Narrative retells in Swedish school-aged children – a clinical pilot study

Abstract

Aim

The aim of this pilot study was to evaluate two new retelling tasks intended for clinical use in terms of language sample size, effects of picture support, and order of presentation.

Methods

Forty Swedish-speaking children in grades 4–6 participated in the study, 31 children with typical language development (TLD, mean age 11;1), and nine children with developmental language disorder (DLD, mean age 11;5). Two oral retells, one with and one without picture support, were analyzed with regards to productivity, syntactic complexity, basic Story Grammar (SG) units, and Internal Responses (IR).

Results

Results showed no systematic order effects in the TLD group, although this needs to be investigated further, and good inter-rater reliability. Both tasks elicited sufficiently large language samples, except from one participant with DLD whose samples were excluded from subsequent comparisons. When appropriate, data were analyzed with ANOVA (productivity, mean length of C-unit/MLCU), otherwise t-tests (TLD-group) or non-parametric tests (DLD-group) were used. As expected, retells from participants with DLD were shorter, with shorter MLCU and fewer SG units compared to the TLD group. There were also task effects: in the task with picture support, all participants had longer MLCU, and participants with TLD also showed a higher proportion of subordinate clauses, indicating that pictures may function as a support for syntactic complexity. The task without picture support, on the other hand, elicited more C-units indicated by a significant main effect, and more SG units, which was a significant effect in the TLD group.

Conclusions

>We conclude that both tasks might be useful for Swedish speech-language pathologists.

Keywords:

• Narrative
• developmental language disorder
• task effects
• school-age
• assessment

Introduction

Narrative skill can be defined as the ability to use language to describe experiences or tell stories in a comprehensible way, and establishing narrative skills is important both from an academic and a social perspective [1]. In school, students are expected to process and produce real and fictive narratives, and narration is also used in everyday interactions to establish and maintain relationships with peers and is important for the development of identity and social-emotional growth [1–3].

Narration requires the simultaneous use of different skills; the speaker needs to choose and structure the content, use grammatical and vocabulary knowledge and consistently adjust the story to the listener [1,4]. Thus, narrative tasks place high cognitive processing demands on all speakers, but may be especially challenging for children and adolescents with developmental language disorder (DLD) [4–6], who struggle with learning their native language(s) despite normal IQ and absence of frank neurological damage, genetic disorders or hearing impairment [7]. In addition, narrative tasks elicit more complex language than conversation, both in children with typical language development (TLD) and children with DLD [e.g. 8]. Taken together, analysis of narrative language is an important tool for clinical speech-language pathologists (SLPs). In Sweden, there is a lack of standardized procedures for eliciting and analyzing narratives for children above the age of nine, even though clinical guidelines recommend that an assessment of oral narrative skills should always be included in the school-age reading, writing, and language assessments [9].

The present study investigates two newly created retelling tasks intended to be used by SLPs who work with children in Swedish grades 4–9 (ages 9–15), one presented with picture support and one without. As a first step towards confirming the clinical value of these tasks, narrative language samples from children with TLD in grades 4–6 are compared to the samples from a smaller group of age-matched children with DLD. We investigate whether each task elicits an acceptable number of utterances, task effects, group differences, and potential effects of order of presentation in the TLD group.

As previously mentioned, narration is especially challenging for children with DLD, who often do not develop narrative skills as expected [6]. Narrative development continues during the school years, and the ability to structure content develops in tandem with the development of grammar, vocabulary, and pragmatic skills [10–12]. Children with TLD are often able to organize the story as a cohesive whole at the age of 10 [13] and can describe the characters'immediate feelings or reactions [12]. Twelve-year-olds with TLD can interpret and describe the characters' emotions in an elaborated way, and explain the cause of the emotions [14]. Ukrainetz et al. [12] showed that 12-year-olds and 9-year-olds produced narratives of similar length, but that 12-year-olds included more dialogue and a higher frequency of words describing feelings.

One way to quantify narrative structure and content is to use Story Grammar (SG), a framework for analyzing narrative macrostructure. SG consists of several specified content units that appear in many fictional stories, and need to occur in a logical order [15]. Within goal-based SG frameworks, a plan-based sequence (e.g. a plan-action-complication sequence) is central to each episode, and more complex stories combine several episodes and elaborations [3,14]. The setting (time and place and character/s), an initiating event, and a story ending are also essential SG units, as well as the feelings of the characters (Internal Responses/IR), although the emotions do not have to be explicitly stated for an episode or a story to be regarded as complete [15].

Children with DLD have been shown to struggle both in terms of macrostructural measures (including e.g. Story Grammar), and microstructural measures (e.g. utterance length and grammatical complexity). Studies have for example shown that children with DLD include less content and produce shorter and less complex utterances in their narratives [8,16–18]. In addition, children with DLD have been shown to use less variation in words related to emotions (Internal Responses/IR) in narrative tasks compared to children with TLD [19], and others have found that the number of IR is significantly correlated with linguistic complexity in children with TLD and DLD [18]. In one of the few studies including older Swedish-speaking children with DLD, Reuterskiöld, Hansson, and Sahlén [5] found that the only group differences in picture elicited narrative language samples were the number of different verbs and percent grammatically correct C-units. The number of SG units, number of words, and grammatical complexity (measured in the proportion of connectives) did not differ between groups.

