Content uploaded by Sandra Lyn Martin-Chang
Author content
All content in this area was uploaded by Sandra Lyn Martin-Chang on Sep 14, 2018
Content may be subject to copyright.
INT J LANG COMMUN DISORD,NOVEMBER–DECEMBER 2014,
VOL.49,NO.6,637–648
Research Report
Thinking aloud: effects on text comprehension by children with specific
language impairment and their peers
Brenna McClintock, Diane Pesco and Sandra Martin-Chang
Department of Education, Concordia University, Montreal, QC, Canada
(Received June 2013; accepted December 2013)
Abstract
Background: Many lines of evidence now suggest that inferencing plays a substantial role in text comprehension.
However, inferencing appears to be difficult for children with language impairments, many of whom are also
struggling readers.
Aims: To assess the effects of a ‘think-aloud’ procedure on inference generation and narrative text comprehension
by children with expressive–receptive specific language impairment (SLI) and age-matched peers with typical
language development (TLD).
Methods & Procedures: An SLI group (n=12; mean age =10;5) and an age-matched TLD group (n=12)
participated in the study. Narrative passages were read silently by participants and simultaneously read aloud by
the examiner in two conditions: (1) uninterrupted reading and (2) a think-aloud, in which children verbalized their
understanding as the text was read. Following the passages in both conditions, children responded to comprehension
questions requiring either literal or inferential information (specifically, ‘informational’ and ‘causal’ inferences).
The children’s comprehension scores were analysed by group, condition and question type. The statements
children generated during the think-aloud were also compared by group and examined in relation to children’s
comprehension scores.
Outcomes & Results: The SLI group scored lower than the TLD group on all questions (literal, informational and
causal), in both conditions. For both groups, however, comprehension scores on all three types of questions increased
when the think-aloud procedure was implemented. During the think-aloud, the SLI group generated a comparable
number of literal statements compared with the TLD group, but fewer informational and causal statements. The
number of causal statements children made correlated with their scores on the inferential comprehension questions.
Conclusions & Implications: Children with expressive–receptive SLI showed poorer comprehension of narrative
texts than children with TLD, as expected. However, both groups’ comprehension improved when participating in
the think-aloud condition. While further investigation is warranted, the think-aloud procedure shows promise as a
strategy to enhance narrative text comprehension in school-age children with, and without, language impairments.
Keywords: specific language impairment, inferencing, narrative comprehension, think-aloud.
What this paper adds?
Think-alouds, defined as verbalizing thoughts during a task, have been proposed as a means of revealing the inferences
children make as they listen to or read texts. Studies additionally suggest that engaging in a think-aloud might benefit
text comprehension. In the present study, children with expressive–receptive language impairments and a typically
developing comparison group participated in a think-aloud about narrative texts, with positive effects on their
comprehension of both literal and implicit (inferential) content. The study is the first to demonstrate that a think-
aloud procedure supports narrative text comprehension for children with SLI. The results show promise for the use
of think-aloud procedures in educational and speech–language therapy settings.
Address correspondence to: Diane Pesco, Department of Education (LB-501–3), Concordia University, 1455 Boul. de Maisonneuve West,
Montreal, QC, H3G 1M8, Canada; e-mail: dpesco@education.concordia.ca
International Journal of Language & Communication Disorders
ISSN 1368-2822 print/ISSN 1460-6984 online C⃝2014 Royal College of Speech and Language Therapists
DOI: 10.1111/1460-6984.12081
638 Brenna McClintock et al.
Introduction
In upper elementary school, building a foundation of
academic knowledge is central to success. To a large
extent this is accomplished by interacting with written
texts. However, many children struggle fully to com-
prehend what they have read. Recent estimates in the
United States, for example, suggest that the majority
of fourth and eighth graders demonstrate only a ‘basic’
level of reading comprehension (National Center for
Education Statistics 2011).
While there has been some discussion about the
number of independent skills that contribute to reading
comprehension, there is wide agreement that in addi-
tion to decoding, oral language abilities are important
(Gough and Tunmer 1986). Clearly, grasping literal in-
formation in a text depends partly on semantic and
syntactic knowledge. However, ‘higher-level’ language,
particularly inferencing—or deducing and integrating
information that is not overtly provided—also plays
a critical role in comprehension (Bowyer-Crane and
Snowling 2005, Oakhill and Cain 2012, van Kleeck
2008). Evidence that children with language impair-
ment make fewer inferences than peers when reading or
listening to texts is thus concerning (Adams et al. 2009,
Botting and Adams 2005, Dodwell and Bavin 2008).
The current study examined the effect of a think-
aloud procedure on the comprehension of literal and
implicit information by children with specific language
impairments and age-matched peers. Here, children an-
swered comprehension questions following two condi-
tions: a standard, ‘uninterrupted’ reading of a story, and
a think-aloud, where the children verbalized their un-
derstanding as they progressed through the story.
Specific language impairment (SLI)
Children with SLI have persistent difficulty with first
language acquisition. Expressive and receptive language
skills may both be affected, or expressive skills may be
the principal difficulty (receptive problems alone are
rare). To be diagnosed with SLI, children must meet
certain exclusionary criteria; they cannot, for instance,
be diagnosed with hearing impairment, neurological dis-
order (e.g., epilepsy), autism or intellectual impairment
(Rice 2009). According to inclusionary criteria used in
research, their scores on norm-referenced language mea-
sures should be at least 1 SD (standard deviation) below
the mean (Rice 2009), but in clinical practice a more
stringent criterion (i.e., 1.5 or 2 SDs below the mean)
is often applied.
In addition to the language difficulties that charac-
terize SLI, studies have shown that children with SLI
are at increased risk for reading difficulties. Catts et al.
(2002) reported that children with SLI were four to five
times more likely to have a reading impairment than
their typically developing peers in Grades 2 and 4, and
that reading difficulties often persisted through high
school (Catts et al. 2008).
