Simmons et al., 1983
Liquid Deviation/Simplification; Medial Consonant
Deletion; Palatal Deviation/Fronting; Stopping;
Vocalization.
Certain error types are more frequent intalking than Andrews & Fey, 1986; DuBois & Bernthal,
citing: Cluster Reduction; Consonant Deletion, Final 1978; Faircloth & Faircloth, 1970; Healy &
Consonant Deletion; Syllable Deletion, Unstressed Madison, 1987; Johnson et al., 1980; Klein,
Syllable Deletion. Substitution errors inciting may 1984; Orr, Blodgett, & Miller, 1983; Paynter
be realized as deletions in talking. & Simms, 1979
Individual differences Several subjects in several studies have DuBois & Bernthal, 1978; Dyson & Robinson,
demonstrated errors on several sounds or error 1987; Healy & Madison, 1987; Johnson et
classes only in talking contexts: Ambisyllabic al., 1980; Klein, 1984; Orr et al., 1983;
Consonant Deletion; Assimilation; Coalescence; Paynter & Sims, 1979
Final Consonant Deletion; Initial Consonant
Deletion; Neutralization; Stopping; Unstressed/Weak
Syllable Deletion.
Several subjects in several studies have Andrews & Fey, 1986; Dyson & Robinson,
demonstrated errors on several sounds and error 1987; Paden & Moss, 1985
classes only in citing: Glide and /I/Deviation; Labial
Assimilation; Stridency Deletion; Velar Deviation.
262 Journal of Speech and Hearing Research 35 259-273 April 192

competence. If we consider the subject and task characteris- base of diagnostic assessments of young, speech-delayed
tics as well as the resources that can be used to perform children, we examined children's responses to citation-form
various communicative tasks, we may be able to unmask
some of the puzzling inconsistency in the language-delayed articulation testing to assess in some detail whether such
child's communicative performance (p. 104). responses reflect their performance in spontaneous conver-
sational speech.
Snyder's perspective suggests that spontaneous conversa-
tional speech may not represent a more difficult context than
citation forms due to a child's opportunity to partially control Method
topic and content, thus ensuring familiarity of words and
meanings. Findings by Menyuk (1980) and Campbell and Subjects
Shriberg (1982) demonstrate variability in phonological ac-
curacy associated with linguistic stress and pragmatic func- Transcripts from 61 of a cohort of 64 speech-delayed
tion, with improved accuracy when children assert or clarify children in a longitudinal study met criteria for inclusion in the
messages in free conversation. Whereas error frequency study. The 64 children had been referred by speech-lan-
may increase in longer, phonologically more complex words, guage pathologists for intelligibility deficits of unknown origin.
unstressed monosyllabic words in conversation (which often Only those children who produced at least 50 nonimitative
carry less communicative force) appear to be most vulnera- utterances in each of two sampling conditions (to be de-
scribed) were included in the present analyses. As shown in
ble to articulation errors (Klein & Spector, 1981).
A third and seldom discussed source of variance in speech Table 3, the gender ratio favoring boys by approximately 2:1
sampling reflects the effects of speaker and task variables on was lower than the approximately 3:1 ratio reported in
phonetic transcription. Response definitions for acceptable previous studies of such children (Shriberg, Kwiatkowski,
articulation in citation forms typically differ from those re- Best, Hengst, & Terselic-Weber, 1986). Consistent with prior
quired to assess articulation in continuous speech forms. The work, approximately 85% of subjects were 4-6 years of age.
latter require familiarity with many assimilatory processes A metric termed Percentage of Consonants Correct (PCC)
that operate in casual and fast speech. For example, if asked (Shriberg & Kwiatkowski, 1982; Shriberg et al., 1986), calcu-
lated on the conversational speech samples used for this
to read the sentence "He put his hat on," normal adult and
child speakers would articulate each of the word-initial h/ study, classified most children's speech delays as falling in
sounds correctly. In spontaneous conversational speech, the mild-moderate to moderate-severe range. All children
however, whether from an adult or child with normal or had hearing within normal limits, showed no significant
sensory-motor involvement, and were developing normally in
disordered speech, deletion of the initial /h/ in the unstressed
pronoun "his" is common and acceptable. The contributions cognitive and social domains. With the exception of one
child, language comprehension was within normal limits, and
of such linguistic variables in continuous speech to speech
the children's expressive language ranged from within nor-
perception outcomes in normal discourse have received
considerable research attention (Klatt, 1989; Nittrouer & mal limits to moderately delayed. The assessment protocol
Boothroyd, 1990). Alternatively, over-careful articulation of (see below) included both standardized measures of lan-
sounds in citation forms (e.g., adding a slight schwa off-glide guage comprehension and production and free speech sam-
to word final voice stops) would generally be recorded as pling procedures (Miller, 1981).
"correct" unless the examiner had reason to believe that the
speaker was not in control of the appropriate allophone (e.g., Procedures
as might be the case for a speaker in an accent reduction
program). Thus, whether using correct-incorrect scoring, All data collection and analyses procedures used in the
broad phonetic transcription, or narrow phonetic transcrip- present study have been developed and described in prior
tion, measurement decisions must be guided by explicit work, including procedures for sampling speech, accomplish-
response definitions for each sampling mode. ing narrow phonetic transcription by consensus, and coding
This paper proceeds from the positions advanced by and entering transcriptions for computer-aided phonological
Henderson (1938) and Jordan (1960), which claim that the analysis (Shriberg, 1986; Shriberg & Kent, 1982; Shriberg &
appropriate referent for the concurrent validity of articulation Kwiatkowski, 1980, 1985; Shriberg et al., 1986; Shriberg,
tests is the construct of customary speech performance in Kwiatkowski, & Hoffmann, 1984; Shriberg, Hinke, & Trost-
contextually rich continuous speech. Drawing from a data- Steffen, 1987). Audiocassette samples of conversational

TABLE 3. Description of subjects.

Age (years) Percentage of Consonants Correct (PCC)'

Standard
Gender n % 3 4 5 6 Mean Deviation Range
Male 40 66 6 18 14 2 61.5 6.9 47.1-74.6
Female 21 34 3 1 6 1 64.5 10.0 38.9-78.3
Both 61 100 9 29 20 3 62.5 8.2 38.9-78.3
Note. aShriberg & Kwiatkowski (1982).
 Morrison & Shriberg: Articulation Testing Versus ConversationalSpeech Sampling 263