There are many ways to elicit narratives. In Sweden, a retelling task is the most common task type used in clinical practice, although currently, no standardized tasks or procedures for administration and scoring exist for older children. Clinically, story retell has the advantage of control over language content and is a relatively efficient way of eliciting a language sample that can be analyzed both regarding microstructural and macrostructural measures, compared to story generation tasks or conversational protocols [8,20,21]. Gazella and Stockmann [20] also argued that a retelling task is the best elicitation option for a language sample used for diagnostic screening since it is possible to control not only content but also story length and grammatical complexity, which is not possible in story generation tasks.

Historically, a minimum of 50 utterances has been recommended for language sample analysis, but Heilmann, Nockerts, and Miller [22] found that a narrative language sample of 3 min had acceptable stability compared to a 7-min sample regarding several measures, including mean length of utterance (MLU), the number of different words and percent of speech disruptions. Their 3-min narrative language samples for 133 children with TLD (6;0–13;3 years old) contained on average 16 utterances with a standard deviation of 4.0, where an utterance was defined as the main clause with all its dependent clauses (a C-unit). Based on their study, a minimum of 12 C-units will be considered a clinically useful narrative language sample in the current study (i.e. within one standard deviation of the mean in the data from [22]).

Retelling tasks can be presented with or without the presence of pictures. Previous studies on the effect of picture support on narrative performance have shown somewhat contradictory results. Westerveld and Vidler [23] compared retelling tasks with and without picture support in children with TLD (6;0–7;11 years old) and reported that picture support generated more information measured in significantly more utterances and unique words. The same study did not find any significant task effects on syntactic complexity, while other studies suggest that pictures may function as cognitive support, thus helping the speaker to attain more complex language [24,25]. Masterson and Kamhi [25] found that retells with picture support elicited more complex syntax than an expository task without picture support in children with TLD (6–9 years old). Bergman and Hallin [24], who used the same tasks as in the present study with 12–16-year-olds with and without ADHD, found that the task with picture support elicited fewer SG units in both groups, but more complex syntax in the TLD group.

The possible supporting effect of pictures on microstructural measures is especially relevant for younger children at earlier stages of language development, as well as children with DLD, who often have impaired verbal short-term and working memory [26]. However, previous research has generally failed to show that picture support enhances the amount of content or language complexity in these groups [20,27–30].

Schneider and Dubé [31] showed that a group of younger children with TLD (M = 5;58 years) introduced fewer characters in a task with picture support, while this effect was not seen in the group of older children (M = 7;81 years). In a subsequent study, they once more showed that fewer characters were introduced in both the younger and the older age group in a task with picture support, even though the characters were included in the pictures [27]. In both studies, a naïve listener, who could not see the pictures during the retelling tasks, was used. They also analyzed the number of SG units, which was somewhat higher in the task with picture support on average, but not to a significant level for any of the groups. Two other studies including younger children with TLD (4–7 years old) did not find any effects of picture support in terms of productivity, content, or complexity [20,28]. For 6–9-year-olds with DLD, dyslexia, and TLD, Masterson and Kamhi [29] found that language samples with referents (pictures and physical objects) present, generated significantly fewer complex sentences and a higher frequency of speech disruptions compared to language samples without referents present. Schneider [30] found that children with DLD (5;7–9;9 years old) included fewer SG units in retellings with picture support compared to retelling without picture support. No significant task effects were found regarding the number of words, MLCU, or number of C-units. Schneider [30] discussed whether young TLD children and children with DLD might exclude content visible in the pictures when retelling with picture support, or that the planning of the story is interfered with by the need to divide attention between the pictures and the retelling.

Important to note is that the type of visual support, as well as task, can play a role. Shapiro and Hudson [32] concluded that picture sequences with embedded problem sequences elicited more complex language and narratives than picture sequences that were action-based, in young school-age children with TLD. Thus, it is possible that the picture material, as well as the elicitation methods used in the studies above, did not support more complex storytelling.

In summary, there are mixed results regarding the effects of picture support on narrative ability, and even though picture support may function as cognitive support in children with TLD [24,25], it may also lead speakers to include less information or interfere with the planning and execution of the narrative [29–31], or make no difference at all [20,28].

The present study investigates narrative language samples from two carefully designed and matched retelling tasks, one with and one without picture support in 9- to 12-year-old children with TLD or DLD and focuses on the following research questions to inform clinical practice:

• Are there any significant order effects when performing both tasks in the same session?

• Do both tasks elicit a sufficient narrative language sample for children with TLD or DLD in this age group (12 complete and intelligible C-units or more)?

• Are there any effects of picture support or group (TLD/DLD) in terms of productivity, grammatical complexity, number of basic Story Grammar units (setting, initiating event, plan-action-complication sequences, story ending), and number of Internal Responses?

