Atypical speech and language development

Speech and language delay and disorders can affect 3–16 percent of children [5]. While some have a known cause, such as hearing loss (13.3%), many do not have an identifiable origin [6]. Developmental Language Disorder (DLD) represents a prominent example as it is a highly prevalent condition where individuals can exhibit deficits in receptive or expressive language without known cause. It has previously been described by many terms, such as ‘Specific Language Impairment (SLI)’ and ‘Primary Language Impairment’. Most recently, there was an effort by the CATALISE consortium [7–9] to establish consensus among experts of different backgrounds, leading to the current label ‘Developmental Language Disorder (DLD)’. It was previously estimated that SLI can affect at least 7% of the population [10]. Affected children can experience long-term problems, including behavioral, psychiatric, emotional, academic and social adaptation [9,11–13]. Children may begin to show language deficits as early as infancy, but DLD cannot be reliably diagnosed until age four [14], meaning the critical period in which intervention could be the most effective may be missed. Additionally, the majority of DLD cases in children are thought to remain unidentified, further indicating a great need to improve the early detection process by establishing risk factors [10,15–17].

DLD has been shown to have a genetic component such that individuals have a significantly elevated risk of developing DLD if there is a familial history [18–20]. For example, close to half of the siblings of individuals with DLD (SLI) exhibit difficulties in language [21], and infant siblings of children with DLD (SLI) scored significantly lower on language measures when they were tested at 3 years of age, compared to infant siblings of children without SLI [22]. Additionally, Hayiou-Thomas et al. [23] found in a large twin study that both environmental effects and genetics can impact early language development, and genetics appear to be the driving force behind individual differences in middle childhood and early adolescence.

Still, individual variability exists even in infants with family risk and much more needs to be learned to precisely predict future onset of DLD in an individual child, which may require taking into consideration both family and individual risk factors. Very few studies so far have examined early predictors of DLD in infancy, and most have focused on behavioral predictors. For example, Hsu & Iyer [24] found that lower usage of deictic gestures (pointing, reaching, showing, and giving) and conventional gestures (e.g., finger to lips for “hush”) as well as decreased vocabulary size at 15 months predicted risk for language impairment at 3 and 4.5 years of age. Similarly, Lüke et al. [25] observed that pointing ability at one year predicts language skills at 5 and 6 years of age, and this effect was mediated by comprehension of iconic gestures (e.g., opening and closing the extended index and middle finger to represent scissors) between 3–5 years. Other behavioral predictors are generally measured after 2 years of age and include receptive and expressive vocabulary, word combinations, and syntactic comprehension (see [14]). While these predictors can be useful, it is challenging to elicit and measure specific behaviors from infants. As an alternative method, neural predictors can provide richer and more nuanced data without requiring the active involvement of infants. Despite this advantage, neural predictors in infancy are yet to be examined.

Given recent findings that early neural speech processing in typically developing infants may predict later individual grammar skills [4], an important next step toward a sensitive and specific neural predictor of DLD is to examine, at the group level, neural speech processing in infants with vs. without first-degree family members who have a history of language delays and/or disorders (Low- vs High-Risk). This will provide a valuable window for us to uncover any already altered neural mechanisms that can later be tested as candidate predictors of atypical speech and language development using regression-based methods. Understanding the group-level differences between High-Risk and Low-Risk infants would not only allow for the expansion of language acquisition theories to encompass atypical development, but also help identify potentially much earlier markers for developmental speech and language disorders. This is a foundational step before we can further examine the exact predictive values of specific early markers for individual children’s later language outcomes.

Language environment as a potential risk factor

In addition to family history, speech input from infants’ language environment may also have an important impact, as it has been repeatedly demonstrated to play a significant role in the language development of typically developing infants. In a seminal lab-based intervention study, only in-person language input in a social environment changed infants’ speech perception, demonstrating the importance of quality language input [26]. This effect was later replicated by analyzing daylong recordings made in typically developing infants’ home environments. The more infant-directed speech (IDS), characterized by more variable and increased vocal pitch, slower speech, and clearer articulation [27], in an infant’s home environment is observed, the stronger their later language outcome [28]. Further, by coaching parents to produce high-quality IDS during the sensitive period (6–12 months), researchers have demonstrated a significant intervention-related benefit in language development, when measured through both examining daylong audio recordings in infants’ home environments and parental surveys up to 30 months [29,30].

