Training Inner Speech in Children With Developmental Language Disorder

Training Verbal Mediation to Support Executive Function in Children With Developmental Language Disorder

The complex and unclear relationship between language and executive function (EF) creates barriers to developing effective interventions for children with developmental language disorder (DLD) whose language difficulties often co-occur with impaired EF. Children and adults with typical language development (TD) facilitate their EF by using self-directed language, or verbal mediation, to guide conscious reflection and override habitual behaviors. Conversely, children with DLD do not use verbal mediation to support EF efficiently or effectively. Promising evidence suggests that language-based training can shape verbal mediation and improve EF task performance in children with TD, which makes it pertinent to determine whether verbal mediation training benefits children with DLD. Specifically, modeling interventions have been shown to promote learning of language forms without taxing the cognitive resources required for learning such as attention or working memory, which are known to be impaired among children with DLD. The long-term goal of the proposed work is to optimize intervention outcomes for children with DLD by elucidating the complex relationship between language and executive functions. The objective of this project is to determine the impact of modeling verbal mediation on shifting task performance in school-aged children with DLD. Shifting, also known as cognitive flexibility, is the ability to alternate between operations or mental sets. It is an important EF because it is the pivot point between multiple goal-directed tasks when language use is critical for guiding action. Children aged 8-10 years will complete three versions of a shifting task over three phases: pre-intervention, intervention, and post-intervention. During the intervention phase, half of the participants with DLD will be exposed to a task model with verbal mediation (training), while the other half will be exposed to a silent task model (control). The investigators will determine the effect of modeling verbal mediation on the subsequent use of verbal mediation (Aim 1) and behavioral and electrophysiological measures of shifting ability (Aim 2). Indirect measures of shifting (i.e., accuracy and reaction time) will be supplemented with an electrophysiological marker of shifting that reflects real-time cue processing. This combination of methods provides insight to changes in processing following intervention that may precede and predict subsequent changes in behaviors. Our central hypothesis is that modeling verbal mediation will facilitate more effective use of verbal mediation and improve shift cue processing in children with DLD. The project will provide a theoretical framework for the role of language in shaping goal-directed behavior and the first examination of electrophysiological change in shifting following a verbal mediation intervention. Results will have a significant impact on clinical practice by advancing knowledge about a promising language-based intervention to support EF and other goal-directed tasks.

This project aims to determine the impact of modeling verbal mediation on the executive function (EF) of shifting in school-aged children with DLD. The investigators will determine the effect of modeling verbal mediation on the use of verbal mediation (Aim 1) and behavioral and electrophysiological measures of shifting (Aim 2). Because accuracy and reaction time are not direct measures of the cognitive process that leads to shifting, they will be supplemented with an electrophysiological marker of shifting that reflects real-time cue processing.

PARTICIPANTS: Seventy-five children (N = 75) aged 8 to 10 years will be enrolled in this study over the course of three years. Target enrollment is 25 children in each group (DLD-training, DLD-control, TD-control) matched on age (+/- 3 months) and gender. Over-recruitment of 85 total participants accounts for up to 14 percent attrition due to excessive artifacts in the EEG recording such as excessive eye blinks and/or movements (estimated at 10 percent), or a participants failure to complete all study tasks and tests (estimated at 4 percent). It is important to note that the project is based on a system for EEG recording that is optimized for use with children, and that the PI has used this system to study children in this age range with minimal loss of data due to artifacts. See the EEG data acquisition and processing sections for details.

Age Justification: This age group was selected because (1) children have made substantial development in spoken language and EF by this time, (2) the early school years are considered a key period in development as children are faced with external challenges that require them to use language and executive skills efficiently, and (3) evidence exists that this age range marks a period with extensive use of verbal mediation to support EF task performance.

GENERAL STUDY PROCEDURE: The study will involve (a) standardized assessments of language and cognitive skills described below and (b) an experimental protocol including three phases: pre-intervention, intervention, and post-intervention. Each phase will require 1 visit to the research lab, with 1-2 weeks between each phase, and a desired timeframe between pre- and post-intervention sessions of 14-21 days.