speech and responses to the Photo Articulation Test (Pen- sistent with other reports in disordered child phonology, with
dergast, Dickey, Selmar, & Soder, 1969) were obtained in narrow phonetic transcription reliable for only certain re-
randomized order by trained research examiners as part of a search questions and broad phonetic transcription coeffi-
21/2-hour diagnostic assessment protocol. Specifically, the cients achieving adequate interjudge and intrajudge levels
ordering of the articulation test and the conversational (cf., Shriberg & Lof, 1991). As discussed in Shriberg and Lof,
speech sample depended on the examiner's judgment on transcription agreement on continuous speech samples ap-
how comfortable each child appeared with the more struc- pears to be somewhat higher than agreement based on
tured articulation test compared to the less structured con- articulation test responses (1-13 percentage points) due to
versational sample. All testing was done in a quiet suite using complex associations among word forms and error types in
high-quality audiocassette tape on Marantz PMD221 audio- each mode. The following findings, with the exception of the
cassette recorders, with matching external microphones allophone data, are based on analyses of the broad phonetic
monitored at a lip-to-microphone distance of approximately transcriptions.
15cm. Responses to the articulation test were obtained
spontaneously using the set of picture cards and evocation
procedures recommended for this standardized, single-word, Structural Comparison
citation-form measure. Conversational samples were ob-
tained by inviting the child to talk about home and social Previous analyses indicate that the types and percentages
activities, using a set of cues and prompts described in of word forms and phoneme distributions differ in continuous
Shriberg and Kwiatkowski (1985). speech compared to articulation test protocols (Shriberg,
Narrow phonetic transcriptions of all samples were com- 1986; Shriberg & Kwiatkowski, 1980). Before proceeding to
pleted by two two-person consensus teams that were well the analyses, it is important to compare the distributions of
trained in a set of computer-aided transcription methodolo- intended (i.e., correct adult) canonical forms and the distri-
gies for young, speech-disordered children (Shriberg, 1986; butions of intended consonants and vowels/diphthongs in the
Shriberg & Kent, 1982; Shriberg & Kwiatkowski, 1980). Using present two sampling modes. The four panels in Figure 1 are
well-maintained Dictaphone 2025 play-back devices, one graphic summaries of the relevant data. The top left panel is
team transcribed the continuous speech samples, and the a display of the percentage of intended canonical forms in the
other transcribed each of the words used as articulation test two sampling modes, with intended percentage of occur-
stimuli. That is, for all analyses of the citation forms to follow, rence of 10 forms sorted left to right from most to least
computations are based on all occurrences of vowel/diph- frequent in conversational speech. Wilcoxon Matched Pairs
thong and consonant sounds that occurred in each of the 76 Signed-Ranks tests indicated that with the exception of the
articulation test words. A 90-70-225 sampling rule ensured percentage of intended CVC forms, all comparisons were
comparable continuous speech samples containing either 90 significant at the .0001 level of confidence. The largest
word types, 70 utterances, or a total of 225 words, whichever apparent absolute differences are in the average percent-
criterion was met first during the glossing-transcription pro- ages of intended CV, two-syllable, and VC words in each
cess. mode (there were no intended V or CnV forms in the
articulation test), with the most notable difference in the
percentage of two-syllable words in continuous speech com-
Reliability pared to articulation tests. A likely source of these differences
is the lack of function words in articulation test stimuli. Simple
Extensive interjudge and intrajudge reliability data for each word shapes, such as those for some determiners, are
of the two-person consensus transcription teams used in this underepresented in standardized articulation test stimuli,
study have been reported, including individual sound-level whereas they account for over 22% of words in comparable
data from six separate studies of children and adults with continuous speech samples (Shriberg & Kwiatkowski, 1985).
speech disorders (Shriberg & Lof, 1991). Extremely conser- Moreover, the need to test sounds in the intervocalic position
vative agreement criteria were used in a computer program in articulation tests is associated with the greater number of
that calculated transcription reliability (Shriberg & Olson, two- and three-syllable words in the articulation test, com-
1987), with narrow phonetic transcription agreement on a pared to their average occurrence in the continuous speech
sound requiring exactly the same configuration of any of the samples. The general picture across the 10 categories is that
42 diacritics available to describe a speech error. Using incomparison to word forms in conversational speech, those
randomly selected transcripts, including approximately 100 in this articulation test present considerably more difficult
to over 250 tokens per comparison, interjudge agreement structural contexts for articulatory performance.
between the two teams on articulation test and conversa- The remaining three panels in Figure 1 provide sampling
tional speech samples ranged from 61.3% to 70.1% for mode comparisons for the percentage of occurrence of
narrow phonetic transcription of consonants and vowels/ intended vowels/diphthongs, singleton consonants, and con-
diphthongs and 84.7% to 88.6% for broad transcription of sonant clusters. The continuous speech data points are
consonants and vowels/diphthongs. Intrajudge agreement again sorted left to right from most to least frequent. The
for each team, calculated on approximately 80 to 1,100 trends suggest that relatively large percentage differences
tokens per comparison, was 65.5% to 81.1% for narrow occurred for only a few sounds in each class. For vowels/
phonetic transcription and 86.7% to 95.1% for broad phonetic diphthongs, the greatest differences in the percentage of
transcription. These transcription agreement figures are con- intended sounds in each sampling mode occurred on the
 264 Journal of Speech and Hearing Research 35 259-273 April 1992

oD 10
)O ,! Conversaionl Speech
O Aticulation Test

30' 30-
ill
C
4
z
W 20 20

Cc
10
o-
O
U 20- 10 C
O 0
C
O *0
y CnCn CnC Syll Cn
C CVINTENDED
2 CAN
ONICALFORMSC C S CnV
I o Ao V l U " o
INTENDED CANONICAL FORMS INTENDED VOWELS AND DIPHTHONGS

40 -

1
aJ

20 20o

-0 0

1 o ,o C) I *
0 I .

t b k d h r g p j v f 0 d3 t t t s n r 1 d k p b .
w 0 v t d3 j h 8
INTENDED CONSONANTS AS SINGLETONS INTENDED CONSONANTS AS CLUSTERS

FIGURE 1. Percentage Intended canonical and segmental forms Inconversational speech and
articulation testing sorted left to right from most to least frequent inconversational speech.

three diphthongs a-i/, /o/, and /-T and the two vowels /a/ parisons were excluded from these analyses; as indicated in
and /ai. These differences can readily be accounted for by Figure 1, these sounds are not included in the pictured-word
both lexical differences in the two sampling modes (e.g., section of the articulation test. The overall accuracy of sound
frequent use in conversational speech of Ai/, "I") and production in both conditions followed predictable develop-
differences associated with lexical and phrasal destressing in mental patterns. These speech-delayed children produced
conversational speech. Trends were generally similar for the vowels and diphthongs most accurately, with substantially
consonant singleton and consonant cluster data, with few lowered accuracy and greater interchild variability on conso-
relatively large differences in the intended occurrence of nant singletons and clusters. As shown by the asterisks,
sounds (e.g., more frequent intended /6/ in demonstratives in vowels/diphthongs and consonant singletons were signifi-
conversational speech) and some intended vowel and diph- cantly more accurate in continuous conversational speech, a
thong sounds in conversational speech not occurring in the finding that is not consistent with trends reported in the
articulation test stimuli used for this study. speech sampling literature. Analysis at the level of individuals
confirmed these significant group-level findings for the con-
sonant data. Specifically, the results of a Percentage of
Results
The nonparametric Wilcoxon Matched Pairs Signed-Ranks 100- | Conversational Speech
test was used as the advisory inferential statistic for the major 90- O Articulation Test
between-modes analyses. Although distributional consider- a 80-
*

ations required the use of nonparametric statistics (including 70-

the large number of 0% and 100% scores that could not be 0
. 60
adjusted using transformations such as the arcsin), paramet-
C 50-
ric descriptive statistics are used in the following figures to 0)
O 40-
best illustrate patterns of central tendency and dispersion.
). 30-
20 -
Overall Accuracy 10-

0-
Figure 2 is a display of the overall accuracy of subjects in vowels and Consonant Consonant
Diphthongs Singletons Clusters
each mode, with means and standard deviations for percent- *p < .01
ages of correct vowels/diphthongs, consonant singletons, FIGURE 2. Overall articulatory accuracy In conversational
and consonant clusters shown for each of the two sampling speech and articulation testing. Significant differences were
modes. The stressed vowel /3- and the diphthong /1/ com- tested using Wilcoxon Matched Pairs Signed-Ranks Test.
 Morrison & Shriberg: Articulation Testing Versus Conversational Speech Sampling 265