Models of text comprehension
According to the simple view of reading by Hoover and
Gough (1990), comprehension of written texts is the
product of decoding ability and linguistic comprehen-
sion (R =D×L). These two components are equally
important and, amongst good readers, positively related;
as the two components improve, so does reading com-
prehension (Hoover and Gough 1990). However, the
correlation between decoding and linguistic compre-
hension does not always hold. Some children with SLI
resemble ‘poor comprehenders’ whose decoding abil-
ity is at, or above, age level, in conjunction with poor
comprehension (Nation et al. 2004). For other children
with SLI, both decoding and comprehension fall be-
low levels expected for their chronological age (Kelso
et al. 2007), but comprehension levels may still be
lower than decoding skills. For example, Bishop and
Adams (1990: 1033) described the reading comprehen-
sion scores of 8-year-old children with SLI as ‘dispro-
portionately poor’ given their decoding skills, and con-
cluded that the deficits observed were better explained
by poor language comprehension than by poor reading
accuracy.
While inferencing is a complex skill influenced by
pragmatic and even non-linguistic factors (Botting and
Adams 2005), there is reason to hypothesize that supe-
rior language comprehension fosters particularly effec-
tive inference generation. According to Kintsch’s (1988)
construction–integration model, text comprehension
occurs in two phases. In the first phase (construction), the
meaning of propositions raised in the text is activated
in the reader’s mind. The activated concepts enable the
reader to generate a number of inferences, some of which
may be incorrect. Thus, during this phase the reader
develops an enriched but incomplete and potentially in-
accurate representation of the text. In the second phase
(integration), the reader uses working memory to carry
information forward, and to select the propositions and
inferences most appropriate to the new section of text.
Additionally at this stage, more specific and controlled
inferences are made by the reader to fill in and make
casual connections between parts of the text. The end
result of the integration phase is a coherent and mean-
ingful representation, comprising many inferences that
both elaborate on, and connect, different parts of the
text. However, if inferencing relies partly on language
Thinking aloud effects on text comprehension by children with SLI 639
comprehension, then it may be compromised for chil-
dren with language impairments, as well as being poten-
tially affected by other factors, such as working memory
and world knowledge relevant to the text. Moreover,
children with language impairments may not utilise in-
ferences to the degree that they are able, but rather focus
on accurate decoding as other ‘poor comprehenders’ ap-
pear to do (Cain and Oakhill 1999).
Inferences and text comprehension
Two types of inferences, as described by van Kleeck
(2008),1were assessed in the present study: causal and
informational. Causal inferences entail linking two parts
of a text by incorporating world knowledge to fill in
missing information about cause-and-effect relation-
ships (Wright and Newhoff 2001). Such inferences help
readers to create a coherent and meaningful representa-
tion of the text. In the context of a story, causal inferences
are particularly important because they link story com-
ponents, i.e., the initiating event or problem of the story;
the protagonist’s internal states, goals and attempts to re-
solve the problem; the consequences of these attempts;
and finally the resolution or outcome of the problem
(van Kleeck 2008).
Studies attest to the importance of generating causal
inferences to text comprehension by children. On both
reading and listening tasks, the majority of inferences
made by third-graders described as ‘good comprehen-
ders’ were causal, while poor comprehenders made fewer
inferences of this type (Laing and Kamhi 2002). Cain
and Oakhill (1999) found that causal inferences pre-
dicted variance in the reading comprehension of good
and poor comprehenders (mean age 7;8), while measures
of literal comprehension did not.
The second type of inference explored here is ‘infor-
mational’. Informational inferences involve using world
knowledge to elaborate information explicitly stated in
the text. Such inferences create a richer representation
of the text, but are considered less critical than causal in-
ferences in creating a coherent representation (Bowyer-
Crane and Snowling 2005, Cain et al. 2001, van Kleeck
2008).
To the best of our knowledge, no studies have com-
pared the effects of instruction targeting particular types
of inferences (e.g., causal versus informational) on chil-
dren’s reading comprehension. However, Clarke et al.
(2010) included generation of various types of infer-
ences as a key element of the text comprehension in-
struction they provided to children with poor reading
comprehension (additional elements were metacogni-
tive strategies, narrative structure, and reciprocal teach-
ing). At post-test, the group receiving the instruction
had higher reading comprehension scores than a wait-
listed comparison group.
Inferencing in SLI
Both typically developing children and children with
language impairments have greater difficulty with infer-
ential than literal comprehension, as measured by com-
prehension questions asked after reading. Children with
SLI, in particular, have been found to perform below
expected levels on inferential comprehension on listen-
ing tasks (Adams et al. 2009, Botting and Adams 2005,
Dodwell and Bavin 2008), and on reading tasks, even
when stories contained familiar content (Norbury and
Bishop 2002). Taken together, the results suggest that
the text comprehension difficulties experienced by chil-
dren with SLI cannot be fully explained by difficulties
with decoding or inadequate world knowledge.
There is also some evidence that certain kinds of
inferential questions are more challenging for children
than others. Wright and Newhoff (2001) hypothesized
that children with language-learning disabilities (LLD),2
along with age-matched and language-matched con-
trols, would be more accurate in responding to infor-
mational inference questions than to causal ones, given
that informational questions require the child to elab-
orate a single premise in a text, while causal questions
require that two premises be integrated. Results con-
firmed their prediction. Moreover, similar performance
by the LLD and language-matched groups confirmed
arelationshipbetweeninferencingandlanguagede-
velopment. Likewise, Botting and Adams (2005), and
Dodwell and Bavin (2008), found that children with
SLI performed similarly to language-matched groups
on story comprehension tasks requiring various kinds of
inferences, but scored significantly below age-matched
participants.
Think-aloud procedure
Inferencing has typically been measured by asking chil-
dren comprehension questions after they read or hear
a story. Think-aloud procedures have been proposed
as a complementary means of investigating inferencing
(Phillips 1988). For our purposes, ‘think-alouds’ refer
to verbal reports of thoughts during or immediately
following a task, also referred to, respectively, as concur-
rent or retrospective think-alouds (Yoshida 2008). They
have been used to investigate a wide range of cognitive
processes, including inferences made by children with
average reading skills (Trabasso and Magliano 1996b),
and good and poor reading comprehension (Laing and
Kamhi 2002).
Applied to text comprehension, think-alouds are
usually implemented by asking participants to verbal-
ize their thoughts after reading or hearing each sentence
of a text. Given that instructions typically do not entail
telling participants to focus on the reading or listening
640 Brenna McClintock et al.
process, verbalizations may simply reveal their thoughts
and reflect information that is consciously available to
them.