The effects of early speech quality and input have also been documented in clinical populations. For example, mothers exhibiting post-partum depression tend to speak less to their infants overall, and their infant-directed speech is reduced in both quantity and quality [31,32]. Mothers of infants at-risk for dyslexia have been found to hyperarticulate vowels in IDS significantly less than mothers of Low-Risk/typically developing infants [33]. It is thus possible that early speech input is significantly related to the family history risk factor, i.e., the quantity and/or quality of the speech input may be reduced if there is presence of speech and language disorder in the family. Such a relation would make the reduced speech input become a risk factor itself.

Therefore, it is important to evaluate the speech input in an infant’s language environment in parallel with the familial history risk factor. Language ENvironment Analysis (LENA) technology has provided an optimal method to assess infants’ speech input in a natural language environment, with the capacity for daylong recordings (i.e., up to 16 hours a day) using a small recorder that infants can wear. From these recordings, we will be able to calculate measures such as the total amount of speech infants hear and, more importantly, how much of that is infant-directed.

Neural measures for assessing the sensitive period for phonetic learning

The recent advances in neural imaging technologies, particularly for infants [34], are now providing us the critical tools to start investigating and understanding the neural mechanisms important for speech learning during the sensitive period. Magnetoencephalography (MEG) is a safe and non-invasive method of evaluating functional brain activity with high temporal and spatial resolution [35]. Two neural indices are the focus of the current MEG measurement protocol, namely, the mismatch response (MMR) and the complex auditory brainstem response (cABR). First documented using the Electroencephalography (EEG), the mismatch negativity/response (MMN/R) is one of the most widely studied neural signatures that indexes neural sensitivity to sound change [36]. For infants during the sensitive period, converging evidence demonstrated similar MMN/R changes in both EEG and MEG [37–40]. MEG technology further allows a closer examination of underlying neural sources, or generators, of the mismatch response (MMR). So far, research has identified MMR in the left inferior frontal region (IF) to be a key process that holds higher predictive value for later language skills [3,4].

Another neural process, namely, the complex auditory brainstem response (cABR) has received increased research interest in recent years as a measure that characterizes the sensory encoding of the acoustic properties of complex sounds (e.g., speech sounds) in the early stages/lower level of auditory processing [41]. In adults, instead of a passive relay for information up through the auditory system, the cABR has been shown to be malleable by higher-level processes, such as early language and music experience, as well as short-term training [42–44]. Critically, our most recent study examined the frequency-following response (FFR), a component of cABR that tracks the periodic portion of the sound, in 7- and 11-month-old infants and found that both language and music experience affected the FFR [45]. This result suggests that cABR could also be an important neural process underlying the sensitive period. However, very limited data currently exists on infant cABR and there is a need to further understand changes in cABR with a focus on the sensitive period using a wide range of speech sounds. Further, methodologically, spatial information has become crucial for cABR, given recent research has suggested cortical contribution to FFR, originally thought to only come from subcortical regions [46]. Yet, most existing cABR relies on recording from a single EEG sensor, which cannot provide such spatial information in the cABR, making it crucial to gather MEG-recorded FFR/cABR.

Current study and planned aims

The overarching goal of this current study is to build a large longitudinal dataset with infants at varying risk level for atypical language development, which will allow us to start addressing important research questions. Critically, this study gathers extensive information to characterize the familial risks not only by collecting detailed background information through parental surveys, but also by comprehensively assessing the language skills of the infants’ older sibling and examining the families’ home language environments. Infants’ neural processing of speech is recorded at three time points corresponding to the beginning (6 months), end (12 months) and a delayed end of the sensitive period (14 months) using magnetoencephalography (MEG) with MMR and cABR as target measures. Infant language development is further tracked with the MacArthur-Bates Communicative Development Inventory every three months between 18–30 months of age, followed by a comprehensive assessment of the same infants’ language skills when they reach elementary school age. In the Methods section, we document the study design, materials, and procedures used in data collection. The current study will be foundational in future potential clinical applications that would deploy brain-based measures very early in development to flag risk for later speech- and language-related disorders.

Our research team aims to address the series of questions listed below based on data collected in this study: Q1. Are there group-level differences between High- vs. Low-Risk infants already detectable through our target neural metrics around the sensitive period for phonetic learning? Q2. Are there differences in speech input, both quantity and quality, between High- vs. Low-Risk infants? Q3. Do High-Risk and Low-Risk infants present different language development trajectories? Q4. Can we improve prediction of individual infants’ speech and language outcomes at school entry by taking both family history and individual neural metrics into consideration?