Consonants Correct calculation on the consonant data in quently in response to articulation testing, compared with
both modes indicated that 47 of the 61 subjects (77%) had their occurrence in conversational speech. Just over half of
better articulation in the conversational speech mode. De- these subjects (32 of 61) had greater than 20% occurrence of
scription of the specific group- and individual-level sources of Stopping in their continuous speech samples, but no occur-
these differences in overall accuracy are explored in the rences of Stopping in the articulation test data. Specifically,
following analyses. most children stopped initial fricatives significantly more often
in conversational speech, and over 50% of these children
demonstrated significantly higher rates of final /t/ deletion in
Natural Phonological Processes Analysis conversational speech. Finally, citation form testing yielded
significantly higher frequencies of occurrence for the least
Natural processes analyses provide a summary of errors frequent of the eight processes, as indicated by data points
aggregated in descriptive units that, as noted previously, below 10% for Unstressed Syllable Deletion and Assimila-
reflect various combinations of error types on certain sounds tion.
in certain word positions. Results of such analyses, using the
definitions for eight natural processes described in Shriberg
(1986) and Shriberg and Kwiatkowski (1980), are shown in Segmental Analysis
Figure 3. The average percentage of occurrence of each of
the eight processes in each sampling condition are plotted in Vowels and diphthongs. Figure 4 includes the descrip-
descending order based on the data for conversational tive data and results of inferential statistical testing for the
speech samples. As shown in Figure 3, the general shape of percentage of correct vowels/diphthongs inthe two sampling
the trends for process occurrence in the two modes were modes. Although the majority of vowels/diphthongs were
similar. However, statistically significant differences at the .01 highly accurate in both samples, significant differences be-
level or greater were obtained on six of the eight Wilcoxon tween samples were obtained for four sounds. The vowels /a/
Matched Pair Signed-Ranks test comparisons. Recall that in and /i/ were produced more accurately incontinuous speech,
Figure 2, as well as in the figures to follow, the group means whereas /ii/ and /a/ were more accurate in articulation
are plotted, whereas the nonparametric statistical tests re- testing. Of these four differences, the findings for /a/ were
flect individual rank-order differences in performance data least expected, yet most robust. The continuous speech
obtained in each sampling mode. Hence, some of the data samples included a number and variety of lexical items
points will appear to be inconsistent with the statistical containing this reduced vowel form, with its occurrence in the
findings indicated by the asterisks. articulation test words limited to the subset of multisyllabic
Consistent with previous studies (Dunn & Davis, 1983; words described above in the analyses of structural differ-
Klein, 1984; Dyson & Robinson, 1987), Cluster Reduction ences between sampling modes. In articulation testing, chil-
and Liquid Simplification were the most frequent process- dren more often deleted /a/ in the unstressed syllable of the
level error descriptors in both sampling conditions. In con- multisyllabic words or replaced /a/ with another stressed
trast to prior reports, however, these descriptive categories vowel. In contrast, for the relatively lower proportion of
significantly more often characterized articulation test re- multisyllabic words in continuous speech, /a/ was more often
sponses. Moreover, as shown in Figure 3, both Stopping and preserved as unstressed. Moreover, /a/ was used often as an
Final Consonant Deletion occurred significantly less fre- acceptable unstressed vowel in casual speech forms.

* Conversational Speech 9 0
o Articulation Test 0
90 * * 0

z e
I- 80 0
LU U0
0
70 *
ar
0
0 60
Z e
0 0 50
*
t-
z Sf z
40
w
LI
To 30-
MU a-
a. 20
* Conversational Speech
o O *4 i
*
10
Io Articulation Test
0
CLUSTER LIUID STOPPING VELAR FINAL PALATALUNSTRESSEOASrIM- n . . .. . . . . . . .
U I II I I I
REDUCTION SIMPLI- FRONTING
CONSONANT FRONTING SYLLABLE ILATION
FICATION DELETION DELETION i Q U3 D W 0 A t I i C U S a'

*p .01 *p < .01

FIGURE 3. Natural phonological process occurrence in conver- FIGURE 4. Mean percentage of correct vowels and diphthongs
sational speech and articulation testing. Mean percent occur- inconversational speech and articulation testing sorted left to
rences is sorted left to right from most to least frequent In right from most to least frequent in conversational speech.
conversational speech. Significant differences were tested us- Significant differences were tested using Wilcoxon Matched
Ing Wilcoxon Matched Pairs Signed-Ranks Test. Pairs Signed-Ranks Test.
 266 Journal of Speech and Hearing Research 35 259-273 April 1992

Stress appeared to be implicated in the other vowel 100

* Conversational Speech
differences as well. For example, the unstressed vowel la,/ * S AO rticulation Test
was rarely correct in continuous speech (mean = 0%; range
'- 80
= 0%-25% accuracy), but it averaged approximately 8% O O
o
accuracy in articulation test responses. This difficult sound
c
for speech-delayed children was produced with somewhat 0 601
greater accuracy in the context of the word stress associated 0
50-
with articulation testing (i.e., isolated, citation-form respons- 0
LU 40
es). Although the stressed rhotic vowel /3,/ occurred with 0
insufficient frequency in articulation test forms to allow for CC
30
DI
statistical comparison, children also produced /3,/ correctly 20
*

