Archiving a large audiovisual dataset of early childhood experiences
Skidmore College, Saratoga Springs NY
Investigators
Abstract
Project Summary Understanding how children learn and especially understanding how children learn language is a critical public health issue. Positive academic outcomes (which are predicted by early language skill) are associated with positive health outcomes like lower rates of substance abuse, higher rates of offspring survival, higher occupational success, and longer lifespans (Hawkins, Catalano, & Miller, 1992? Low, Low, Baumler, & Huynh, 2005? McGregor et al., 2007? Serbin, Stack, & Kingdon, 2013). While we know that early social, cognitive, and language ability predict educational outcomes (McGregor et al., 2007? Marchman & Fernald, 2008), many basic questions about early child development remain unanswered, making it challenging to design effective early educational policy and interventions. One important question is how a child?s input (e.g., the things that they see and hear in daily life) predicts what they end up learning. This question is relevant to researchers interested in all aspects of development. Answering such a question, however, requires actually measuring a child?s input something that until recently was technologically impossible. PI Sullivan and Drs. Frank and Perfors (see letters of support) created a large longitudinal dataset of videos from the child?s perspective to measure input. Our goal is to make this dataset available and accessible to other researchers. Using a headmounted camera, we recorded everything that participants saw and heard from their perspective for approximately 2.5 hours a week over the course of two years (from infancy through toddlerhood), and continued through toddlerhood. Recordings were naturalistic, and included a wide array of contexts and activities that have never previously been recorded. This resulted in a dataset of over 325 hours of audiovisual recordings, along with a dense collection of cognitive, social, and linguistic measures that were also collected longitudinally. This dataset is the first of its kind, and is unique in its size, scope, and perspective. This R03 proposal has three main aims. First, we aim to post the entirety of our dataset to Databrary, an NIHfunded host for video data relevant to child development research. This will allow other researchers to access our rich dataset. This will require converting the videos into a standardized format and collaborating with Databrary to host the videos. Our second aim is to attach the appropriate metadata to the videos so that they can be searchable by researchers. To this end, we will hire a fulltime research assistant who will create a database of both videolevel data (e.g., the time of day and year, the child?s age, the people featured in the video) and timestamped data (e.g., names of locations, activities, and objects salient in the child?s visual field). Third, we will transcribe and provide a detailed coding of a sizeable subset of our corpus (including verbatim transcriptions, descriptions of the objects being touched and seen, and coding of the referents of each speech act) that will allow other researchers to immediately begin answering questions of our data.
View original record on NIH RePORTER →