Participants

Eligibility.

Participants are infant-sibling pairs who share at least one biological parent and were enrolled by their parent(s). The following describes our enrollment criteria (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Enrollment criteria.

https://doi.org/10.1371/journal.pone.0335596.t001

Ethics statement

This study was approved by the University of Washington Institutional Review Board under STUDY00015376. Written consent was obtained from all participants for all study procedures.

Overview

Fig 1 illustrates the stages of this study. Interested parents were invited to complete a screening form and, if deemed eligible, were contacted via phone by a research assistant to discuss the study and enroll. Next, families were asked to complete a parental questionnaire before attending a brief Zoom call conducted by one of our licensed research SLPs, who verified their responses to the questionnaire and further discussed family health and language history with the parent(s). Then, the participating older sibling of the infant completed their language and behavioral assessment with the SLP during an in-lab visit before the infant turned 6 months of age. Next, the infants begin their magnetoencephalography (MEG) appointments, which are completed within two weeks of their six-, twelve-, and fourteen-month birthdays. At nine months, some families elect to be sent a Language ENvironment Analysis (LENA) recording device to record their infants’ home language environments for one typical day (around 12.5 hours and up to 16 hours). Additionally, families are asked to complete online versions of Communicative Development Inventories (CDIs) when their infants are 18, 21, 24, 27, and 30 months of age to track vocabulary and grammar development. Finally, when infants reach 6 years of age (school entry), they will be recruited to return for a language assessment. By incorporating extensive data collection methods in a longitudinal format, this study will provide a rich dataset that will allow researchers to thoroughly examine early predictors of (a)typical language development.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Study timeline.

https://doi.org/10.1371/journal.pone.0335596.g001

Screening process

Interested families completed a REDCap screening form to provide information about eligibility criteria (ages, due dates, birth weights, etc.), risk status (language history), and contact information (email, phone number, and best times to contact). Parents could choose to skip any or all questions. Screening forms were reviewed by the research team, and qualifying families were called by a research assistant (RA). During the call, the RA went over each procedure in detail and emphasized the longitudinal nature of the study involving four in-person visits to the lab (UW I-LABS) and optional at-home measures. Furthermore, the RA went over the compensation system, which is done through e-gift cards. Participants could ask questions during this call and, if they decided to enroll, their first appointment (i.e., the older sibling’s language assessment) was scheduled. Upon enrollment, participants were sent a confirmation email containing detailed information and instructions about their upcoming appointment as well as links to their REDCap consent form and questionnaire. Screening forms from families who were not enrolled were deleted by an RA, who noted the reason for non-enrollment in a secure document.

Informed consent

Participating families are provided detailed study information prior to and throughout their participation. First, before enrolling, a research assistant gave detailed descriptions of the in-person procedures of the study and answered any questions over the phone. Additionally, prior to each in-person appointment, the research SLP or RA goes over the appointment process again with the families. For the at-home components, detailed descriptions of the measure are provided through email. Participating families are reminded at each stage of the study that they can ask questions, end appointments early, and/or unenroll from the study at any time. Finally, a parent or legal guardian of each participant is required to complete a written consent form for all study procedures they elect to participate in detailing their involvement as well as the study’s purpose, risks, and benefits.

Parental questionnaire

The parental questionnaire was administered through REDCap and collected information about the infants’ parents, such as their age, gender, race, education, economic status, living arrangements, and health history. It also included questions about the infants and their biological siblings, including their health and development (pregnancy/birth complications, milestones, medical screenings, medications, allergies, language development), childcare, auditory environment (additional languages, music exposure), and sleep schedules. Participants were asked to complete their questionnaire and consent form before meeting with the SLP to discuss their responses. Both structured and unstructured text data were generated from the questionnaire and securely stored on the ITHS REDCap server with access limited to team members who are actively involved in the project.