In addition to making readers’ thoughts public,
think-aloud procedures might influence comprehen-
sion. On the one hand, they might reduce understanding
by increasing cognitive load and by potentially introduc-
ing meta-linguistic demands during reading, thus dis-
rupting the processes readers usually engage in (Yoshida
2008). On the other hand, think-aloud procedures likely
increase the time spent on the reading task. While in-
creased time on a reading task is not always associated
with better comprehension (Yoshida 2008), it might
have positive effects for children.3Simply put, when
asked to complete a think-aloud, children might spend
more time considering a text than they would otherwise,
and the increased time could enhance comprehension.
Additionally, children might spontaneously use a range
of comprehension strategies when thinking-aloud about
a text: inferencing, as well as other strategies critical to
reading comprehension, such as activating prior knowl-
edge and monitoring comprehension (National Institute
for Literacy 2001). The think-aloud does not provide
children with explicit instruction in using these strate-
gies, but most studies have involved brief demonstra-
tions that may provide children with an improved sense
of ‘what it means to read and think about text’ (Kucan
and Beck 1997: 289). Additionally, adult prompting,
although often minimal, is inherent to the procedure
when implemented with children. These characteristics
may encourage children to engage actively with text and
to employ comprehension strategies that are within their
repertoire but not used routinely.
Indeed, there is some evidence that exposure to
think-alouds via direct instruction, modelling, and/or
practice have positive effects on the reading compre-
hension of typically developing children (for reviews,
see Kucan and Beck 1997, and Yoshida 2008). Facil-
itative effects have also been found for children with
below average language or literacy. Laing and colleagues
showed that think-aloud procedures improved the literal
comprehension of expository text of children with SLI
(Laing et al. 2009), and augmented the inferences made
by children with average and below average reading skills
in a narrative context (Laing and Kamhi 2002).
The present study
The present study is the first to examine the effects of a
think-aloud on the literal and inferential understanding
of narrative texts by children with SLI, compared with
age-matched peers with typical language development
(TLD). Children in the SLI and TLD groups were asked
questions to assess text comprehension following two
conditions: a standard ‘uninterrupted’ reading of a text,
and an oral ‘think-aloud’ protocol. In both conditions,
children read the story silently and simultaneously heard
it read aloud. This procedure was implemented in order
to minimize difficulties with decoding that might arise
during independent reading.
Our main hypothesis was that both the SLI and
TLD groups would have higher comprehension scores
after completing the think-aloud procedure, based on
the evidence discussed above (see the section on ‘Think-
aloud procedure’). We were also interested in testing
four additional predictions. First, given the poor syn-
tactic and semantic knowledge exhibited on norm-
referenced language tests administered as part of the
study, we predicted that the SLI group would have
lower scores than the TLD group on the comprehension
questions.
Second, we predicted that scores on inferential ques-
tions (causal and informational) would be lower than
scores on literal questions, and, moreover, that the scores
on causal questions would be lower than scores on in-
formational questions. These predictions were based on
the patterns reported by Wright and Newhoff (2001)
for children with milder deficits in receptive language,
and on Bishop and Donlan’s (2005) finding that chil-
dren with mixed SLI have particular difficulty encoding
causal relationships in narrative.
The last predictions focused on the types of state-
ments made during the think-aloud task and their asso-
ciation with comprehension. Based on previous studies
(see ‘Inferencing in SLI’ above), we anticipated that the
children with SLI would make fewer inferential state-
ments than the group with TLD. If group differences
were found for inferential but not literal statements,
then fewer inferential statements would be taken to re-
flect poorer inferencing rather than solely being a con-
sequence of expressive language difficulties. We also ex-
pected the number of causal statements (versus literal
or informational) to correlate most strongly with scores
on the comprehension questions, given the important
role causal inferences play in reading comprehension
and in narrative comprehension, in particular (Cain
and Oakhill 1999, Laing and Kamhi 2002, van Kleeck
2008).
Methods
Research design
The current study used a mixed analysis of variance
(ANOVA) with group (SLI and TLD) as a between-
subjects factor, and condition (uninterrupted and think-
aloud) and question type (literal, causal inference, in-
formational inference) as within-subjects factors. The
number of literal, informational and causal statements
made during the think-aloud were also compared across
Thinking aloud effects on text comprehension by children with SLI 641
Table 1. Descriptive statistics for language measures, by group
SLI group (n=12) TLD group (n=12)
CELF-4aMean (SD) Mean (SD) t d.f.
Core Language Score 60.92 (7.40) 107.42 (19.37) 7.77∗∗∗ 14.2b
Receptive Language Index 70.25 (5.79) 100.58 (18.12) 7.28∗∗∗ 13.2c
Expressive Language Index 63.83 (6.85) 110.83 (14.84) 9.96∗∗∗ 22
Notes: aStandard scores; all measures have a population mean of 100, SD =15.
bAdjusted degrees of freedom (d.f.) as equal variances are not assumed; Levene’s test F,4.556,p=0.044.
cLevene’s test F,5.089,p=0.034.
∗∗∗p<0.001.
the two groups and analysed in relation to the compre-
hension question scores.
Pilot participants
The procedures were piloted to refine their implemen-
tation and to hone a scoring key for the comprehension
questions. Pilot participants were recruited by word of
mouth. Individual consent was obtained from adults,
and parental consent and child assent was obtained
for children. Twenty adults, all of whom had some
university-level education, participated: eight in a first
phase (mean age =29;8 in years;months, SD =2;1),
and 12 in a second phase (mean age =26;10, SD =
4;8). Two children also participated: one girl aged 7;10
and one boy aged 9;11.
Study participants
Participants were recruited through letters distributed
in schools, word of mouth and flyers posted in public
settings, following ethics approval for the study from the
host university. Prior to beginning the study, consent was
obtained from each child’s parents and verbal assent was
given by each child.
The participants included 12 English-speaking up-
per elementary students (eight boys, four girls: mean
age 10;5) who were regularly receiving speech–language
services. The participants had previously met criteria
for mixed (expressive and receptive) SLI; hearing im-
pairment, neurological disorders and intellectual im-
pairments had been ruled out, and all had a history
of language test scores 1 SD or more below the mean.