• Classification to DLD and TD Groups: Children with DLD will have a history of delayed language development as reported by a caregiver during an initial phone screening and will meet the criteria for DLD as determined by administration of standardized assessments of language and cognitive skills. Children will be confirmed as having DLD if they obtain an Identification Core Score (ICS) of less than 34 on the Test of Integrated Language and Literacy Skills (TILLS). Children with TD will not have a history of delayed language development and they will obtain a score of 34 or higher on the TILLS. All children must obtain a standard score greater than 70 on the Fluid Subtest of the Kaufman Brief Intelligence Test, 2nd edition (KBIT-2) to exclude those with cognitive impairments.
• Assignment to Training or Control Conditions: The study is an individually randomized trial with a parallel intervention design. All TD participants will complete the control condition. DLD participants will be randomly assigned to either the training or control condition. Randomization will occur after confirming classification to the DLD group and before beginning the intervention phase of the experimental protocol. The PI will oversee a block randomization protocol to assign participant IDs to either the training or control condition. Distinct teams of research assistants (RAs) will be trained to (A) conduct pre- and post-intervention sessions, (B) conduct intervention sessions, and (C) code verbal mediation data collected during the intervention session. The PI will inform members of Team B about participant assignments using controlled access features of the REDCap database. Teams A and C will be blinded as to whether children with DLD were in the training or control group.

SHIFTING TASK OVERVIEW: Each participant will complete three versions of a shifting task. The task is a modified dimensional change card sort (DCCS) task. The objective is to correctly sort stimulus images according to one of two rules, indicated by a visual cue. Each task will contain two sorting rules and a total of four bivalent stimulus images. Shifting rules will be simple and familiar concepts (i.e., color, shape, number, and size) for school-aged children. Cues will be the written name of each rule, for example "color," presented before each stimulus While participants are not expected to have any difficulty reading the rule names, the cue will be both spoken and written during all practice trials to eliminate any potential effects of word reading ability. To minimize demands on working memory, the written cue will remain on the screen while the stimulus is presented. The three task versions will have the same basic structure but different sets of rules and stimuli (counterbalanced across phases and participants). The task will always begin with two single-rule blocks, allowing children to practice each rule individually with a goal of learning to associate each rule with its cue and button response options. All shifting tasks will be computer-administered and presented using E-Prime 3.0 software. Responses will be collected using a Chronos response box. The NIH Toolbox version of the DCCS had excellent developmental sensitivity, test-retest reliability, and convergent validity in a sample of children aged 3-15 years. The DCCS has also been adapted for use with EEG in children, adolescents, and adults.

AIM 1 (TRAINING VS. CONTROL INTERVENTION): Determine the effects of modeling verbal mediation on the use of verbal mediation during a shifting task in children with DLD.

To achieve Aim 1, half of the participants in the DLD group will be exposed to an experimental verbal mediation training while the other half (and all TD participants) will be exposed to a control condition, embedded in a version of the shifting task.

Eliciting and recording verbal mediation during the intervention phase: The task design allows for eliciting and recording verbal mediation behavior before, during, and after the training or control conditions. Audio and video recordings will be used for offline coding and analysis of the participants self-directed spoken language. After exposure to each single-rule block, participants will complete several mixed-rule blocks with pre-randomized trials described below:

• Pre-/Post-Training: Participants will complete a total of 20 mixed-rule trials before training/control and 20 mixed-rule trials after. During these trials, all participants will be prompted to say what they are thinking and planning to do before they enter a response. Examiners will remind participants to speak out loud as needed but will not provide any language modeling.
• Experimental training (modeling with verbal mediation): Participants will watch trials being completed while Max, a friendly character who knows the task very well, narrates his thoughts and actions to provide a model of effective verbal mediation. The language demonstrated will be future-oriented and structured as an implementation intention (i.e., if/then phrase) to support goal attainment. For example, participants will hear: "If this is the color game, then I should press the button for blue." An arrow will point toward the correct response.
• Control (modeling only): These participants will watch Max complete trials with no verbal mediation spoken aloud (i.e., an arrow will point toward the correct response). This control condition will allow the isolation of the effect of verbal mediation training because both conditions provide the same modeling of actions.
• Performance trials: In both the training and control conditions, every 10 modeling trials will be followed by 10 performance trials during which participants will be prompted to: "Tell Max what you plan to do before you do it." This alternating structure of modeling and performance will be repeated six times for a total of 60 model trials and 60 performance trials. This structure will allow the examination of change in verbal mediation throughout the intervention phase.