more often in articulation testing, thus adding additional 10-

support to the association of stress with articulation accu- a

racy. NASALS GLIDES STOPS AFFRICATES FRICATIVES LIQUIDS

Consonants and features. Figure 5 displays the group
*p < .01
means for individual consonants in the two sampling modes.
The dashed line divides the 23 consonants into those aver- FIGURE 6. Mean percentage of correct segments by manner
aging greater and less than 50% correct in the two sampling features In conversational speech and articulation testing
sorted left to right from most to least frequent In conversational
modes. As shown by the trends in Figure 5, most sounds speech. Significant differences were tested using Wilcoxon
averaging greater than 50% accuracy were more accurate Matched Pairs Signed-Ranks Test.
when sampled in conversational speech, whereas those
averaging less than 50% correct were more correct when
sampled by articulation testing. Of the former group, average better in articulation testing. The reduced number of pair-
percentage correct differences were statistically significant at wise contrasts for the affricates (14 children did not attempt
the .01 level or greater for /j/, /rm/, and /n/; of the latter group, /tf' in conversational speech; 17 did not attempt /d3/) lowered
average percentage correct differences were statistically the statistical power for these comparisons.
significant for //, /s/, zl, /1/, and /r/. The one exception to this Error type and word position. Two additional segmental
trend was /g/. Although g/ was produced with greater than analyses compared the types of consonant errors made in
50% accuracy in both samples, it tended to be more accurate each word position. The six panels in Figure 7 display the
in articulation test responses. percentage of deletions, substitutions, and distortions in
Figure 6 displays the consonant data shown in Figure 5, as word-initial, word-medial, and word-final positions separately
aggregated by six manner features. Nasals, glides, and stops for consonant singletons and consonant clusters. In word-
were predictably more accurate in both sampling conditions initial position, the error-type trends for each sampling mode
compared to affricates, fricatives, and liquids. However, as are extremely similar to one other, with considerably more
indicated in Figure 6, average performance within these substitutions occurring on singleton consonants in both
trends on nasals and glides was significantly better in con- modes and relatively equal proportions of each error type
versational speech compared to articulation testing; average occurring in initial clusters. In contrast, in word-medial posi-
performance on affricates and fricatives was significantly tion, the interactions of error-type and singleton-cluster with
sampling mode are too complex for ready summary. In
1005 word-final position, the trends for singletons and clusters are
l Conversational Speech i
i'! 0 o Articulation Test
essentially similar: proportionally more deletion and distortion
90
1~~~~~~~~~~~~~~~ errors occur in continuous speech, but more frequent substi-
I- tution errors occur in articulation testing.
LA 0 · For an additional perspective on error types at the level of
01
50 * individual subjects, relative error-type percentages were cal-
0
o culated for each of the three error types. For this calculation,
I- a subject's total number of consonant errors was used as the
z
40 - denominator and numbers of each error type as numerators,
30 I * * 0 thus controlling for subjects' severity of involvement. The
LIJ
a- O *
three panels in Figure 8 provide subjects' relative error
20
scores in each sampling mode for deletions, substitutions,
'eile 0 *
10 and distortions. These subject data clearly indicate the trend
0
for children's errors to be transcribed proportionally more
m b n w h p d t k g f v t d3 e s z 1 r often as substitutions when sampled from the articulation
*p < .01
tests. Specifically, over 90% of the children had a higher
percentage of relative substitution errors in response to the
FIGURE 5. Mean percentage of correct consonants in conver- articulation test stimuli, compared to the percentage of rela-
sational speech and articulation testing sorted left to right from
most to least frequent in conversational speech. Significant tive substitution errors obtained in spontaneous continuous
differences were tested using Wilcoxon Matched Pairs Signed- speech. In comparison, relative deletion and distortion errors
Ranks Test. were more nearly equal across sampling modes: 39% of
 Morrison & Shriberg: Articulation Testing Versus ConversationalSpeech Sampling 267

* Conversational
Speech °s o0-
9 AteCulEionTest | 90 go
00~U~l~TI 501 0
,,· 90-60

0 70- 70 501 S

4
0 60-

so
o ~ to
o .4
so o-
M

ro,
so-
o 3
oI ~
0
.... SO
' ·

:o·

'°~io' o| .
*

,°1
o
l0 O

.
i·
2DJ

o i~~~~"

DELETIONS SUBSTITUTIONS DISTORTIONS DELETIONS SUBSTITUTIONS DISTORTIONS DELETIONS SUBSTITUTIONS DISTORTIONS

INITIAL SINGLETON ERRORS MEDIAL SINGLETON ERRORS FINAL SINGLETON ERRORS

lO
1. 1

10I 904
19 -
Q
50 I.,
50
50-
II
'U
. 400 C 001
S
.
O
a
20

I 1 1 o

o
DELETIONS SUBSTITUTIONS DISTORTIONS DELETIONS SUBSTITUTIONS DISTORTIONS DELETIONS SUBSTTUTIONS DISTORTIONS

INITIAL CLUSTER ERRORS MEDIAL CLUSTER ERRORS FINAL CLUSTER ERRORS

FIGURE 7. Error-type analyses of consonant singletons and consonant clusters In conversational speech and articulation
testing.

children had a higher percentage of relative deletion errors, findings in each of the two speech sampling modes. The
and 46% of children had a higher percentage of relative model for hese analyses was the clinical paradigm of a
distortion errors in response to the articulation test stimuli. pass-fail analysis, wherein each subject's status on each of
the consonant sounds is dichotomized as a pass or a fail, that
is, as a sound currently mastered or not yet mastered. As
Subphonemic Analysis there is no standard or "true" criterion against which to
Narrow phonetic transcription allowed inspection of allo- compare the sensitivity and specificity of each of the current
phone-level differences in children's productions of vowels/ sampling modes, the analyses were constructed to yield
diphthongs and consonants. These changes included the concordance rates between modes.
percentage of occurrence of casual and fast speech Procedures. Percentage correct summaries for each sub-
changes, as well as clinically relevant sound distortions and ject in each sampling mode were assembled to yield totals for
additions. Figure 9 shows the average percentage of occur- each of the consonant sounds in word-initial and word-final
rence of 34 diacritic symbols in the transcripts from the two singletons and clusters. Articulation accuracy in word-medial
speech sampling modes. Nearly half of the advisory rank- or intervocalic singletons or clusters was excluded from
order statistical comparisons were significant at or beyond these analyses due to low frequencies of occurrence in the
the .01 level; recall that the data points in these figures reflect articulation test stimuli. As in the preceding analyses, all
means data. The pattern of diacritic use transcribed for the eligible occurrences of sounds in word-initial and word-final
articulation test responses in Figure 9 can be characterized positions on the articulation test stimuli were included.
as showing strengthened articulatory movements, particu- For each consonant sound in each of the two word
larly on targets in word-initial and word-final positions. In positions, matrices were constructed to categorize the sound
particular, three sound changes appeared significantly more as 100% correct in one or both modes, 0-99% correct in one
often in articulation test responses: lengthening of vowels or both modes, or having missing data in one or both modes.
and consonants, additions of sounds represented by syn- The low absolute frequency of occurrence of some sounds in
chronic ties, and use of off-glides in productions of word-final each word position required the use of the 100% correct
vowels and consonants. In contrast, the allophones occurring criteria for "pass"; matrices with missing data in at least one
in the conversational speech samples can be characterized cell were excluded from the analyses. The individual data
as acting to assimilate and reduce sounds, evidenced by from each of these matrices were summed to yield for each
findings of increased use of the diacritic symbols for weak- of the 24 sounds, the percentage of children whose data
ening, labialization, and dentalization of consonants and indicated (a) a pass on both sampling modes, (b) a fail on
centralization of vowels. both modes, (c) a pass on articulation test, but fail on
continuous speech, and (d) a fail on articulation test, but pass
Pass-Fail Analysis on continuous speech.
Results. As shown in Table 4, results for word-initial
A series of analyses was conducted for a perspective on singletons and clusters and word-final singletons and clus-
the types of research and clinical decisions made from ters were similar. Averaging findings from the two word
268 Journal of Speech and Hearing Research 35 259-273 April 1992

3.5-
* Conversational Speech
cn DELETIONS
O Articulation Test
I
0 UJ 3.0
| Conversational Speech Z
0 Articulation Test
2.5

0 2.0
-J 0 O O
0

l- i_ 1.5 O
z 00 0 z
Lii 0
V 10 - O 1.0
nr 1 0
w
a. i- 0 0
0.5
00 d o0-
5 10 15 20 25 30 35 40 45 50 55 60 e°°0 ..........
0 Qmeni^ 0
50 nn
0
U) O SUBSTITUTIONS 0 o
-0 Oo; O o
0 0 0 0 90p09 o 0 0 0 00: O `0 OO0 OOw O
0o DIACRITIC SYMBOL
40 N
lr 0 0 00 0 FIGURE 9. Subphonemic analysis of diacritic usage In conver-
uJ sational speech and articulation testing. The diacritic symbols
30 o% o o
0o O o (described In Shriberg & Kent, 1982) are arranged from right to
left from most to least frequent In conversational speech.
-J s Aim O o0 0 0 0 0