As shown in table 1, the participants with SLI scored,
on average, more than 2 SDs below the mean on the
Core Language subtests of the Clinical Evaluation of
Language Fundamentals—4 (CELF-4).
Tw e l v e E n g l i s h - s p e a k i n g s t u d e n t s w i t h T L D a l s o
participated: eight boys and four girls, matched by sex
and age to the SLI group. Their mean age was 10;3
(SD =0;6). There was no significant difference in age
between the two groups: t(22) =–0.562, p=0.580.
TLD status was confirmed by scores of 85 or above
(within 1 SD of the mean) on the Core Language
subtests of the CELF-4 (table 1), and corroborated by
teacher or parent report.
Measures and apparatus
Apparatus
All tests requiring verbal responses from the participants
were audio recorded using a Panasonic RR-US-570 dig-
ital recorder. The story stimuli were pre-recorded into
mp3 files and saved onto an iPod Touch mp3 player. A
portable X-Mini II capsule speaker was used to play the
story recordings while participants read.
Standardized language test
CELF-4 is a standardized, norm-referenced test designed
to assess the oral language abilities of individuals aged
5–21 years (Semel et al. 2003). The Core Language
Score of CELF-4 has been shown to discriminate typ-
ical and language-disordered performance well, and is
also the basis of the test’s sensitivity and specificity, both
of which are high (Semel et al. 2003). This score was
thus selected to confirm SLI or TLD status (table 1).
The score is a composite, derived from scores on recep-
tive subtests (Word Classes—Receptive and Concepts
& Following Directions) and expressive subtests (Word
Classes—Expressive, Recalling Sentences, and Formu-
lated Sentences), which respectively yield Receptive and
Expressive Index scores, also reported in table 1.
Experimental stories and comprehension questions
Think-aloud demonstration stories. The examiner
used two stories (one in each session) to demonstrate
the think-aloud procedure. One (Rachel and the Giant;
51 words) was adapted from Trabasso and Magliano
(1996a), and a second story was written to parallel it:
Ivan and the Dragon; 55 words. The fantastical story
themes contrasted with the realistic themes of the ex-
perimental stories, so as to minimize carryover effects
from the demonstrations to the children’s own think-
alouds. Stories were written at a second- and third-grade-
level respectively, as measured by the Spache Readability
642 Brenna McClintock et al.
Formula, which takes into account the difficulty of
words used and average number of words per sentence
(Spache 1953). Participants read the story from a printed
copy, and simultaneously heard an audio recording of
the story, played over a speaker at a conversational vol-
ume.
Experimental stories. Four stories, two from Bishop
and Adams (1992) and two new stories (see the appendix
for the sample story), were used to measure story com-
prehension and inferencing. Children read each story
from a printed copy and simultaneously heard a record-
ing of the story read aloud. The stories ranged from 108
to 114 words and were written at a second-grade level
(Spache 1953). The stories were written below grade
level to avoid the potential effects of vocabulary knowl-
edge, syntactic complexity or decoding difficulties on
the variables of primary interest. Story themes were se-
lected to maximize the probability that story content
would be familiar to all students.
Comprehension questions. Nine comprehension ques-
tions were written for each of the four stories: three
literal questions, three informational and three causal,
resulting in 36 questions in total (see the appendix for
examples). Literal questions were based on information
that was stated explicitly within the text. Informational
questions required that the participant expand on a sin-
gle proposition in the text. Causal questions required
that the participant connect two propositions, establish-
ing cause-and-effect relationships between them. One of
the 12 literal questions was not recorded due to experi-
menter error, resulting in 35 questions in total.
Procedure
A trained examiner conducted individual testing in a
quiet area of the child’s school or home, using stan-
dardized procedures. The two testing sessions were con-
ducted within the same week, separated, on average, by
3days.Eachsessionlastedapproximately1handwas
audio-recorded for later review. The experimental stories
and measures preceded administration of the CELF-4
subtests in the two sessions.
Tw o s t o r i e s w e r e p r e s e n t e d p e r s e s s i o n . Co n d i t i o n
was counterbalanced, so that each child began one ses-
sion with the uninterrupted reading, and the other with
the think-aloud task. Stories were also counterbalanced
across the uninterrupted and think-aloud conditions so
that each story appeared equally in each condition.
Immediately prior to the think-aloud procedure in
each session, the examiner gave a scripted demonstration
of the task using one of the two demonstration stories
(counterbalanced between sessions and within group).
After the demonstration, children were presented one
of the four experimental stories and told that it was
now their turn to ‘tell what [they] understood about the
story’ after each story sentence. The child was asked to
read the written story as a recording of the story (read
by the examiner) was simultaneously played on a digital
audio player. The examiner followed the words on the
printed copy in front of the child with her finger. After
each sentence she pointed to an image (a light bulb in a
thought bubble) to remind the child of the task. After
the children completed the think-aloud procedure, they
answered the comprehension questions orally. In the
uninterrupted condition, the child read and listened to
the story without interruption and proceeded to the
comprehension questions.
During questioning, the examiner consulted a scor-
ing key devised for the study. The general prompt ‘Can
you tell me a little more?’ was used when participants
gave partially correct answers. Participants were not
prompted if they gave correct or clearly incorrect an-
swers. When a participant did not respond, the prompt
‘Do you want to guess?’ was used.
Coding and scoring
Standardized tests of language. The CELF-4 raw
scores were calculated and converted to standard scores
as per test guidelines, yielding a Core Language Score,
Receptive Language Index and Expressive Language In-
dex.
Comprehension questions. For each question type (lit-
eral, informational and causal), responses were scored
according to a 3-point system adapted from Bishop and
Adams (1992). Two points were awarded for a fully cor-
rect response, 1 point for a response that was correct
but omitted an important detail, and 0 points for an
incorrect response, or no response.
Think-aloud statements. Think-aloud statements
were transcribed and coded. A statement was broadly
defined as a clause that expressed a main idea, and
thus allowed a child to receive credit for both main
clauses and attached ones (i.e., subordinate, condi-
tional and -wh clauses) that provided novel infor-
mation. The transcribed statements were categorized
as either literal statements, which consisted of full
and partial repetitions and paraphrases of the sen-
tence; informational inferences which elaborated a story
proposition (i.e., gave ‘what, who, where, when’ infor-
mation); or causal inferences which expressed cause–
effect relationships that were not explicitly stated and
required the integration of two story propositions. Chil-
dren were given credit for statements related to the sen-
tence they had just read as well as statements related to
earlier sentences.
