The Center for Effective Global Action (CEGA) is a hub for research, training and innovation headquartered at the University of California, Berkeley. We generate insights that leaders can use to improve policies, programs, and people’s lives. Our academic network includes more than 150 faculty, 65 scholars from low- and middle-income countries, and hundreds of graduate students–from across academic disciplines and across the globe–to produce rigorous evidence about what works to expand education, health, and economic opportunities for people living in poverty.
Differences in academic achievement between high and low socioeconomic status (SES) children arise at a very early age. Our study team, led by PIs Supreet Kaur (Economics) and Mahesh Srinivasan (Psychology), will test the idea that the psychological experience of poverty leads parents to engage less with their young children, hampering early child development. We focus on parent-child verbal interactions, which differ markedly by SES in observational data, and are the most prominent proxy for parental engagement in developmental psychology. Our study will work with families in the Bay Area to gather in-home long-form audio recordings (4-16 hours long) of parent-child interactions for analysis using NLP tools.
With the guidance of the research team, the Discovery team would help apply Natural Language Processing algorithms and tools (like voice type classifiers) to analyze raw speech audio data collected through the project. The position will involve:
- Extracting features of both child (toddler) and adult speech from long-form audio recordings using NLP tools
- Constructing measures of parent-child interactions such as conversational turns and adult word count
- Generating other potential outcomes of interest using the speech data around adult conversational interactions, such as affect or valence
- Validating algorithm performance, for example, in the presence of background noise in recordings such as tv or music inside participants’ households
- Data processing and handling a high volume (100+ hours) of raw speech audio
