Poster #46 - Developing an iPad-based adaptation of the Looking while Listening Task for the NIH Baby Toolbox

• In Event: Poster Session 08

Abstract

The Looking while Listening (LwL) paradigm is a widely used and robust method of testing language comprehension and processing in infancy (Fernald, Zangl Portillo, & Marchman, 2008). Based on the visual world paradigm (Allopenna, Magnuson & Tanenhaus, 1998), it relies on infants’ tendency to direct their gaze toward a visual referent (e.g., a car) when hearing that referent’s label in natural speech (e.g., Look at the car!). The LwL task has been used with infants as young as 3 to 6-months (Burgleson & Swingley, 2012), and as early as 18 months provides measures that are predictive of later language and cognition (Marchman & Fernald, 2008). However, the LwL paradigm is predominantly used in laboratory research, and has never been part of a formal standardized developmental assessment. Our goal here is to describe the development of a novel version of LwL , adapted for the upcoming NIH Baby Toolbox as part of the Language Domain. This assessment will be administered entirely on an iPad, using automated gaze detection, tracking, and scoring.

The first step in adapting LwL for the NIH Baby Toolbox was to develop a method of automated gaze-tracking and data processing that could be completed entirely within the iPad. To do this, we utilized iOS ARKIT, a software framework for augmented reality applications, to extract estimated look-at points that are paired with known locations on the screen during an individual calibration process. Calibration is used to correct subsequent look-at-points values, allowing the program to differentiate left-right looking. The resulting software requires minimum training to administer. Data available for exports includes both raw x, y coordinates collected at a rate of 60 frames per second, as well as processed scores.

LwL task stimulus development involved multiple steps. Word items were selected using Wordbank, which archives child vocabulary data collected with the MacArthur-Bates Communicative Development Inventory (MB-CDI). We selected highly imageable nouns to form word-pairs matched by average age of acquisition as well as by their difficulty and discrimination parameters in an item-response theory model (Frank et al., 2021). A pilot study (N=80) with infants from 4-24 months (M age = 13.3 months, SD = 6.4) using an initial set of images revealed a strong correlation between accuracy and age r= .745, p<.01, suggesting excellent potential. Based on that preliminary data, new images were drafted to better control image saliency. A second pilot using these images will be completed in December 2022.

The final design, which will include both an English and Spanish version, will be normed starting May 2023, in a sample of 2,550 infants representing the racial and ethnic diversity of the United States. The resulting scores will contribute to the language composite score for the NIH Baby Toolbox which will be available for public use in the coming years. By designing a language comprehension measure for infants ranging from 0-3 years, the NIH Baby Toolbox provides a much-needed tool for researchers, clinicians and educators engaged in serving the developmental needs of our nation’s youngest children.

Authors

• Miriam A Novack, Northwestern University
  Presenting Author

• Aaron J Kaat, Feinberg School of Medicine, Northwestern University

• Elizabeth M Dworak, Northwestern University

• Adriana Weisleder, Northwestern University

• Jenny R Saffran, University of Wisconsin - Madison

• Michael C. Frank, Stanford University

• Maria Varela Diaz, Northwestern University

• Sandra R Waxman, Northwestern University