Thinking aloud effects on text comprehension by children with SLI 643
Table 2. Mean scores (in percentage points) for comprehension questions, by group, condition and question type
SLI group TLD group
Uninterrupted, Think-aloud, Uninterrupted, Think-aloud,
Comprehension questions mean (SD) mean (SD) mean (SD) mean (SD)
Literal questions: percentage points 67.08 (10.87) 76.93 (15.27) 86.67 (12.06) 92.42 (8.80)
Informational questions: percentage points 53.81 (22.45) 61.34 (18.15) 82.38 (18.83) 90.36 (10.20)
Causal questions: percentage points 52.08 (13.82) 68.81 (13.30) 81.25 (10.73) 88.89 (8.21)
Interscorer reliability. The first author scored all com-
prehension question responses (24 participants ×4sto-
ries) and think-aloud statements (24 participants ×2
stories). A trained assistant, blind to group, scored 25%
of the material, selected randomly from a set that equally
represented the groups and the four stories. Interscorer
agreement was 94.6% (kappa =0.91) for comprehen-
sion questions, and 85.4% (kappa =0.72) for think-
aloud statements. Disagreements were resolved via dis-
cussion between the scorers and authors.
Results
Standardized tests of language
Table 1 shows the mean scores by group on the standard-
ized language measure, and independent t-tests com-
paring the means. As was expected, the Core Language
Score, the Receptive Language Index and the Expressive
Language Index of the CELF-4 were significantly lower
in the SLI group than in the TLD group.
Comprehension questions
Table 2 provides mean scores as a function of group,
condition and question type for descriptive purposes.
Comprehension scores are shown as percentages due to
the unequal number of literal questions (as noted in the
‘Methods’ section, one question was not recorded due
to experimenter error). Box plots, presented in figure 1,
illustrate the range and distribution of the scores.
As expected, the children with TLD answered more
questions correctly than did children with SLI. This
observation was confirmed with a mixed 2 (group)
×2 (condition) ×3(questiontype)ANOVA,with
a significant main effect of group, F(1, 22) =56.58,
p<0.001, partial η2=0.720. In accord with our hy-
pothesis, both groups answered more questions correctly
in the think-aloud compared with the uninterrupted
condition. The ANOVA showed a significant main ef-
fect of condition: F(1, 22) =15.48, p=0.001, partial
η2=0.413. Individual scores were in line with the group
pattern. Specifically, in the SLI group, eight out of 12
children scored higher in the think-aloud condition than
in the uninterrupted condition on the literal and causal
questions, and nine scored higher on the informational
questions. In the TLD group, eight of 12 participants
scored higher in the think-aloud on the literal and causal
questions, while six of the participants scored higher on
the informational questions. The group ×condition
interaction was non-significant.
In keeping with our third prediction, the ANOVA
also revealed a main effect of Question Type: F(2, 21) =
7.183, p=0.002. Planned paired t-tests (Bonferroni
adjusted α=0.016; all one-tailed, in accord with hy-
potheses) showed that percentage scores on literal ques-
tions (mean =80.78, SD =12.65) were higher than
scores on informational questions (mean =71.97, SD =
19.68): t(23) =2.84, p=0.002. The mean score for lit-
eral questions was also higher than for causal questions
(mean =72.76, SD =15.14), t(23) =3.31, p=0.001.
There was no significant difference between informa-
tional and causal comprehension scores. The condition
×question type interaction and group ×condition ×
question type interaction were both not significant.
Think-aloud statements
The final predictions were in regard to the statements
made during the think-aloud. First, we expected the
SLI group to make fewer inferential statements than
the TLD group. Findings from independent t-tests sup-
ported this prediction (Bonferroni adjusted α=0.016,
one-tailed). The SLI group made significantly fewer in-
formational statements (mean =5.38, SD =2.54),
compared with the TLD group (mean =9.04, SD =
4.85): t(22) =2.32, p=0.015. The SLI group also made
significantly fewer causal statements (mean =21.38,
SD =6.26), relative to the TLD group (mean =35.08,
SD =14.05): t(15.2) =3.09, p=0.004. The groups
did not differ significantly in terms of the number of
literal statements: SLI mean =6.21 (SD =3.91), TLD
mean =7.77 (SD =3.23).
Table 3 presents correlations between the statements
made during the think-aloud and the comprehension
question scores, also in the think-aloud condition. The
number of causal statements correlated significantly
with both the informational and causal comprehension
scores, but not the literal comprehension scores; the
results are thus partially in line with our prediction that
causal statements would correlate with comprehension.
644 Brenna McClintock et al.
Figure 1. Box plots of comprehension question scores (%), by group, condition and question type. Analyses of variance excluding the outlier
in the SLI group and the most extreme outlier in the TLD group led to the same pattern of results reported in the text; the outliers were thus
retained in the final analyses.
Table 3. Pearson product-moment correlations of think-aloud
statements and language measures with comprehension
questions in the think-aloud condition
Comprehension question scores
Literal Informational Causal
Number of TA statements
Literal statements n.s. n.s. n.s.
Informational statements 0.37∗n.s. n.s.
Causal statements n.s. 0.40∗0.40∗
Language measures (CELF-4)
Receptive Language Index 0.57∗∗ 0.61∗∗ 0.59∗∗
Expressive Language Index 0.57∗∗ 0.74∗∗ 0.72∗∗
Note: ∗p<0.05, one-tailed; and ∗∗p<0.01, one-tailed.
The number of informational statements correlated
significantly with literal comprehension, but not with
performance on informational or causal questions. The
number of literal statements did not correlate signif-
icantly with comprehension question scores. As table
3 also shows, the Receptive and Expressive Language
indices of the CELF-4 were significantly and strongly
correlated with scores on the three comprehension
question types.