20 ~o
Oo°"~ oo 0 o o 0
data presented in Table 4, with the top and lower panels
Ir
0 o including data for word-initial and word-final positions, re-
0 10 % spectively. For each consonant sound, the four bars indicate
Wu
11
(L the percentage of subjects who passed both sampling
a- - 1` - - -r ·---- r --- -- ·- -- · ·- ·--
1- ·-I ·· ··i · ···I ··---
·
modes, failed both modes, passed articulation testing but
0 5 10 15 20 25 30 35 40 45 50 55 60
failed in continuous speech, and failed the articulation test
5n0-
DISTORTIONS but passed in continuous speech. Percentages for each
V) sound were derived from the total number of children for
0
, 40- whom sampling mode comparisons were possible. Consis-
tent with the group and individual analyses, the data in Figure
W 10 indicate that the unconditional frequencies at which
0-
sounds are articulated correctly does not completely predict
0
uJ concordance rates (Kearns & Simmons, 1988). That is,
%0
2 0- 0 0 concordance rates are not highest for all of the "easy" and
I3 a00
0 .,,... 0 "hard" sounds to articulate and lowest for sounds of inter-
*~O 0
0--
mediate difficulty. Rather, as divided into the four outcome
categories, concordance status is marked by extensive inter-
a.
uJ~~~~
oo~ ~~
0 0
o 00
-''qo 0
e,. 00 o0
actions among sounds and word positions. The data in
v0 5 10 15 20 25 30 35 40 45 50 55 60
Figure 10 allow for a sound-by-sound inspection of these
interactions. Review of the transcripts indicates that these
SUBJECTS
subject-level concordance rates for each sound would be
FIGURE 8. Relative error type analyses of consonant errors in even more discrepant if broken out by consonant singletons
conversational speech and articulation testing for individual and consonant clusters.
subjects.

TABLE 4. Pass-fall analysis for each of the 24 consonants

positions, approximately 71% (17) of the consonant sounds articulated by 61 children In articulation testing and conversa-
received similar clinical decisions in the two sampling tional speech.
modes, approximately 22% (5) of the consonant sounds
passed the articulation test, but failed on the continuous Percentage of
Position of sounds
speech sample, and approximately 7% (2) of the consonant singleton or
sounds failed the articulation test, but passed on the con- Pass-fail status cluster In word M SD
tinuous speech sample. Thus, on average, dichotomous
categorization based on the two sampling modes was Pass or fail both Initial 72.5 13.4
sampling modes Final 69.1 11.8
similar for approximately two thirds of the 24 consonant
sounds. Based on the standard deviations in Table 4, Pass articulation test; fail Initial 21.0 11.1
outcome discrepancies between the two sampling modes conversational speech Final 23.4 12.4
for two thirds of the 61 children occurred on as few as 13
sounds to as many as 20 sounds. Fail articulation test; pass Initial 6.7 6.4
conversational speech Final 7.3 6.8
Figure 10 is a display of the sound-level information on the
 Morrison & Shriberg: Articulation Testing Versus ConversationalSpeech Sampling 269

FIGURE 10. Pass-fall analyses of consonant singletons and clusters inarticulation testing (AT)
and conversational speech (CS). The upper panel Includes the available data for word-initial
position; the lower panel includes the available data for word-final position.

IndividualDifference Analysis ables: age, average words per utterance, percentage of

vowels correct, percentage of singleton consonants correct,
A final series of analyses was designed to explore whether percentage of consonant clusters correct, and intelligibility. A
concordance between modes was associated with several total of 12 analyses were performed, with the data based on
individual difference variables that were available in the word-initial and word-final positions intended as cross-vali-
database. dational. The question was whether children's status on any
Procedures. The data from Table 4 and Figure 10 were of these demographic, language, or speech variables was
used to assign each of the 61 children to one of five associated with the relative number of sounds on which the
categories for each word position. Skew and kurtosis values two sampling modes yielded different pass/fail outcomes in
and associated plots supported a parametric approach to each of the two word positions.
categorization, wherein children could be assigned to four Results. Table 5 is a summary of the results of the
groups based on their location in the distribution of scores individual differences analyses. No clear trends are apparent
relative to group means and standard deviations. Children across the four groups. The only difference among the 12
whose percentage of inter-mode concordance for consonant comparisons to approach statistical significance, using a
sounds was above one standard deviation from the group conservative a level adjusted for multiple tests, was for the
were designated group A, children who scored within a word-initial data on Average Words Per Utterance (H =
standard deviation of the mean on each of the three agree- 10.45; p < .02), a language metric that correlates in the high
ment figures were designated group B, children who scored .90s with mean length of utterance (Shriberg, 1986). The fact
above one standard deviation from the group mean on that the median data for word-final position are opposite in
percentage of sounds passed on the articulation test and direction to the word-initial data for groups C and D-the
failed in continuous speech were designated group C, chil- groups with the largest median differences-suggests that
dren who scored above one standard deviation from the this trend may not be reliable. Essentially, these final analy-
group mean on percentage of sounds failed on the articula- ses indicated that none of the individual differences assessed
tion test and passed in continuous speech were designated was strongly associated with concordance rates for perfor-
group D, and children who scored above one standard mance in the two sampling modes.
deviation from the mean on both articulation pass/continuous
speech fail and articulation test fail/continuous speech pass
were designated group E. Group E was discarded because Discussion
only one child in each word position met the description.
Each of the other four groups included 7 to 34 children, with Several methodological considerations should be kept in
approximately half of the children meeting criteria for group B mind in review of these findings. First, the speech analyses
for both word-initial and word-final analyses. were based solely on perceptual methods, with associated
With membership in one of the four groups used as the consequences for issues of validity and reliability (Shriberg
independent variable, Kruskal-Wallis One-Way Analyses of and Lof, in press). Compared to data obtained by acoustic,
Variance were performed on each of six dependent vari- kinematic, or physiological methods, each of which also has
270 Journal of Speech and Hearing Research 35 259-273 April 1992

TABLE 5. Individual difference analyses results based on the four-category, pass-fall data". See text for an explanation of the four
categories.
Percent Percent
Average Percent correct correct
Age words per correct consonant consonant Intelligibility
(months) utterances vowels singletons clusters (%)
Category Brief description Initial Final Initial Final Initial Final Initial Final Initial Final Initial Final
A Above one standard deviation 53.0 52.0 4.1 3.0 92.4 92.2 64.0 58.1 50.0 40.1 92.5 82.0
agreement between modes
B Within one standard deviation 49.5 48.0 3.7 3.5 92.0 92.4 66.1 66.7 52.1 50.5 92.2 91.3
agreement between modes
C Above one standard deviation 55.0 55.0 4.6 3.7 92.9 92.1 64.6 64.8 45.5 51.3 90.0 91.8
for sounds passed on
articulation test and failed in
conversational speech
D Above one standard deviation 49.5 54.0 3.2 4.3 90.6 91.1 67.2 73.6 48.1 55.6 79.2 92.7
for sounds failed on
articulation test and passed
inconversational speech
Note. a All table entries are medians; speech entries are taken from the conversational speech data. b Word-initial: p < .05 (Kruskal-Wallis H
value adjusted for ties).