Language and behavioral assessment: Older sibling

Magnetoencephalography (MEG) sessions

For all MEG appointments, we follow the best practices outlined by Clarke et al. [56]. MEG assessments are scheduled early in the day following the infant’s nap and mealtime to increase the likelihood of a cheerful and cooperative infant during testing. Once the parent arrives and before beginning the session, the RA verbally checks in with the parent to ensure their infant does not need to be fed or changed, and inquires about any notable updates in their development since enrollment or the last visit. Next, the parent, infant, and research assistant simultaneously complete a “metal check”, in which they stand and move with the infant near the MEG machine sensors to ensure that no ferrous metal is present, as this metal can interfere with data quality. If interference is detected, its source is determined (clothing, jewelry, etc.) and parents are asked to remove the item from themselves or their infant if they are comfortable doing so before performing a second check. Adult- and infant-size scrubs are kept in the testing area, and parents are given the option to wear them if they do not have metal-free clothing on hand. Some parents have metal on their person that they are unable or unwilling to remove (e.g., jewelry or a permanent retainer); testing still occurs if it is small and the parent agrees to remain still in their seat.

After the metal check is complete, the first stage of the appointment begins: MEG preparation and digitization. During this segment, the infant is securely strapped into a highchair with Velcro straps while their parent sits out of sight. The parent is told they can interact with and soothe their infant whenever they like during the appointment, but are asked to remain just out of sight of the infant during the preparation process (i.e., sitting in a chair behind them).

To begin preparation, the MEG technician places a soft stretchy cap over the infant’s head with the strap sitting beneath their chin. Head Position Indicator (HPI) coils are taped to the cap in fixed locations to provide data about the infant’s head position relative to the MEG sensors. Next, the MEG technician uses a Polhemus 3-D magnetic digitizer pen to mark points on the infant’s head, including the locations of the HPI coils and the cardinal points. Electrocardiography (ECG) and electrooculography (EOG) electrodes are taped on the infant’s chest and back to measure muscle activity and heart rate, and on the outsides of their face to measure blinking activity. Next, small soft earbuds are placed in both of the infant’s ears and taped down with skin-safe tape to ensure they remain in place. Finally, a foam halo is placed around the infant’s head to secure the coils, protect their head, and reduce head movement in the machine. Throughout this stage, a research assistant speaks to the infant and engages them with various toys to keep them calm. The MEG technician takes detailed notes of the digitization process.

Once the preparation stage is complete, the infant is securely strapped into the MEG infant seat with Velcro straps, and the seat is adjusted so that the bottom of the dewar containing the superconducting sensors is near their shoulders. The parent is once again seated out of their infant’s sight but within reach of them. When the infant is secure, adjusted, and calm, one of six presentation sequences (randomly selected via RANDBETWEEN Excel function and tracked in a spreadsheet) initiates sound presentation, which employs a novel and time-efficient paradigm that presents speech sounds in 200 clusters, allowing for simultaneous assessment of cortical MMR and subcortical FFR. Each cluster has 8–14 tokens of the same speech stimuli. In each speech sound cluster, the presenting rate accelerates in the first few sounds and decelerates in the last few sounds, with jittered stimulus onset asynchrony to minimize predictability. See Cheng & Zhao [57] for further details.

During this time, a separate workstation records raw data from the MEG for later offline analysis. Throughout testing, a research assistant silently engages the infant with toys along with a video screen showing infant faces to keep them calm and alert. If the infant becomes fussy, the RA (and occasionally the parents) attempt to soothe them. If they do not calm down within a few minutes, the testing is paused or halted. If the infant remains calm, the appointment ends once the 15 minutes of trials finish. Video and audio are recorded from three different angles to provide context during data analysis. The audio/video recordings and the MEG recordings are uploaded to a password-protected server after each session. Additionally, the MEG technician monitors the infant’s behaviors and takes detailed notes. Participants are compensated for each completed MEG appointment, regardless of whether data collection was successful.

Language ENvironment Analysis (LENA) recording at home

The LENA recorders are small recording devices that infants wear in specialized vests and are capable of recording up to 16 hours of audio in infants’ naturalistic sound environments. In addition, LENA software can process these recordings to compute a series of metrics to characterize the language input, including the number of words and conversational turns the infants experienced. LENA devices have many strengths. For example, they capture the full range of people who speak to and around infants (parents, grandparents, siblings, nannies, etc.). Additionally, they are a more subtle form of observation than lab visits, meaning participants are less likely to modify their natural behaviors due to bias. Finally, the day-long recording allows characterization of input across various contexts.

The recordings are scheduled for a typical weekend day within 3 weeks of infants’ nine-month birthdays, defined as days in which both caregivers are present for the majority of the day and the families are not sick, out of town, etc. Participating families are sent a package containing the LENA device, an infant vest specially made to hold the device, written instructions, and an activity diary. The instructions detail how to use the recorder and tell parents to remove the vest during naps, car rides, and baths for safety purposes, and to place the recorder near the infant in these situations if possible. To encourage participation in this measure, the option to have the recording package hand-delivered and retrieved by the research team is offered. Once the research team receives the returned LENA device, the family is compensated.