Materials and methods

This study is part of a larger research project headed by the third author (Expository Language and Oral Retelling in Swedish Adolescents: EXPLORE-SA) which is approved by the Stockholm Regional Ethical Review Board (#2017/49-31/4). Participation in the study was voluntary, and participants and caregivers could withdraw their consent at any time without explanation. No compensation was given to participants recruited in the early stage of the study. Participants recruited after some external funding was secured were compensated with a cinema ticket (and the necessary ethical review update to allow this was made).

Participants

Forty children in grades 4–6 were recruited for the study through school invitations and social media posts: 31 children with typical language development (TLD), and nine children with diagnosed developmental language disorder (DLD).

The TLD group consisted of 11 boys and 20 girls aged 9;10–12;10 (M = 11;1). Inclusion criteria were parent-reported typical speech and language development, that the child had no previous contact with a speech-language pathologist (SLP) or psychologist, no hearing impairment (hearing is routinely checked in Swedish schools), and no current or previous special accommodations in school. All children were native speakers of Swedish, born in Sweden, and had at least one Swedish-speaking parent, and five children spoke an additional language at home. The participants were quasi-randomized into two groups: one group completed the task with picture support first (group 1, n = 16), and the other group the task without picture support first (group 2, n = 15). Data from five of the oldest children in group 1 were also included as a comparison group in Bergman and Hallin [24].

The DLD group consisted of five boys and four girls aged 10;7–12;10 (M = 11;5). All had been diagnosed by an SLP with expressive or mixed (expressive-receptive) developmental language disorder within a year before participating. One of the participants had an additional diagnosis of ADHD. All children were native speakers of Swedish, born in Sweden, and had at least one Swedish-speaking parent, and two children spoke an additional language at home. Eight of the participants attended a language unit (with special accommodations for children with DLD) integrated within a mainstream school, and teachers reported that four of the participants had worked on narrative skills as part of the curriculum. Participants in the DLD group all received the task with picture support first. One participant with DLD had to be excluded from all analyzes since the transcribed material was too limited to analyze (a few utterances in the task with picture support, no utterances in the task without picture support). Thus, retells from eight participants with DLD were included in the subsequent analyzes (boys n = 5, girls n = 3, mean age 11;3).

All participants were tested with TROG-2 (Test for Reception of Grammar 2, Swedish version [33]) and Recalling sentences from CELF-4 (Clinical Evaluation of Language Fundamentals, Swedish version [34]).

Participants with TLD in group 1 (n = 16) and group 2 (n = 15) were compared regarding age and test results. Independent t-test showed no significant differences between the groups (age p = .222, TROG-2 p = .300, Recalling sentences p = .261). The TLD (n = 31) and DLD (n = 8) groups were also compared, and as expected, there were no significant group difference in age (p = .74), but the TLD group had significantly higher results on TROG-2 (percentiles) TLD M = 54.29 (SD 20.16), DLD M = 19.14 (SD 20.88), t₍₃₇₎ = 4.28, p < .001, and Recalling sentences (scaled score) TLD M = 10.06 (SD 2.24), DLD M = 4 (SD 2.27), t₍₃₇₎ = 6.82, p < .001.

Materials

The retelling tasks consisted of two recorded stories with three parts, one task always presented with picture support and one always presented without. The three-part structure (with three corresponding pictures to each part for the task with picture support) was inspired by the material Multilingual Assessment Instrument for Narratives (MAIN) [35], intended for pre-school and early elementary school children. This multi-part structure provides more than one opportunity to include the central Story Grammar units. The two stories were carefully matched on number of words (487), number of C-units (41), speaking pace (length of each recording: 4 min), mean length of C-unit/MLCU (11.88), the average number of clauses per C-unit (Subordination Index/SI: 1.66/1.68) number of characters (3), number of plan-action-complication sequences (5), total number of Story Grammar (SG) units (25), and number of explicitly stated Internal Responses (IR) (9). The SG units included in both stories follow the terminology and structure of the SG units in the narrative intervention material SKILL, for elaborated stories with more than one plan-action-complication sequence [36]. In both stories, Internal Responses were considered and scored as a separate content category, since the emotional responses of characters do not have to be explicitly stated for a story to be complete [15], and IR is something that both younger children and children with language disorders may have particular difficulty with [e.g. 14,18]. Figure 1 shows the basic SG units and all IR that were included and scored for both retell tasks.

Figure 1. Overview of the three parts, basic Story Grammar units (bullets, maximum score 25) and Internal Responses (hearts, maximum score 9) that were included in both narrative retell tasks.

Each story included five complete plan-action-complication sequences, was about an everyday event and was created to be relatable for adolescents living in Sweden regardless of age and cultural background. The narrative with picture support was about a girl waking up to her alarm the day of school photography (initiating event) and deciding to wear her favorite jeans. First, she could not find the jeans (complication 1), then her brother spilled milk all over them (complication 2), and then she could not find any other pants she wanted to wear (complication 3). When she tried to take the stains out, her jeans got even bigger wet stains (complication 4), and then her dad did not allow her to go to school with wet pants (complication 5). The wrap-up was that she happily left for school in a pair of other pants with her still wet jeans in a bag. The picture support consisted of three black and white drawings per part, that explicitly depicted the problem sequence [32].