associated validity and reliability constraints, the present The specific structural and content differences between
data rely solely on the auditory-perceptual vigilance and and within the two speech sampling modes assessed in this
consensus decisions of persons well trained in phonetic study are difficult to capture in one framework. Put most
transcription of disordered speech. Second, as reviewed generally, in contrast to the monosyllabic and multisyllabic
previously, the continuous speech and citation form data nouns that comprised the articulation test stimuli, the simple
were provided by different research transcription teams. canonical structures in the recurrent function words occurring
Although detailed examination of transcription reliability for in the conversational speech included proportionally more of
these consensus transcription teams suggests excellent in- the earlier developing consonant sounds. Accordingly, a
terteam agreement at the level of broad phonetic transcrip- significant portion of the variance in the findings displayed in
tion, transcription differences are a potential source of vari- Figures 5 and 6 is assumed to reflect asymmetries in the
ance. Third, the findings are based on comparisons of lexical contexts for earlier and later occurring consonants,
continuous speech data to only one of the dozens of stan- with the trend for the earlier developmental sounds to be
dardized articulation tests, using whole-word transcription more correct in continuous speech due to the lexical stimuli in
rather than transcriptions of only the targeted sounds. Lastly, which they are tested. The present database is not appropri-
there were asymmetries in the token frequencies of sounds, ate for a well-balanced inspection of the validity of this
with more tokens for most, but not all, sounds available from hypothesis. Controlled studies would require adequate to-
the conversational speech samples. kens of comparable stimuli in each mode.
Potential variance associated with transcriber tasks fo-
Research Considerations cuses on the relative perceptual salience of articulatory
targets in relation to the maintenance of stable response
The clear findings of this study are that the two sampling definitions for correct and incorrect articulation. The cogni-
modes yield significant differences in the speech profiles of tive-perceptual demands of sentential transcription com-
speech-delayed children. As in prior literature, the group- pared to word-level phonetic transcription tasks have not
level and individual graphs and statistical effects do not all been explored. Stress-related sound changes in continuous
indicate that articulation testing yields higher performance speech production, such as reduced loudness and segment
scores. Rather, sampling mode differences in both directions durations, may be associated with more liberal response
were obtained at the linguistic levels of sound class, manner definitions. In contrast, the clarity of word boundaries in
feature, phonological process, phoneme, error type, word single-word articulation tests may require subtle but system-
position, and allophone. The complex pattern of speech atically more articulate production to be considered correct.
sampling findings defies a consolidated explanation; that is, Moreover, in the richer contrasts provided in continuous
each finding cannot be neatly attributed to collateral effects speech, a child's error pattern on stressed compared to
associated with either perceptual (transcriber) or production unstressed contexts could influence transcribers' perceptual
(speaker) processes. Rather, an eventual account of the criteria for both contexts. Thus, the relatively homogeneous
influence of sampling mode on research and clinical speech contexts of repeatedly stressed nouns in citation forms may
data will likely require a multifactorial framework. Therefore, provide the perceptual context for stable response defini-
the following few comments only extend the research impli- tions, compared to those engendered by the constantly
cations for sampling, transcriber, and speaker variables changing stress patterns of continuous speech. Such poten-
discussed previously. tial transcription processes are undocumented in the clinical
 Morrison & Shriberg: Articulation Testing Versus ConversationalSpeech Sampling 271

and research speech pathology literature (see Shriberg & as a child moves to the segmental and suprasegmental
Lof, 1991, for extended discussion). demands of conversational speech. With approximately 80%
The speaker's contribution to the sampling mode differ- of this clinical population having associated language pro-
ences also involve the individual and interactive contributions duction problems, and approximately 25% having associated
of cognitive-linguistic and pragmatic processes. The pattern prosody-voice involvement (Shriberg, 1991; Shriberg et al.,
of increased consonant deletions in continuous speech, 1986; Shriberg, Kwiatkowski, & Rasmussen, 1990), con-
which is consistent with the relatively higher rates of constit- straints in continuous speech could be associated with many
uent omissions reported in the grammatical analyses of levels of psycholinguistic and motor-speech processing.
language-disordered children, may reflect a necessary shift Conversational speech samples would appear to be the only
in attention and speaking strategies across sampling con- source of integrated speech, language, and prosodic analy-
texts. In the articulation test format, emphasis is placed on ses needed to assess, plan intervention for, and monitor the
lexical and phonological accuracy, that is, on retrieving the progress of these children's individual phonological error
appropriate word and articulating it correctly. In contrast, in patterns.
conversational samples, emphasis shifts to maintaining the
accuracy and flow of ideas, often at the expense of less
salient or important elements of form. In the present data, Conclusions
children produced developmentally earlier sounds better in
spontaneous conversational speech and developmentally Reliable estimates of the interactive effects of sample,
later sounds better in response to articulation test stimuli. transcriber, and speaker variables in speech assessment will
Thus, when composed of familiar meaning, intentions, and require carefully designed and controlled studies, moving
forms, spontaneous conversational speech may provide the well beyond the level of descriptive findings available from
most optimal setting for production of well-established analyses of the present data and those cited throughout this
sounds or structures. In contrast, for less well-established report. Notwithstanding the lack of a comprehensive explan-
sounds, the extensive linguistic, motor-speech, and prag- atory account, the current findings do suggest that attention
matic demands of that context may lead to decreased to the potential effects of sampling mode is a crucial meth-
articulatory accuracy. odological need in child phonology research. As reviewed
previously, single-word articulation tests and citation-form
protocols continue to be used as the primary speech sam-
Clinical Considerations pling mode for many contemporary research programs in
developmental and nondevelopmental phonological disor-
Clinical implications of these data warrant brief comment, ders. Apparently, such methodological decisions are de-
as most intervention programs include stages progressing fended on utility and efficiency criteria and especially on the
from single-word training to naturalistic continuous speech. assumption of equivocal speech sampling findings in the
The present findings are consistent with the clinical observa- archival literature. The present findings suggest that, in fact,
tion that single-word training often is associated with exag- statistically, clinically, and potentially theoretically significant
gerated speech, including lengthening of both the target speech performance differences are associated with sam-
sound and the vowel nucleus, addition of on-glides and pling mode. If these subjects' speech had been sampled
off-glides at word boundaries, affrication of initial and final using only citation-form stimuli, some portion of the data and
obstruents, and other featural and segmental distortions. subsequent conclusions would be incomplete and possibly
Such speaker behavior may, of course, at least partly mirror misleading relative to their performance in conversational
exaggerations in the stimuli provided for imitation by the speech samples obtained during the same assessment
examiner or speech-language pathologist. In contrast, fast or session.
casual speech requires reduction, not addition, of features; Put most strongly, the present findings suggest that cita-
the transitions from word to word, which turn final consonants tion-form testing yields neither typical nor optimal measures
into syllable-releasing segments, cannot be exaggerated of speech performance. Thus, as long as researchers and
without a disruption in prosody. practitioners continue to base assessment results solely on
For children pressed to progress from single-word to different citation form tests, including the standardized artic-
continuous speech tasks, their options may be either to ulation tests that continue to proliferate, sampling procedures
produce the newly learned forms with exaggerated begin- will continue to be a major source of bias in phonological
nings and endings or delete the segments entirely. In the assessment (Butcher, 1990; Smit, 1990, 1991). Especially in
present study, transcribers reported (annotated on the tran- view of emerging developments in nonlinear phonological
script) a "choppiness" or discontinuity between words in the theories, with their promise for integrated analyses of
continuous speech of many of these speech-delayed chil- speech-language-prosody (Goldsmith, 1990; McGregor &
dren. These comments alluded to aspects of precision and Schwartz, 1991; Shriberg, 1990; Shriberg et al., 1990) and
timing that reflect the integration of segmental processing the utility of conversational speech samples for measures of
with suprasegmentals, including lexical, phrasal, and em- intelligibility and severity of involvement (e.g., Garrett &
phatic stress. Such observations are not unfamiliar to Moran, in press), conversational speech sampling would
speech-language pathologists, wherein deletion of well-es- seem to be the measurement procedure of choice. For
tablished consonants such as /t/, /d/, and /n/ and stopping of continuous speech sampling of children with marked intelli-
well-established sounds at word boundaries occurs as soon gibility problems, Kwiatkowski and Shriberg (1991) describe
272 Journal of Speech and Hearing Research 35 259-273 April 1992