Verbal Mediation Measure: Each utterance, defined as a unit of speech with no temporal or semantic discontinuities (i.e., pauses greater than 2s or a change of content), will be coded for relevance and function. Relevance will be classified as: (a) irrelevant to the task, (b) relevant to task and self-directed, or (c) relevant to task but directed to examiner (e.g., asking questions to clarify rules, planned actions, etc.). Relevant utterances (b and c) will also be classified as one of three functions: (1) rehearsal, (2) regulating, or (3) affective. Rehearsal utterances would include simple statements of the rules or labeling of the stimulus. Regulating utterances would include the use of specific action verbs (e.g., switch) or the if/then framing provided in the model. Affective utterances would include emotional responses, evaluations of accuracy (e.g., oops), or mental states (e.g., I think…).

Fidelity Procedures: The PI will provide training on the coding procedures to study staff (RAs in Team C) who are blind to group membership. A random 25 percent of participants will be coded a second time to evaluate inter-rater reliability. The PI will monitor whether intraclass correlation for each utterance function category is greater than .75 and provide additional training as needed. Descriptive analyses will be used to quantify the amount of each utterance type produced. Relevant self-directed utterances produced during the 100 trials of Pre/Post-Training Verbal Mediation (20 trials pre, 20 trials post) and Performance (6 sets of 10 trials) will be further analyzed.

Data Analysis - Aim 1: To evaluate change in verbal mediation throughout the intervention phase, a random-intercept model will be performed with relevant utterance quantity regressed on (1) indicator for group (with DLD-control as the reference group; with parameters for the indicator for DLD-training and TD-control indicating differences for these groups from the reference group), (2) indicator for function of verbal mediation (with rehearsal as the reference group), (3) indicator for three task stages: pre-training (first 20 trials), training/control (60 performance trials) and post-training (last 20 trials), and (4) interaction of group indicator and task stage indicator (to evaluate differences between groups across stages). Multiple measures (of utterance quantity; controlling for utterance function) for each person will be accounted for by including the person-specific random intercept. Analysis will be conducted using RStudio and lme4 package. The investigators will also evaluate pre- and post-training differences in verbal mediation ability among children with and without DLD, by examining the number of each utterance type produced by all children with DLD (training + control) compared to TD-control participants using a series of independent samples t-tests with cluster-robust standard errors and using Bonferroni method to adjust for multiple comparisons.

Anticipated Results - Aim 1: modeling with verbal mediation will have a greater effect on the subsequent use of verbal mediation than modeling alone. Specifically, by the end of the intervention phase, it is expected that children in the DLD-training group will produce more relevant rehearsal and regulatory utterances than the DLD-control group, demonstrating more effective verbal mediation. The investigators also predict that at post-training, the DLD-training group will demonstrate fewer differences from the TD-control group than at pre-training; there will be a significant interaction between the relevant pairs of indicators. This comparison will serve to validate the novel outcome measures and further demonstrate effective change in verbal mediation for the DLD-training group.

AIM 2 (PRE- VS POST-INTERVENTION): Determine the effects of modeling verbal mediation on shifting in children with DLD.

To achieve Aim 2, DLD participants will complete two versions of the shifting task before and after the intervention phase while EEG is recorded. After learning and practicing each rule individually, participants will complete several mixed-rule blocks, or shifting blocks, that require participants to shift between the two rules on a trial-by-trial basis. When a color trial is followed by a shape trial (or vice versa), the latter is a switch trial; when a color trial is followed by a color trial, the latter is a stay trial. Importantly, the visual cue indicating what rule to use will continue to be presented before each trial (in a pre-randomized order), allowing for the categorization of the cue as a switch cue (cue that precedes a rule change) or a stay cue (cue that precedes a rule repetition).

Trial Structure: Each trial will begin with a fixation for 1000 ms, followed by a visual cue (e.g., color) for 1500 ms, followed by the stimulus (e.g., a blue circle) until the participant makes a response (or up to 7 s ) by pressing the left- or right-most button on the Chronos response box, and finally, a blank screen for 1500 ms before the next trial begins.

Behavioral Measure of Shifting Performance: During the pre-/post-intervention shifting blocks, two measures will be calculated: (1) accuracy or the percentage of correct trials and (2) reaction time in milliseconds by trial type (switch, stay). A main effect of trial type (stay, switch) in which switch trials are less accurate or have longer reaction times than stay trials will be interpreted as a switch effect.