The narrative without picture support was about a girl finishing her last class for the day (initiating event) and deciding to go to the mall to meet a friend. First, the bus to the mall did not show up (complication 1), then she could not find her cellphone that was ringing (complication 2), and then her battery died (complication 3). Then she saw her mom who did not know she was in the mall (complication 4), and then she fell when she was trying to hide (complication 5). The wrap-up was that her friend came by, and they went to get ice cream.

Both narratives were recorded by a female speech-language pathologist in a studio with the software Phog (2.500.1003; Neovius Data and Signal System AB, Lidingö, Sweden) onto a recorder at a sampling rate of 16,500 Hz, normalized to 65 dB.

Procedure

The participants met a trained SLP student or an SLP individually, in the participant's school, home, or at an SLP clinic, and completed the tasks in one session (∼60 min). The testing started with one of the retelling tasks, followed by the language and memory measures, and finally the second retelling task. The tasks were administered with a laptop (in Microsoft PowerPoint^©) through headphones, and the retells were recorded with a Tascam recorder DR22-WL.

In the task with picture support, the participants were instructed to choose one of three stories, review the pictures from the chosen story, listen to the story while the same pictures would be presented on the screen again, and retell the story to the test leader with as many details as possible while looking at the pictures a third time. Then, the test leader started the digital presentation and sat so that s/he could not see the screen. The numbers 1–3 appeared on the screen, and the participant chose a story by clicking on a number with a mouse click. The participants were unaware that regardless of what number they chose, the same story was presented. This method was used to elicit a more detailed retelling, as participants regarded the test leader to be a naïve listener unaware of what story they had chosen, and to avoid eliciting a simpler story because of shared and seen information (see [37], for a similar procedure and [30] for a discussion). After the participant had selected a number s/he reviewed the nine pictures at his/her own pace (three pictures/one part at a time), and then s/he listened to the story. For the duration of each part, the three corresponding pictures were shown again on the screen. After the audio recording, the participant got a reminder on the screen to retell the story, and the pictures corresponding to the first part were shown again on the screen. The participant clicked with the mouse to move on to the next three pictures when s/he had finished the retelling of one part.

The retelling task without picture support followed the same procedure, except that there was no picture series to review before the audio recording, and the recorded story and retelling were presented without any picture support, only a blank screen and the written reminder to start retelling after the recorded narrative had finished.

The examiner was careful to give only minimal prompting and feedback (e.g. listening actively, looking interested, nodding, and occasional positive humming) during the retelling of both tasks. If a participant did not start the retelling, the examiner would wait for 10 s and give one of the following prompts: “Now it’s your turn to tell the story”/“How would you like to start your story?”/“Do you remember what happened initially in the story?” If the participant did not start the retelling after the first prompt, the examiner would wait for 10 s and give another of the prompts above. This was carried out a maximum of three times, and if the participant had not yet started, the task was aborted. If a participant forgot the name of a character and wanted help, the examiner responded that the exact names did not matter and that it was ok to use any name (for scoring of the SG unit “Who,” see below). If a participant ended the retell abruptly, the examiner would ask “is the story finished/can you tell me something more that happened?” to allow the participant to add more to their narrative.

Transcription and coding

The transcribed recordings were analyzed in the software SALT (Systematic Analysis of Language Transcripts, Research version 18, SALT software LLC). This software, with versions for both researchers and clinicians, can automatically provide measures, such as MLU and grammatical complexity from an orthographic transcription, given that it is segmented and coded according to the guidelines. For this project, the published SALT guidelines [38] together with a Swedish adaptation of transcription and coding rules [39] were used. The language samples from each participant were transcribed orthographically, segmented into C-units, and coded by the transcribers. Examples of SALT coding relevant to this project is that unintelligible speech that the transcriber did not comprehend after listening three times were marked with “X” and excluded from all following analyzes, that speech disruptions were put in parentheses and excluded from the word count and complexity measures, and abandoned utterances and parenthetical remarks were coded and excluded from all measures. The transcribers were blinded for each participant’s age, sex, and order of presentation of the tasks. The following measures from each transcript were then obtained through SALT: a total number of words, complete and intelligible C-units, mean length of C-unit (MLCU), and subordination index (SI: average number of clauses/C-unit).

The coding of the number of Story Grammar units (SG units) and the number of Internal Responses (IR) was developed by the first two authors. One point was given for each included SG unit and IR (see Figure 1). To get an SG unit point for “Who,” the participant had to state the character’s name or label (e.g. “the girl/boy”/”brother”/“mom”/“friend”). Points were also given if a participant used other names than those presented in the retell tasks, as long as they were used consistently. For the main character,” she” was also accepted. For all other characters, no points were given if only “he/she” was used and no other introduction was included. For example, the sentence “Maya met her friend, he was…” was scored with points both for Maya and the friend, but “Maya went to the mall, then he came…” only generated points for Maya. One point was given for each introduced character, with a maximum of three points per story. To get an SG unit point for “Plan”, the participant had to use a metacognitive verb (e.g. “think,” “plan,” “decide,” “get an idea,” “want”).