procedures that might be useful to augment examiner Fudala, J. B. (1974). Arizona Articulation Proficiency Scale (Rev.
glosses with those provided by a caregiver. ed.). Los Angeles: Western Psychological Services.
Gallagher, R., & Shrlner, T. (1975). Contextual variables related to
inconsistent /s/ and /z/ production in the spontaneous speech of
children. Journal of Speech and Hearing Research, 18, 623-633.
Acknowledgments Garrett, K. K., & Moran, M. J. (in press). A comparison of phono-
logical severity measures. Language, Speech, and Hearing Serv-
We express our sincere thanks to a number of persons who ices in Schools.
provided competent research and editorial assistance at different Goldman, R., & Fristoe, M. (1969;1972). Goldman-Fristoe Test of
stages of this work: Barri Babow, Maria Cavicchio, Patricia Enge- Articulation. Circle Pines, MN: American Guidance Service.
bose, Frederic Gruber, Rebecca Hinke, Sara Hoffman, Gregory Lof, Goldsmith, J. A. (1990). Autosegmental and metrical phonology.
Jane Loncke, Amparo Ortiz, Carmen Rasmussen, Dorothy Rorick, Oxford, UK: Basil Blackwell, Ltd.
Catherine Trost-Steffen, and Carol Widder. We also acknowledge Harris, J., &Cottam, P. (1985). Phonetic features and phonological
the continuing support of the Madison Metropolitan School District, features in speech assessment. British Journal of Disorders of
Integrated Student Services, with special thanks to Beth Daggett and Communication, 20, 61-74.
Susan Albert. This work was supported by grants from the United Healy, T. J., & Madison, C. L. (1987). Articulation error migration: A
States Department of Education, G008400633 and the National comparison of single word and connected speech samples. Jour-
Institute on Deafness and Other Communicative Disorders, nal of Communication Disorders, 20, 129-136.
DC00496. Henderson, F. M. (1938). Accuracy in testing articulation of speech
sounds. Journal of Educational Research, 31, 348-356.
Hodson, B. (1980). The assessment of phonological processes.
Danville, IL: The Interstate Press.
References Hymes, D. H. (1971). Competence and performance in linguistic
theory. In R. Huxley & E. Ingram (Eds.), Language acquisition:
Andrews, N., & Fey, M. E. (1986). Analysis of the speech of Models and methods (pp. 3-28). New York: Academic Press.
phonologically impaired children in two sampling conditions. Lan- Johnson, J., Winney, B., & Pederson, . (1980). Single word
guage, Speech, and Hearing Services in Schools, 17, 187-198. versus connected speech articulation testing. Language, Speech,
Bankson, N. W., & Bernthal, J. E. (1982). A comparison of phono- and Hearing Services in Schools, 11, 175-179.
logical processes identified through word and sentence imitation Jordan, E. P. (1960). Articulation test measures and listener ratings
tasks of the PPA. Language, Speech, and Hearing Services in of articulation defectiveness. Journal of Speech and Hearing
Schools, 13, 96-99. Research, 3, 303-319.
Bernthal, J. E., & Bankson, N. W. (1988). Articulation and phono- Kearns, K. P., & Simmons, N. N. (1988). Interobserver reliability
logical disorders (2nd ed.). Englewood Cliffs, NJ: Prentice-Hall. and perceptual ratings: More than meets the ear. Journal of
Bernthal, J. E., Grossman, F. M., & Goll, A. E. A. (1989). Phono- Speech and Hearing Research, 31, 131-135.
logically delayed children's responses to three types of pictured Kenney, K., Prather, E., Mooney, M., & Jeruzal, N. (1984). Compar-
stimuli. Journal of Childhood Communication Disorders, 12, 137- isons among three articulation sampling procedures with preschool
143. children. Journal of Speech and Hearing Research, 27, 226-231.
Blodgett, E., &Miller, V. (1986, November). The effect of utterance Klatt, D. H.(1989). Review of selected models of speech perception.
length and imitation on disordered phonology. Paper presented at In W. Marslen-Wilson (Ed.), Lexical representation and process
the Annual Convention of the American Speech-Language-Hear- (pp. 169-226). Cambridge, MA: The MIT Press.
ing Association, Detroit. Klein, H. B. (1984). Procedure for maximizing phonological informa-
Butcher, A. (1990). The uses and abuses of phonological assess- tion from single-word responses. Language, Speech, and Hearing
ment. Child Language Teaching and Therapy, 6, 262-276. Services in Schools, 15, 267-274.
Campbell, T., & Shrlberg, L. D. (1982). Associations among prag- Klein, H. B., & Spector, C. (1981, November). The relationship
matic functions, linguistic stress, and natural phonological pro- between serial-position/stress level interaction and consonant
cesses in speech-delayed children. Journal of Speech and Hear- errors. Paper presented at the Annual Convention of the American
ing Research, 25, 547-553. Speech-Language-Hearing Association, Los Angeles.
Chapman, R., &Ting, A. (1971). The effect of mode of elicitation in Klein, H. B., & Spector, C. (1985). Effect of syllable stress and serial
articulation testing. Technical Report No. 154. Wisconsin Re- position on error variability in polysyllabic productions of speech-
search and Development, Center for Cognitive Learning, Univer- delayed children. Journal of Speech and Hearing Disorders, 50,
sity of Wisconsin-Madison. 391-402.
Compton, A. J., & Hutton, S. S. (1978). Compton-Hutton Phono- Kohn, S. E. (1988). Phonological production deficits in aphasia. In
logical Assessment. San Francisco: Carousel House. H. A. Whitaker (Ed.), Phonological processes and brain mecha-
DuBols, E., & Bernthal, J. E. (1978). A comparison of three nisms (pp. 93-117). New York: Springer-Verlag.
methods for obtaining articulatory responses. Journal of Speech Kwiatkowskl, J., & Shrlberg, L. D. (in press). Intelligibility assess-
and Hearing Disorders, 43, 295-305. ment in developmental phonological disorders: Accuracy of care-
Dunn, C. (1982). Phonological process analysis: Contributions to giver gloss. Journal of Speech and Hearing Research.
assessing phonological disorders. Communicative Disorders, 7, Leopold, W. (1947). Speech development of a bilingual child. A
147-163. linguist's record. Vol. 2. Sound-learning in the first two years.
Dunn, C., & Davis, B. (1983). Phonological process occurrence in Evanston, IL: Northwestern University Press.
phonologically disordered children. Applied Psycholinguistics, 4, Levelt, W. J. M. (1989). Speaking: From intention to articulation.
187-207. Cambridge, MA: The MIT Press.
Dyson, A. T., & Robinson, T. W. (1987). The effect of phonological Locke, J. (1983). Phonological acquisition and change. New York:
analysis procedure on the selection of potential remediation tar- Academic Press.
gets. Language, Speech, and Hearing Services in Schools, 18, Lowe, R. (1986). Assessment link between phonology and articula-
364-377. tion. Moline, IL: Lingui-Systems.
Elbert, M., Dinnsen, D. A., Swartzlander, P., &Chin, S. B. (1990). McGregor, K. K., & Schwartz, R. G. (in press). Converging evi-
Generalization to conversational speech. Journal of Speech and dence for underlying phonological representation. Journal of
Hearing Disorders, 55, 694-699. Speech and Hearing Research.
Faircloth, M.A., & Faircloth, S. R. (1970). An analysis of the Menyuk, P. (1980). The role of context in misarticulations. In G.
articulatory behavior of a speech-defective child in connected Yeni-Komishian, J. Kavanaugh, & C. Ferguson (Eds.), Child
speech and in isolated-word responses. Journal of Speech and phonology: Vol. 1.Production (pp. 211-226). New York: Academic
Hearing Disorders, 35, 51-61. Press.
 Morrison & Shriberg: Articulation Testing Versus ConversationalSpeech Sampling 273