Discussion
Effects of a think-aloud procedure
Think-alouds have been proposed as a window into
children’s thinking (Trabasso and Magliano 1996a); the
evidence reported here both supports and extends this
analogy. Here, the think-aloud procedure served as a
‘window’ by making aspects of the children’s think-
ing public. However, engaging in the think-aloud also
seemed to reflect the children’s thoughts back to them
in a way that shaped, and ultimately benefited, compre-
hension. Our data show that children in both groups
(SLI and TLD) had higher scores on comprehension
questions in the think-aloud condition, relative to the
uninterrupted condition. Given the counterbalancing
of conditions and stories, it is unlikely that the observed
Thinking aloud effects on text comprehension by children with SLI 645
pattern reflect practice effects. Rather, we posit that the
think-aloud encourages children to spontaneously use
strategies known to benefit comprehension. Specifically,
when describing what they ‘understood about the story’,
children tended to fill in important but unstated parts
of the text (i.e., generated inferences).
Consistent with previous studies using think-alouds
with children (e.g., Laing and Kamhi 2002, Laing et
al. 2009, Wright and Newhoff 2001), we asked chil-
dren to verbalize their understanding of the story. The
examiner also demonstrated a think-aloud using a train-
ing story, and presented a visual ‘thought-cloud’ to re-
mind the child of the task and focus his or her atten-
tion. These features of the procedure, which can be
viewed as promoting active and sustained ‘meaning-
making’, may have sparked generation of inferences that,
in turn, contributed to higher scores on the comprehen-
sion questions. In addition, the verbalization inherent to
the think-aloud procedure may have transformed infer-
ences from implicit to explicit information, making the
information easier for participants to store and retrieve
from memory when responding to the comprehension
questions.
The positive effects observed in the present investi-
gation converge nicely with a small set of studies show-
ing positive effects of think-aloud protocols on com-
prehension for children with difficulties with language
and/or reading (Laing and Kamhi 2002, Laing et al.
2009, Wright and Newhoff 2001). The study adds to
the literature by demonstrating, for the first time, that a
think-aloud enhanced the comprehension of narratives
of children with SLI, all of whom, moreover, had marked
difficulties with both expressive and receptive language.
Furthermore, the benefits of the think-aloud (versus un-
interrupted reading) extended to an age-matched TLD
comparison group.
Group comparisons (SLI versus TLD) and question
type
Despite the positive effects of the think-aloud, children
with SLI were, as we expected, significantly less accurate
than same-age peers in their literal understanding of nar-
rative texts, and in their responses to questions requiring
inferences. Given that the participants simultaneously
read and listened to the text during testing, and that
the stories were written below grade-level about familiar
or accessible topics, our findings are unlikely uniquely
due to difficulties with decoding, or inadequate world
knowledge. Rather, the results are more likely due to par-
ticipants’ language levels, and their levels of inferencing
during reading, addressed in the next section. The con-
tribution of language is supported by the correlational
data showing a strong and positive relationship between
standardized language measures and scores on the ex-
perimental comprehension questions. The findings are
also in line with previous research demonstrating poorer
text comprehension in children with SLI, regardless of
whether texts are written or oral (Bishop and Adams
1990, Botting and Adams 2005, Dodwell and Bavin
2008, Norbury and Bishop 2002, Wright and Newhoff
2001).
In keeping with our predictions, the literal ques-
tions were nonetheless less difficult than the inferential
questions (informational and causal), for both groups.
Scores on the two inferential question types did not dif-
fer in either group, in contrast with Wright and Newhoff
(2001) who found that scores on informational ques-
tions exceeded those on causal questions for groups with,
and without, language-learning disabilities. The discrep-
ancy in findings might be explained in three ways. First,
the stories in the present study were intentionally writ-
ten with gaps in order to elicit inferences. Wright and
Newhoff’s stories may have had greater coherence as they
were also designed to assess story retelling; the informa-
tional inferences invited by their texts may have been less
challenging, leading to better performance on the infor-
mational questions than on the causal ones. Second, the
LLD group in their study had better receptive skills than
the SLI group in our study, possibly leading to better
overall understanding and subsequently, better infor-
mational inferencing. Thirdly, both informational and
causal inferences rely to some degree on world knowl-
edge. When language impairments are mixed (i.e., re-
ceptive and expressive), such knowledge is likely more
seriously compromised, perhaps rendering the two in-
ference types equally difficult and explaining the lack of
differences for our SLI group. For the TLD group, the
similarity of scores for both informational and causal
questions in our study might have a different explana-
tion; the participants may have simply found the two
question types equally easy, given the stories used here,
though only a minority scored at ceiling.
Think-aloud statements and correlations with
comprehension
The number of causal statements produced by chil-
dren in the think-aloud condition correlated with scores
on both the informational and causal comprehension
questions, while the number of informational state-
ments did not. Perhaps causal statements were associated
with both types of inferential comprehension questions
because they helped children create a coherent story
with a causal chain of events that could subsequently
be supplemented with elaborative information (i.e., in-
formational inferences). At the same time, children in
the SLI group made fewer inferential statements than
the TLD group; this relative weakness in causally con-
necting and elaborating text propositions during reading
646 Brenna McClintock et al.
could help explain their weaker performance on infer-
ential comprehension questions.
No significant group differences were found for the
number of literal statements, nor did literal statements
correlate with comprehension scores. This finding is
broadly consistent with Cain and Oakhill’s (1999) find-
ing that inferencing contributed unique variance to
comprehension whereas measures of literal recall did
not. The lack of correlation of literal statements with
comprehension might also be partly due to their relative
rarity; the results indicated that both groups produced
fewer literal statements than inferential ones, possibly
because they interpreted the think-aloud task as one
in which they were being asked to make sense of and
elaborate the text, rather than reiterate it.
The CELF-4 scores (Core Language, Receptive Lan-
guage and Expressive Language) correlated more con-
sistently and strongly than the number of think-aloud
statements with comprehension question scores. One
plausible interpretation of these findings is that syntac-
tic and semantic abilities (measured by the CELF-4)
have a greater impact on text comprehension than in-
ferencing, at least as measured here by calculating in-
ferential statements. This does not, however, mean that
the think-aloud procedure is not useful, given that in-
tervention might target both basic language skills and
‘higher-level’ language.