Points for IR were only given when the participant described the main character's emotional state/internal response, not for other characters’ (e.g. in the utterance “Maya felt ashamed as her mother angrily yelled at her”, points were given for “felt ashamed” but not for “angrily”). It is worth noting here that there was only one instance per story where other characters showed any emotions (the mother in the story without picture support and the father in the story with picture support).

The ending of one recording was missing for one participant with TLD due to technical issues, therefore this participant was excluded from the SG unit and IR analyzes. Thus, the recordings from 30 TLD participants were analyzed with regard to SG units and IR (boys n = 11, girls n = 19), and all 31 participants were analyzed in terms of the microstructural measures.

Statistical analysis

Statistical analyses were made in IBM SPSS Statistics for Macintosh, Version 27.0. Possible order effects in the TLD group were investigated with paired-sample t-tests, with a significance level of p ≤ .05. For the effects of picture support and group, a two-way ANOVA was performed for the variables where assumptions of normality and homogeneity were met (namely number of C-units, number of words, and MLCU), also with a significance level of p ≤ .05. For SI, SG units, and IR, potential effects of picture support were explored with paired-samples t-test for the TLD group, and for the DLD group, the corresponding non-parametric test was used (Wilcoxon). Potential group differences for these variables were investigated with Mann–Whitney U-tests. For these three variables, the significance level was set to p ≤ .0125 (.05/4 statistical tests per dependent variable) to correct for multiple comparisons and avoid type 1 errors. Since the statistical power is reduced due to the small sample size and Bonferroni corrections, all p-values and effect sizes (Cohen’s d) are presented for both statistically significant and non-significant comparisons. For significant effects in the ANOVA analyses, η² is reported.

Reliability

Inter-rater reliability was calculated for all variables. For the microstructural variables, both retelling tasks from eight randomly selected participants (four from the TLD group and four from the DLD group) were transcribed and coded by both transcribers, and then the measures were obtained from SALT were compared. All 16 transcripts were also coded for macrostructural variables by both transcribers and then compared. For the transcriptions from both the TLD and the DLD group, inter-rater reliability was at least 93.0% or higher for all microstructural variables, which is considered high reliability for transcribing/coding in SALT [40]. For SG units and IR, inter-rater reliability was 97.4% or higher across both groups.

Results

Order effects

Comparing all measures between the two groups of TLD participants with different task order presentation, no systematic order effects were found (see Table 1). For the task with picture support, there were no significant differences between the groups based on order of presentation, with moderate effect sizes: #C-units: t₍₂₉₎ = −0.89, p = .38, d = 0.32, #Words: t₍₂₉₎ = −0.45, p = .65, d = 0.16, MLCU: t₍₂₉₎ = 1.07, p = .29, d = 0.39, SI: t₍₂₉₎ = 0.97, p = .34, d = 0.35, SG units: t₍₂₈₎ = −1.33, p = .19, d = 0.49, and IR: t₍₂₈₎ = 1.33, p = .19, d = 0.49. For the task without picture support, there were no significant differences and moderate effect sizes five of six variables #C-units: t₍₂₉₎ = 1.01, p = .32, d = 0.36, #Words: t₍₂₉₎ = 1.37, p = .18, d = 0.49, MLCU: t₍₂₉₎ = 1.61, p = .23, d = 0.58, SG units: t₍₂₈₎ = −1.12, p = .27, d = 0.41, and IR: t₍₂₈₎ = 0.52, p = .61, d = 0.19. Subordination Index (SI) had a large effect size, however, and the difference was statistically significant, with higher SI for the group with this task presented last, t₍₂₉₎ = 2.63, p = .01, d = 0.95. After ensuring that the conclusion of the paper did not change (for SI or any other variable) when only participants with one order of presentation were included in the analyses, the whole TLD group was combined into one group for the rest of the results section.

Table 1. Means and standard deviations for the groups of TLD children with different order of presentation, N = 31).

	W: With picture support, M (SD)		WO: Without picture support, M (SD)
	1: W first (n = 16)	2: W last (n = 15)	1: WO last (n = 16)	2: WO first (n = 15)
Total number of C-units (#C-units)	25.81 (7.38)	27.73 (4.06)	31.06 (6.53)	28.60 (7.06)
Total number of words (#Words)	284.81 (78.73)	295.13 (40.42)	301.38 (85.35)	262.07 (73.62)
Mean length of C-unit (MLCU)	11.57 (1.86)	10.96 (1.20)	9.83 (1.22)	9.19 (0.99)
Subordination index (SI)	1.69 (0.27)	1.61 (0.17)	1.54 (0.14)	1.41 (0.13)
Story grammar units (SG units)	18.67 (2.74)*	19.80 (1.82)	22.40 (2.53)*	21.47 (2.00)
Internal responses (IR)	3.20 (1.97)*	2.40 (1.24)	2.07 (1.53)*	1.80 (1.26)

*n = 15.