Miller, J. (1981). Assessing language production in children. Austin, Shrlberg, L. D., & Lof, G. L. (1991). Reliability studies in broad and
TX: Pro-ed. narrow phonetic transcription. Clinical Linguistics and Phonetics,
Nlttrouer, S., & Boothroyd, A. (1990). Context effects in phoneme 5, 225-279.
and word recognition by young children and older adults. Journal Shriberg, L. D., & Olson, D. (1987). PEPAGREE: Programs to
of the Acoustical Society of America, 87, 2705-2715. compute transcriber agreement. Madison, WI: Waisman Center on
Orr, E., Blodgett, E., & Miller, V. (1983, November). Phonological Mental Retardation and Human Development.
analysis of spontaneous and imitated speech samples. Paper Siegel, R., Winltz, H., & Conkey, H. (1963). The influence of testing
presented at the Annual Convention of the American Speech- instruments on articulatory responses in children. Journal of
Language-Hearing Association, Cincinnati. Speech and Hearing Disorders, 28, 67-76.
Paden, E., & Moss, S. (1985). Comparison of three phonological Simmons, R., Blodgett, E., & Miller, V. (1983, November). Assess-
analysis procedures. Language, Speech, and Hearing Services in ment of phonological disorders in conversation and single word
Schools, 16, 103-109. picture naming. Paper presented at the Annual Convention of the
Paynter, E., & Sims, F. (1979, November). A comparison of two American Speech-Language-Hearing Association, Cincinnati.
types of articulation assessment. A paper presented at the Annual Smit, A. B. (1990, November). Three kinds of variability (inconsis-
Convention of the American Speech-Language-Hearing Associa- tency) in children's speech. Paper presented at the Annual Con-
tion, Atlanta. vention of the American Speech-Language-Hearing Association,
Pendergast, K., Dickey, S., Selmar, J., & Soder, A. (1969). Photo Seattle.
Articulation Test (2nd ed.). Danville, IL: The Interstate Publishers. Smit, A. B. (1991). Children's mimicry of voiced and voiceless
Schmitt, L. S., Howard, B. H., & Schmitt, J. F. (1983). Conversa- English stops: An investigation of relative ease of production.
tional speech sampling in the assessment of articulation profi- Manuscript submitted for publication.
ciency. Language, Speech, and Hearing Services in Schools, 14, Snyder, L. (1984). Communicative competence in children with
210-214. delayed language development. In R. Schiefelbusch (Ed.), The
Selkirk, E. (1984). Phonology and syntax: The relation between acquisition of communicative competence (pp. 423-478). Balti-
sound and structure. Cambridge, MA: The MIT Press. more: University Park Press.
Shrlberg, L. D. (1986). PEPPER (Programs to Examine Phonetic Stemberger, J. (1988). Between-word processes in child phonol-
and Phonologic Evaluation Records). Hillsdale, NJ: Lawrence ogy. Journal of Child Language, 15, 39-61.
Earlbaum. Stoel-Gammon, C. (1987). Phonological skills of 2-year-olds. Lan-
Shriberg, L. D. (1990). Measurement validity. In L. B. Olswang, guage, Speech, and Hearing Services in Schools, 18, 323-329.
C. K. Thompson, S. F.Warren, & N. J. Minghetti (Eds.), Treatment Stoel-Gammon, C. (1988). Evaluation of phonological skills in
efficacy research in communication disorders (pp. 63-78). Rock- preschool children. Seminars in Speech and Language, 9, 15-25.
ville, MD: American Speech-Language-Hearing Foundation. Templin, M.C., & Darley, F. L. (1969). The Templin-Darley Tests of
Shrlberg, L. D. (1991). Directions for research in development Articulation (2nd ed.). Iowa City: Bureau of Educational Research
phonological disorders. In J. F. Miller (Ed.), Research on child and Service.
language disorders: A decade of progress (pp. 267-276). Austin, Van Demark, D. R. (1964). Misarticulations and listener judgments
TX: Pro-ed. of the speech of individuals with cleft palates. Cleft Palate Journal,
Shrlberg, L. D., Hinke, R., & Trost-Steffen, C. (1987). A procedure 1, 232-245.
to select and train persons for narrow phonetic transcription by Vihman, M., & Greenlee, M. (1987). Individual differences in pho-
consensus. Clinical Linguistics and Phonetics, 1, 171-189. nological development: Ages one and three years. Journal of
Shrlberg, L. D., & Kent, R. D. (1982). Clinical phonetics. New York: Speech and Hearing Research, 30, 503-521.
Macmillan. Watson, M. M. (1989). Comparison of three methods for eliciting
Shrlberg, L. D., & Kwlatkowski, J. (1980). Natural Process Analy- phonological processes. Perceptual and Motor Skills, 69, 771-
sis: A procedure for phonological analysis of continuous speech 778.
samples. New York: Macmillan. Weiner, F. (1979). Phonological Process Analysis. Baltimore: Uni-
Shrlberg, L. D., & Kwiatkowski, J. (1982). Phonologic disorders Il: versity Park Press.
A procedure for assessing severity of involvement. Journal of Weiss, C. (1978). The Weiss Comprehensive Articulation Test.
Speech and Hearing Disorders, 47, 256-270. Boston: Teaching Resources.
Shriberg, L. D., & Kwlatkowski, J. (1985). Continuous speech Wren, C. T. (1985). Collecting language samples from children with
sampling for phonologic analyses of speech-delayed children. syntax problems. Language, Speech, and Hearing Services in
Journal of Speech and Hearing Disorders, 50, 323-334. Schools, 16, 83-102.
Shrlberg, L. D., Kwiatkowskl, J., Best, S., Hengst, J., &Tersellc-
Weber, B. (1986). Characteristics of children with speech delays
of unknown origin. Journal of Speech and Hearing Disorders, 51,
140-160. Received January 31, 1991
Shriberg, L. D., Kwlatkowski, J., & Hoffmann, K. (1984). A proce- Accepted July 29, 1991
dure for phonetic transcription by consensus. Journal of Speech
and Hearing Research, 27, 456-465. Contact author: Lawrence D. Shriberg, PhD, Phonology Project,
Shrlberg, L. D., Kwlatkowski, J., & Rasmussen, C. (1990). The Waisman Center on Mental Retardation and Human Development,
Prosody-Voice Screening Profile. Tucson, AZ: Communication University of Wisconsin-Madison, 1500 Highland Avenue, Madison,
Skill Builders. WI 53705.