A second interpretation is that the number of state-
ments shows only a modest association with compre-
hension scores because it is the reflection entailed in
the think-aloud that matters, rather than the statements
per se. One could test this hypothesis by comparing
the think-aloud procedure to other forms of concurrent
reflection. A third possibility is that the quality of chil-
dren’s inferences might be more strongly correlated with
comprehension than the number of inferences; this is
an interesting issue to explore in future work.
Limitations and additional directions for future
research
Some directions for future research have been alluded to
in our discussion of the correlational data. In the present
study, the inclusion of a younger, language-matched
group might also have provided additional insights into
the relationship between language skill and inferencing,
though including a younger group would have intro-
duced a new challenge of controlling the stories to suit
a wider range of skills. Adjusting the level of the texts
upwards might also have been an informative avenue;
indeed, had children in the present study been presented
with grade level texts and asked to read them indepen-
dently, the gap between children with SLI and TLD may
have been even wider than that observed here.
The design in the present study might also be en-
hanced in future studies by having two think-aloud con-
ditions: one in which participants would hear texts read
aloud, and another in which they would read silently.
In the present study, participants read a text silently as
the examiner read it aloud to minimize comprehension
difficulties that might arise from inadequate decoding.
Children with SLI might plausibly inference less in in-
dependent reading, but we have no reason to believe
that this would render the think-aloud ineffective. If, as
we have suggested, the think-aloud improves children’s
comprehension by prolonging engagement with a text,
stimulating inferencing, and promoting comprehension
monitoring, then it should affect comprehension re-
gardless of text presentation (aural, written, or com-
bination). Nevertheless, this hypothesis awaits further
testing.
Practical implications
The present study confirms that a think-aloud task can
provide insights into the nature of inferences made by
school-age children with TLD and SLI, and can also
positively influence their comprehension of narrative
texts. The feasibility of using the think-aloud to assess
and promote inferencing in instructional or intervention
contexts, particularly for children with SLI, is a promis-
ing direction for future research. There is an ample body
of work demonstrating that direct explanation, mod-
elling, guided practice, reciprocal teaching, and cooper-
ative learning are all fruitful avenues for teaching reading
comprehension strategies in classrooms (National Insti-
tute for Literacy 2001). Think-alouds, in particular, have
been shown to enhance comprehension strategies, with
positive effects on reading comprehension for children
and adolescents (Kucan and Beck 1997). With respect to
children with SLI, a think-aloud procedure might be in-
tegrated to a classroom setting or to small group sessions
held in therapeutic settings. With effective guidance
from a teacher, teaching assistant, or speech–language
therapist, gains in comprehension might be made for a
number of students at once. One might also implement
and study paired think-alouds (involving two children),
or self-directed ones, where children would be taught
the procedure and gradually exercise increasing indepen-
dence in using it to boost their reading comprehension.
Conclusion
In the present study a think-aloud procedure was used
with children with SLI and their age-matched peers
to elicit their interpretation of narrative texts during
reading. This procedure improved comprehension
relative to an uninterrupted condition, as reflected in
children’s responses to literal, informational and causal
Thinking aloud effects on text comprehension by children with SLI 647
comprehension questions. The think-aloud also allowed
examination of the types and number of inferences
children generated. The number of causal inferences
children made correlated significantly with their
responses to both informational and causal comprehen-
sion questions. These findings are striking, given that
causal inferences play a critical role in creating a coherent
mental representation of texts, particularly narratives.
The think-aloud procedure requires further investi-
gation but shows promise as an instructional or inter-
vention strategy to improve literal and inferential com-
prehension in children with language impairments, as
well as their typically developing peers. It is amenable
to implementation by clinicians or educators, and could
be gradually taught to children so that they can take an
active role in ‘reading between the lines’ and in the pro-
cess, enjoy and learn more from the texts they encounter
in and out of school.
Acknowledgements
Declaration of interest: The authors report no conflicts of interest.
The authors alone are responsible for the content and writing of the
paper.
Notes
1. A number of inference taxonomies have been proposed in the
literature, most of which distinguish between inferences that
connect text by filling in gaps or elaborate the text in some way.
2. The participants had greater deficits in expressive language than
receptive; their mean standard score on receptive language tests
was 90.5, within normal limits but lower than both age- and
language-matched peers.
3. Yoshida (2008) uses the term ‘reactivity’ to describe the changes
to cognitive processes that may occur when individuals perform
a task and concurrently verbalize their thoughts.
References
ADAMS,C.,CLARKE, E. and HAYNES, R., 2009, Inference and sen-
tence comprehension in children with specific or pragmatic
language impairments. International Journal of Language and
Communication Disorders,44,301–318.
BISHOP, D. V. M. and ADAMS, C., 1990, A prospective study of the
relationship between specific language impairment, phono-
logical disorders, and reading retardation. Journal of Child
Psychology and Psychiatry,31,1027–1050.
BISHOP, D. V. M. and ADAMS, C., 1992, Comprehension problems
in children with specific language impairment: literal and
inferential meaning. Journal of Speech and Hearing Research,
35,119–129.
BISHOP, D. V. M. and DONLAN, C., 2005, The role of syntax in
encoding and recall of pictorial narratives: evidence from spe-
cific language impairment. British Journal of Developmental
Psychology,23,25–46.
BOTTING, N. and ADAMS, C., 2005, Semantic and inferencing abil-
ities in children with communication disorders. International
Journal of Language and Communication Disorders,40,49–
66.
BOWYE R-CRANE, C. and SNOWLING, M. J., 2005, Assessing children’s
inference generation: what do tests of reading comprehension
measure? British Journal of Educational Psychology,75, 189–
201.
CAIN, K. and OAKHILL, J., 1999, Inference making ability and its
relation to comprehension failure in young children. Reading
and Writing,11,489–503.
CAIN,K.,OAKHILL,J.V.,BARNES, M. A. and BRYANT,P.E.,
2001, Comprehension skill, inference-making ability, and
their relation to knowledge. Memory and Cognition,29,850–
859.
CATT S, H. W., BRIDGES,M.S.,LITTLE, T. D. and TOMBLIN,J.
B., 2008, Reading achievement growth in children with lan-
guage impairments. Journal of Speech, Language, and Hearing
Research,51,1569–1579.