Number of complete and intelligible C-units

All TLD participants met the benchmark of at least 12 C-units in both tasks (range 14–49). Except for the excluded participant, all participants with DLD met the benchmark of at least 12 C-units in the task with picture support (range 15–33), and all participants but one in the task without picture support (range 11–38).

Effects of task and group

Table 2 shows all groups by task means and standard deviations.

Table 2. Descriptive data for all variables for children with TLD (n = 31) and children with DLD (n = 8).

	W: With picture support, M (SD)		WO: Without picture support, M (SD)
	TLD	DLD	TLD	DLD
#C-units	26.74 (5.99)	22.00 (7.03)	29.87 (6.80)	24.75 (8.80)
#Words	289.81 (62.36)	186.25 (43.21)	282.35 (81.06)	192.50 (66.61)
MLCU	11.28 (1.58)	9.39 (1.76)	9.52 (1.14)	8.37 (1.54)
SI	1.65 (0.23)	1.44 (0.16)	1.47 (0.15)	1.32 (0.19)
*SG units	19.23 (2.36)	15.38 (2.13)	21.93 (2.29)	19.0 (3.67)
*IR	2.80 (1.67)	0.88 (0.99)	1.93 (1.39)	0.75 (0.89)

*TLD n = 30.

Total number of C-units

An ANOVA showed a significant main effect of task on the total number of C-units, F_(1,37) = 4.356, p = .044, η_p² = .105, with more C-units in the task without picture support on average (M = 28.82) than the task with picture support (M = 25.77). There was also a significant main effect of group, F_(1,37) = 4.734, p = .036, η_p² = .113, where the TLD group had longer retells on average (M = 28.31) than the DLD group (M = 23.38). The interaction between group and task was not significant (p = .894, η_p² = .000).

Total number of words

An ANOVA showed a significant main effect of group on total number of words, F_(1,37) = 15.09, p < .001, η_p² = .290, where the TLD group had longer retells on average (M = 286.08) than the DLD group (M = 189.37). There was no significant main effect of task (p = .960, η_p² = .000) and no significant interaction (p = .567, η_p² = .009).

Mean length of C-unit

An ANOVA showed a significant main effect of group on MLCU, F_(1,37) = 10.28, p = .003, η_p² = .217, where the TLD group on average had a higher MLCU (M = 10.40) than the DLD group (M = 8.88). There was also a significant main effect of task, F_(1,37) = 19.64, p < .001, η_p² = .347, where the task with picture support had higher MLCU (M = 10.89) than the task without picture support (M = 9.28). There was no significant interaction between group and task (p = .247, η_p² = .036).

Subordination index

A paired-samples t-test showed that for participants with TLD, the task with picture support yielded a higher SI on average, compared to the task without picture support, t₍₃₀₎ = 3.88, p < .001, d = 0.70. This significant effect of task on SI was not seen in the DLD group (p = .161, d = 0.68). There were no significant group differences (despite large effect sizes) in the task with picture support (p = .014, d = 1.00) or in the task without picture support (p = .040, d = 1.01).

Story grammar units

A paired-samples t-test showed that for participants with TLD (n = 30), the task with picture support elicited fewer SG units on average, compared to the task without picture support, t₍₂₉₎ = −5.44, p < .001, d = 0.99. This significant effect of task on SG units was not seen in the DLD group, despite a large effect size (p = .035, d = 1.21). There were also significant differences between groups in the task with picture support, U = 27.50, p < .001, d = 1.67, and the task without picture support, U = 50.50, p = .011, d = 1.12. The TLD group included more SG units in both tasks (with picture support Mdn = 19, without picture support Mdn = 23) compared to the DLD group (with picture support Mdn = 15, without picture support Mdn = 20.5).

Internal responses

A paired-samples t-test showed that there were no significant effects of task on the number of IR in the TLD group (p = .014, d = 0.48) or the DLD group (p =.317, d = 0.13). There was a significant group difference in the task with picture support, however, (U = 38.50, p = .003, d = 1.23), where the TLD group included more IR (Mdn = 3) than the DLD group (Mdn = 0.5). This significant group difference was not seen in the task without picture support (p = .024, d = 0.91).

Discussion

This study aimed to investigate the clinical value of two novel elicitation tasks for narrative retell intended for Swedish SLPs, one with and one without picture support. The study investigated language sample length, order effects, the effect of picture support, and group differences in 9–12-year-olds with typical language development (TLD) and developmental language disorder (DLD). The results showed clinically promising results in terms of the number of elicited utterances for participants in both groups, and no systematic order effects (due to practice or fatigue) in the TLD group. As expected, retells from children with DLD were shorter and had a shorter mean length of utterance/C-unit (MLCU) compared to children with TLD. In the TLD group, the task with picture support elicited fewer basic Story Grammar (SG) units but a higher Subordination Index (SI) than the task without picture support. The main effects of task showed that participants increased their MLCU in the task with picture support, but had a higher number of C-units in the task without picture support. In addition, participants with DLD included significantly fewer SG units in both tasks, and fewer Internal Responses (IR) in the task with picture support only compared to participants with TLD.