The LENA device is simple to use. After the families power on the recorders in the morning and place them in the infant vests, the devices automatically record for a full day and can turn themselves off if the storage limit is reached, though many parents manually end recordings after their infant has gone to bed. In addition to the recording, participants complete an activity diary to provide brief context for the recording in half-hour increments and make any relevant notes (e.g., breakfast, playing with sibling, nap time). Furthermore, families are provided a “Notice of Recording Script” which they can use to inform others that their child’s sound environment is being recorded for research. After their recording is complete, families mail the recorders, vests, and diaries back to the lab using our pre-paid return packaging. The encoded audio can only be uploaded using LENA’s secure cloud-based software onto our lab account. Data are automatically deleted from the devices after uploading and stored on our secure hard drive and server. Parents are assured that they may omit specific portions of the recording if desired.

MacArthur-Bates Communicative Development Inventories

The Words and Sentences form of the MacArthur-Bates Communicative Development Inventories [58] is a validated and common way for evaluating language skills between ages 16–30 months [59], including the onset of certain vocabulary, word combinations, and grammatical forms. This survey method is advantageous as parents provide their assessment of their children based on continuous observation in diverse settings, as opposed to a brief laboratory visit that may not accurately represent a child’s behavior in their daily environment [60]. After families consent to participate in this measure prior to their infants’ 18-month birthday, REDCap is programmed to automatically send emails containing their unique survey links on their infants’ designated birthdays (18, 21, 24, 27, and 30 months of age). Parents can complete the survey remotely whenever it is convenient and may choose to save their partial response to finish later if needed. Additionally, if parents do not complete the form, a maximum of three reminder emails are automatically sent out every three days until the form is completed. If they still do not complete the form, this is considered a missing data point. Families are compensated for each completed survey, and their responses are stored in REDCap and scored using Python scripts. The use of CDIs to track our participants’ language development through 30 months of age will allow us to analyze how the various predictive factors we are examining correlate with their language scores.

Comprehensive language assessment: Infant six-year follow-up

Parents will be recruited for an in-lab follow-up language assessment session when the infant participants have turned six years of age and completed kindergarten. This assessment will follow the same protocol as the older sibling language assessment with two modifications. Firstly, the SLP will have the parent(s) complete a brief follow-up survey similar to the parental questionnaire and provide informed consent in-person at the beginning of the appointment rather than via Zoom. Secondly, the Kaufman Brief Intelligence Test, Second Edition (KBIT-2; [61]) will be used in place of the TONI-4 for assessing non-verbal intelligence, which takes approximately 20 minutes to complete. This is because the TONI-4’s ability to accommodate 3-year-olds is no longer required and the KBIT will allow this dataset to merge with previous datasets using the same protocol. By following up with infant participants at six years of age to assess their language abilities, we can better examine how the measures taken in infancy predict language development and later language difficulties. The same data types and data storages will be used as those employed for behavioral assessment for older siblings.

Piloting and training

Prior to data collection, pilot MEG appointments were completed with infants from each age group to determine the feasibility of the MEG appointment protocol. Additionally, both research assistants who ran the MEG appointments (first and second author) completed 15–20 practice sessions with infants in an MEG simulator to learn techniques and best practices in infant neuroimaging studies [56], ensuring the success of data collection.

Data management plan

Data information associated with specific study procedures was described in each procedure section. All data files are de-identified and are designated a participant ID. The only link between identifiable information and de-identified data is securely stored in REDCap which only the research team has access to. The PI conducts weekly meetings with key study personnel to ensure the timeliness and quality of data entry.

Raw data are stored on a local password-protected server with highly controlled access as they may still contain private information, as allowed by the informed consent and the institutional certification. The informed consents contain language permitting secondary use with broad data sharing (de-identified) for general research use. Participating families will not be contacted or re-consented for future sharing or accessing data through repositories.

De-identified processed data, with parents’ consent for sharing, will be deposited at the Open Science Framework (OSF) repository under a project that can be findable and identifiable with a DOI created by OSF. The data deposit will occur when results generated from specific data are published. Data will be preserved within the repositories for at least three years following the completion of the research grant.

Data preprocessing and planned analysis