CATT S, H. W., FEY,M.E.,TOMBLIN, J. B. and ZHANG, X., 2002,
A longitudinal investigation of reading outcomes in children
with language impairments. Journal of Speech, Language, and
Hearing Research,45,1142–1157.
CLARKE,P.J.,SNOWLING,M.J.,TRUELOVE, E. and HULME,C.,2010,
Ameliorating children’s reading-comprehension difficulties: a
randomized controlled trial. Psychological Science,21,1106–
1116.
DODWELL, K. and BAVI N, E. L., 2008, Children with specific lan-
guage impairment: an investigation of their narratives and
memory. International Journal of Language and Communica-
tion Disorders,43,201–218.
GOUGH, P. B. and TUNMER, W. E., 1986, Decoding, reading, and
reading disability. Remedial and Special Education,7,6–10.
HOOVER, W. A. and GOUGH,P.B.,1990,Thesimpleviewofreading.
Reading and Writing,2,127–160.
KELSO,K.,FLETCHER, J. and LEE, P., 2007, Reading comprehension
in children with specific language impairment: an examina-
tion of two subgroups. International Journal of Language and
Communication Disorders,42,39–57.
KINTSCH, W., 1988, The role of knowledge in discourse comprehen-
sion: a construction–integration model. Psychological Review,
95,163–182.
KUCAN, L. and BECK,I.L.,1997,Thinkingaloudandread-
ing comprehension research: inquiry, instruction, and so-
cial interaction. Review of Educational Research,67,271–
299.
LAING,S.P.,GILLAM,S.,FARGO, J. D. and ST.CLAIR ROBERTSON,
K., 2009, Comprehension of expository text: insights gained
from think-aloud data. American Journal of Speech Language
Patholog y,18,82–94.
LAING, S. P. and KAMHI,A.G.,2002,Theuseofthink-aloudpro-
tocols to compare the inferencing abilities in average and
below-average readers. Journal of Learning Disabilities,6,436–
447.
NATI ON ,K.,CLARKE,P.,MARSHALL, C. M. and DURAND,M.,2004,
Hidden language impairments in children: parallels between
poor reading comprehension and specific language impair-
ment? Journal of Speech, Language, and Hearing Research,47,
199–211.
NATIONAL CENTER FOR EDUCATION STATISTICS,2011,The Nation’s
Report Card: Reading 2011. NCES 2012–457 (Washington,
DC: Department of Education).
NATIONAL INSTITUTE FOR LITERACY, 2001, Put Reading First: The
Research Building Blocks for Teaching Children To Read (avail-
able at: http://www.nationalreadingpanel.org//Publications
/publications.htm) (accessed on 5 January 2013).
NORBURY, C. F. and BISHOP, D. V., 2002, Inferential processing
and story recall in children with communication problems:
a comparison of specific language impairment, pragmatic
648 Brenna McClintock et al.
language impairment and high-functioning autism. Interna-
tional Journal of Language and Communication Disorders,37,
227–251.
OAKHILL, J. and CAIN, K., 2012, The precursors of reading ability
in young readers: evidence from a four-year long longitudinal
study. Scientific Studies of Reading,61,91–121.
PHILLIPS, L. M., 1988, Young readers’ inference strategies in
reading comprehension. Cognition and Instruction,5, 193–
222.
RICE, M. L., 2009, Children with specific language impairment:
bridging the genetic and developmental perspectives. In E.
Hoff and M. Shatz (eds), Blackwell Handbook of Language
Development (Malden, MA: Blackwell), pp. 411–431.
SEMEL,E.,WIIG, E. H. and SECORD,W.A.,2003,Clinical Evaluation
of Language Fundamentals, Fourth Edition, CELF-4 (Toronto,
ON: Pearson Assessment Canada).
SPACHE , G., 1953, A new readability formula for primary-grade
reading materials. Elementary School Journal,55,410–
413.
TRABASSO, T. and MAGLIANO, J. P., 1996a, Conscious understand-
ing during comprehension. Discourse Processes,21,255–
287.
TRABASSO, T. and MAGLIANO, J. P., 1996b, How do children under-
stand what they read and what can we do to help them? In
M. F. Graves, P. van den Broek and B. M. Taylor (eds), The
First r: Every Child’s Right to Read (New York, NY: Teachers
College Press), pp. 160–188.
VAN KLEECK, A., 2008, Providing preschool foundations for later
reading comprehension: the importance of and ideas for tar-
geting inferencing in storybook-sharing interventions. Psy-
chology in the Schools,45,627–643.
WRIGHT, H. H. and NEWHOFF, M., 2001, Narration abilities of
children with language-learning disabilities in response to
oral and written stimuli. American Journal of Speech–Language
Patholog y,10,308–319.
YOSHIDA,M.,2008,Think-aloudprotocolsandtypeofreadingtask:
the issue of reactivity in L2 reading research. In M. Bowles, R.
Foote, S. Perpi˜
n´
an and R. Bhatt (eds), Selected Proceedings of
the 2007 Second Language Research Forum (Somerville, MA:
Cascadilla Proceedings Project), pp. 199–209.
Appendix: Sample story with comprehension
questions
Anna and her Dog
Anna was walking her dog. They stopped by the swings
to play. Anna’s dog was very good at playing fetch. Anna
threw the ball over and over again. She threw the ball
too far. It went into the woods. Anna waited but her dog
didn’t come back. Anna looked by the water fountain.
Her dog was not there. She looked by the baseball field.
The baseball players had not seen her dog. Anna looked
until it started to get dark. Finally, she started to walk
home. She passed by the swings on her way. Her dog
was waiting for her with the ball. Anna ran and gave her
dog a hug.
Literal questions
!Where did Anna and her dog stop to play with
the ball?
!What game did Anna play with her dog?
!Where did Anna look for her dog?
Causal questions
!Why didn’t Anna’s dog come back?
!Why did Anna’s dog wait for her by the swings?
!How did Anna feel when she found her dog?
Informational questions
!Where did Anna take her dog for a walk?
!Who did Anna ask if they had seen her dog?
!What time of day was it when Anna’s dog got lost?
Copyright of International Journal of Language & Communication Disorders is the property
of Wiley-Blackwell and its content may not be copied or emailed to multiple sites or posted to
a listserv without the copyright holder's express written permission. However, users may
print, download, or email articles for individual use.