Language sample size

Both tasks elicited language samples with a sufficient number of complete and intelligible C-units according to the benchmark we set based on Heilmann, Nockerts, and Miller [22]. In fact, their 3-min narrative language sample contained on average 16 utterances, while the language samples in the present study contained on average 22–30 utterances for both groups of children, with an average sample length of around 2.5 min for participants with TLD and around 2 min for participants with DLD. Seven of eight included participants in the DLD group had at least 12 complete and intelligible utterances in both tasks, and one participant nearly met the benchmark with 11 utterances in the task without picture support. One participant with DLD was excluded completely from the analysis due to too few utterances, however. This participant also had the lowest results on the standardized language tests of all children participating in this study, indicating that the retelling task was too complex. Given that Swedish SLPs are recommended to always include a narrative task in the language, reading, and writing assessments of school-age children [9], the results of the present study indicate that for most children, these retell tasks will elicit a large enough sample. The results also indicate that the task with picture support can function as a scaffold for children with DLD, but the scaffold might not be enough for the children with the most severe language disorders.

Order effects

Possible order effects were investigated in the TLD group. A plausible learning effect could have been that the participants would show higher productivity or greater complexity in the last task, and the reverse could be a plausible effect of fatigue. The results indicated that there were no systematic order effects across these variables. One variable showed a significant group difference, however, where SI in the task without picture support was higher for the group with this task presented last. This effect on SI was not seen for the other task. It is also worth noting that this group had an even higher SI in the task with picture support (which was presented first), as well as higher MLCU overall compared to the other group. Taken together, this does not indicate that the difference was due to a practice effect. Still, additional analyses were made before collapsing all TLD children into one comparison group to ensure that results and conclusions did not change if only the participants with tasks presented in the same order were included. When more data from children with TLD has been collected for a reference database, order effects should be investigated again before making the material available for clinical use. Order effects can of course not be excluded in the DLD group either, since the tasks were always given in the same order, and these tasks can be expected to be harder and more tiring for children with DLD. The patterns of results in the DLD group do not indicate any clear effects of fatigue, however, and the impression during testing was that all participants were able to do their best in both tasks.

Task and group effects

The results show that there were some statistically significant effects of task, with the main effect of task on MLCU which indicated that the task with picture support elicited longer utterances in both groups and a higher SI in the TLD group. This is in line with the results from some previous studies including children and adolescents with TLD, and indicates that visual scaffolding may function as cognitive support [24,25]. In the present study, the participants with DLD also increased MLCU, which supports this notion, even though previous studies including younger children and children with DLD did not find this effect [20,27–29,31]. In the present study (and Bergman & Hallin [24] which used the same materials), the pictures were designed to depict explicit problem sequences, which might have encouraged more complex language [32]. On the other hand, the fact that there was a main effect of task on the number of C-units with more C-units in the task without picture support, and that both groups included more SG units in the same task (for TLD this difference was statistically significant) might indicate that picture support leads to less complete or complex narrative content. This is in line with the results from Schneider [30] who found that slightly younger children with DLD included fewer SG units in retellings with picture support compared to retelling without pictures, and hypothesized that the reason might be that children might exclude information visible in the pictures. Since the participants were instructed to include as many details as possible in the current task, and the examiner was perceived as a naïve listener, this is not a likely explanation in this case. A more likely explanation is that a task without picture support allows the speaker to tell the story from their mental representation, without being restricted by external specific pictures, and this might aid in including more content [30]. Yet another explanation is that for the tasks included in this study, differences in content play a role, which will be discussed in the limitations section.

In line with several previous studies including slightly younger children, retells from participants with DLD were shorter, had shorter utterances, and had fewer SG units than participants with TLD [e.g. 16,18]. Group differences in IR were only seen in the task with picture support, however. This might be an artifact of weak statistical power, and the need to correct for multiple comparisons, which might inflate the risk of type 2 errors [41], but could also have another explanation. Previous studies have shown that terms describing emotions are often challenging for young children and children with DLD [e.g. 14,18]. It might be that the participants with TLD were able to use the facial expressions in the pictures to a larger extent than participants with DLD [42] and that this contributed to the group difference (the characters’ feelings are shown in the pictures for six out of nine feelings mentioned). Half of the participants with DLD did not include any IR at all. In the TLD group, all but four participants included IR in the task with picture support and all but six in the task without. This also shows that the scoring of IR separately from other SG units makes sense for narrative tasks intended for clinical use since the inclusion of characters’ emotions in a narrative can be a meaningful and important intervention target [10].

Four of the eight participants with DLD had worked explicitly with Story Grammar as part of their curriculum, but an analysis of the retells from these four participants showed that they did not include a higher amount of SG units of IR compared to the other half of the DLD group. It should be noted, that in the only previous Swedish narrative study including 10-year-olds with and without DLD, no group differences were found in terms of SG units, story length, or grammatical complexity [5], so in comparison, the present study found more group differences. Task design could play a role, since Reuterskiöld, Hansson, and Sahlén [5] used a picture elicited narrative rather than a retelling, and the analyzed language samples were notably shorter with a range of 6–16 C-units for the children with DLD and 5–13 C-units for the children with TLD. Thus, from a clinical perspective, the number of C-units, as well as the significant group differences in the present study, are promising.