EEG Data Acquisition: EEG will be recorded using the GES 400 system by Electrical Geodesics, Inc. (EGI). A 32-channel HydroCel Geodesic Sensor Net from EGI will be applied on the participants scalp to allow non-invasive EEG recording. The EEG will be continuously recorded at a 1000 Hz sampling rate, and electrode impedances will be kept below 50 kilo ohm. EEG data will be processed offline, using EEGLAB.

EEG Data Processing: Offline processing of the EEG data will include 0.1-40 Hz band pass filtering, segmentation of the EEG into 1000 ms long epochs (200 ms before cue onset and 800 ms after cue onset), artifact rejection of trials contaminated by eye blinks, artifact correction through independent component analysis (ICA), signal averaging, baseline correction (-200 ms to 0 ms), and re-referencing to the average. Data will then be submitted to a Temporal Principal Component Analysis (TPCA) using MATLAB to allow the separation of potentially overlapping components, and for an analytic reduction of the temporal dimensionality of the data. The TPCA methodology is commonly used for the analysis of ERP components, including the P3. Factors that account for 95 percent of the variance in the input data set will be retained for Promax rotation. Factor scores will be obtained for each participant and task condition (switch, stay). These factor scores reflect the level of activation (magnitude) of a specific ERP component of interest for each individual.

Electrophysiological Measure of Cue Detection: The cue-P3 is a well-studied ERP component that has been examined in adults and children, including children with dyslexia. During the pre-/post-intervention shifting blocks, amplitude of the cue-P3 ERP component for switch and stay cues will be measured. Factor scores from the PCA will serve as the amplitude measure of the cue-P3. A main effect of cue type (switch, stay) in which the amplitude for switch cues is larger than for stay cues will be interpreted as the cue-P3 switch effect.

Data Analysis - Aim 2: To evaluate change in shifting post-intervention, a repeated-measures ANOVA will be performed for each outcome measure (accuracy, reaction time, cue-P3 amplitude) with one between-subjects variable: group (DLD-training, DLD-control) and two within-subjects variables: time (pre-intervention, post-intervention) and trial type or cue type (switch, stay). Since there are multiple dependent variables to compare, investigators will adjust for multiple comparisons using the sharpened False Discovery Rate (FDR) q-value method.

Anticipated Results - Aim 2: a greater reduction in reaction times in the DLD-training group than in the DLD-control group post-intervention. These findings will indicate a more efficient use of the shift cue after training. Specifically, the investigators predict that post-intervention switch cost will be larger for the DLD-training group than the DLD-control group, which would indicate a more typical and effective pattern of a differential response to switch and stay cues. ERP results will elucidate changes in the processing of switch cues that are not detected by behavioral measures. The investigators also predict that the cue-P3 switch effect will be larger for the DLD-training group than the DLD-control group post-intervention, indicating that modeling verbal mediation influences how cues are detected and interpreted.

POWER ANALYSIS. AIM 1: With N = 75 (25 in each of the three groups) treated as three clusters and with 100 measures within each person, with power of 80 percent and at an alpha level of 0.05 with a two-tailed test, and assuming the ratio of between-person variance to total variance between 0.30-0.40, the investigators estimate minimum detectable effect size (MDES; in standard deviation units) at 0.36-0.42.

AIM 2: With N = 50 (25 in each of the two DLD groups), with power of 80 percent and at an alpha level of 0.05, using mixed ANOVA and assuming the correlation between the pre- and post-intervention measures at 0.5, the investigators will be able to detect an effect size as small as Cohen's d = 0.41 (between small and medium).

Inclusion criteria (children with developmental language disorder; DLD):

• Age 8;0-9;11 years English as the primary language
• Pass hearing screen bilaterally (20 dB HL, at 1K, 2K, & 4K)
• TILLS Identification Core Score < 34

Inclusion criteria (children with typical development; TD):

• Age 8;0-9;11 years English as the primary language
• Pass hearing screen bilaterally (20 dB HL, at 1K, 2K, & 4K)
• TILLS Identification Core Score > or = 34

Exclusion criteria (DLD & TD groups):

• KBIT-2 Fluid Subtest Standard Score < 70
• ADHD, Autism, TBI, or other neurological deficits or disorders