Archiving Natural Language Data Sets of Parent-Child Conversations

University Of Michigan At Ann Arbor, Ann Arbor MI

Investigators

Abstract

PROJECT SUMMARY A critical task of childhood is acquiring a set of concepts to organize information efficiently, guide inductive inferences, and serve as building blocks for larger systems of knowledge. This process is powerfully social and interactive, permitting humans to pass down important discoveries and ideas from one generation to the next. Children are exquisitely sensitive to language testimony and statistical features of the input to guide learning. There is thus a great need to supplement experimental, lab-based studies with observational data on parent- child conversations. Gathering and analyzing such data sets requires extensive investments of time and effort; thus it would benefit the field for individual researchers who have gathered such data sets to make them available to other researchers. The goal of this research project is to archive, document, and make publicly available 11 existing data sets of parent-child conversations for use by the broader research community. The data sets were supported by two NICHD grants to S. Gelman (HD23378 and HD36043), including 1,034 participants and over 180,000 utterances, with typically developing children ranging in age 1.5 to 7 years and their parents. These data sets are innovative in several respects: (a) They provide parent-child conversations regarding concepts and categories for a large sample of children within an understudied age range. (b) A quasi-naturalistic method was employed, in which parents and children conversed freely, but the stimuli and contexts were standardized. This approach has the advantage of providing dense sampling of topics of interest (e.g., categorization) and control over factors such as item frequency, familiarity, and salience. (c) The data sets are particularly innovative in their inclusion of two understudied languages, Marathi (an Indo-Aryan language) and Southern Quechua (an endangered indigenous language spoken in Peru and Bolivia). Distribution will be via the Child Language Data Exchange System (CHILDES), which provides sophisticated search and analytic tools. These rich datasets will permit important secondary analyses investigating competing models of concept learning and representation, and characterizing the learning environment in depth. The data will also permit establishing norms regarding typically developing use of language to talk about abstractions, which have been found to be impaired in children with ASD and SLI. Thus our specific aims are: Aim 1: To archive, document, and make publicly available (in the CHILDES database) 11 data sets of parent- child conversational interactions that were collected with NICHD support. Aim 2: To enable researchers to conduct secondary analyses on these data, in order to address research questions regarding concepts, categories, language development, and parental input. Aim 3: To provide full documentation of the stimuli created for these data sets (pretested, standardized, researcher-created picture books) for future researchers who wish to conduct extensions of these projects in different languages, different socioeconomic and cultural contexts, or different populations.