Limitations and future studies

The DLD group was small and heterogeneous and consisted of children with more severe forms of DLD (since most of them were going to specialized language units), which means that the results cannot be generalized to all children with DLD. The participants do represent the variability that SLPs will encounter, however. It was positive that despite this variability, language samples from eight out of nine participants with DLD were sufficiently large, and that the findings were in line with previous research. Bonferroni corrections might have increased the risk of type 2 errors, however, since effect sizes revealed that some medium to large effects (mainly between groups) did not reach statistical significance.

When interpreting the results regarding the effect of picture support, it is important to remember that the content of the stories was different, even though they were as closely matched as possible on all other aspects (including the specific Story Grammar units). It is impossible to know whether one of the stories was more engaging to the participants than the other, which may have affected the results. Earlier studies have shown that children speak more when talking about some subjects compared to others, which might be due to background knowledge, interest, and motivation [1]. Variables measuring linguistic complexity (e.g. MLCU and SI) should not be affected by engagement, however, so the effect on MLCU seen in this study can (and should) be viewed as an effect of picture support. In addition, there was no effect of task on story length, which speaks against that the participants engaged more with one of the stories. The fact that the TLD group included significantly more SG units in the task without picture support, could potentially be explained by content differences, or SG unit scoring, however. The scoring was developed for this study to give the child credit for the different sub-parts of the “Orientation”-unit, and to omit scoring for the second “plan”-unit in part 3 of both narratives, which is only implied in the original narratives (this differs from the scoring in Bergman and Hallin [24]). When a larger sample has been collected to provide clinical comparison data, the scoring of SG units is something that should be investigated and potentially developed further, before the tasks and manuals are made available for clinicians (alongside an investigation of potential order effects as previously mentioned). Future studies will also investigate the effect of age on the performance of these tasks, as well as the effect of gender since there is some indication that girls around this age may have stronger narrative skills than boys [37].

Conclusions and clinical implications

The purpose of this study was to investigate different aspects of two newly created retell tasks intended to be used by SLPs working with school-aged children and designed to capture narrative language reliably and efficiently. The conclusion is that both tasks work as intended, and both tasks might be useful in the clinic for slightly different purposes. To elicit a language sample with as syntactically complex language as possible, the task with picture support is recommended based on the results from the present study together with the results in Bergman and Hallin [24]. The task with picture support is also recommended for children with more severe language difficulties. The task without picture support, elicited on average more Story Grammar units, however, and might be a better choice if an analysis of content is in focus. Arguably, the retelling task without picture support also has a higher ecological validity, since visual support of the type included in this study rarely is available in everyday life.

Since there were no systematic order effects, an SLP who wants to include both tasks in one session can do them in any order. The tasks should not be given back-to-back, however, but other tasks should be administered in between to make sure that the child can do his/her best on both tasks. Each task takes 7–10 min to administer, and the SLP can code for SG units and IR during the retells. Transcription and coding of microstructural measures of one task take 20–30 min, where additional measures easily can be coded and analyzed using SALT, e.g. speech disruptions and morphosyntactic errors. Compared to other standardized tests, 30–40 min for the administration and analysis of one task may seem like a long time, on the other hand, the SLP gets a broad picture of the child’s ability to put many cognitive and linguistic skills together in narration [4], in a way that is more naturalistic than many other standardized language tests.

In conclusion, we believe that these two narrative retell tasks could be highly useful for Swedish SLPs. Clinically, there might not be resources or time to administer more than one narrative task per client, and the SLP should then choose a task depending on the focus of the assessment as outlined above. The SLP could also choose the task based on the needs and abilities of the client, since some children might be overwhelmed by the retelling task without pictures and might need picture support to be able to produce a language sample. Finally, the standardized presentation and scoring of the narrative tasks presented in this paper enable a more objective narrative assessment than previously have been possible in Swedish for this age group. These results bring us one step closer to improving language evaluations for Swedish school-aged children.

Acknowledgements

We want to thank the participating schools, clinics, parents, and children. We also want to thank the two anonymous reviewers whose suggestions helped improve and clarify this manuscript.

Disclosure statement

The authors report no conflict of interests.

Additional information

Notes on contributors

Elin Hällström

Elin Hällström, has a Degree of Master in Speech Language Pathology from Karolinska Institutet, Stockholm, Sweden. She is currently a doctoral student in Engineering Psychology at the Department of Health, Education and Technology, Luleå University of Technology, Luleå, Sweden.

Jenny Myr

Jenny Myr has a Degree of Master in Speech Language Pathology from Karolinska Institutet, Stockholm, Sweden and work as a school based SLP at the Stockholm International Academy Häggvik, Sollentuna, Sweden.

Anna Eva Hallin

Dr. Anna Eva Hallinis an Assistant Professor in Speech Language Pathology at Karolinska Institutet, Stockholm, Sweden. Her research focuses on language and literacy development and disorders.