Examining the Implementation of a Spanish-English Bilingual Conversational Agent to Promote Shared Reading at Home

• In Event: AI in Early Childhood: Impacts on STEM and Language Learning

Abstract

Objective
We developed a tablet-based interactive e-book system for Latinx Spanish-English bilingual children featuring two culturally relevant books. The system included Spanish and English versions of each book. These books were integrated with a bilingual AI-powered conversational agent (CA) embodied in the Sesame Street character Rosita. Throughout the reading process, the CA engaged children and their parents with two types of questions: child-directed questions, designed to promote comprehension of story content; and family-directed questions, designed to facilitate conversation between parents and children about story-related topics. We conducted a home study with 15 parent-child dyads to investigate the effectiveness of the e-book system. Results indicated that the books promoted children’s language learning and stimulated meaningful conversations among parents and children.

Theoretical Framework
The benefits of dialogic reading for young children’s language and literacy development are widely acknowledged (e.g., Mol et al., 2008). By engaging in question-answering and back-and-forth conversations during reading, children can build oral language skills that set a foundation for literacy acquisition and academic achievement. Capitalizing on rapid advances in conversational AI technology, recent research has begun to examine the prospects of leveraging CAs to promote dialogic reading (Xu et al., 2022). However, these studies have typically focused on monolingual (e.g., English-only) agents that do not afford multilingual learners the opportunity to draw on their full linguistic repertoire.

Method
This study consisted of two home visits, spaced one week apart, to Spanish-English bilingual parents and children in the southern California region. During the first visit, families were given tablets equipped with the e-book system, and provided with guidance on how to use the system. Families were advised to read each book at least once throughout the week, but extended use was encouraged. In the initial visit, parents completed a survey on home literacy practices, and children completed a pre-test to gauge their prior knowledge of the target vocabulary from the e-books (in both Spanish and English). During the follow-up visit, children were administered a post-test to assess their learning of target vocabulary and comprehension of story content. Additionally, we conducted semi-structured interviews with parents to understand their perceptions of the e-book system.

Results
Pre- and post-tests indicated that the children learned Spanish vocabulary (t = 2.06, p = .049), and that this learning was associated with children’s response rate to Rosita’s questions during Spanish reading sessions (𝛽 = 1.81, p = .007). For family-directed questions, significant associations were found between parents’ conversational strategies and children’s language production. Specifically, more elaboration in parents was associated with a greater quantity and length of utterances among children. Finally, thematic analysis based on interview data revealed six key themes that captured the perceptions of parents and children on reading with Rosita (see Figure 2).

Scholarly Significance
Parent-child joint reading is an enriching activity that can be effectively supported with AI-based CAs. In order to serve children from diverse backgrounds, consideration should be given to leveraging technological advancements for linguistically and culturally relevant design.

Authors

• Kunlei He, University of California - Irvine

• Kelsyann Cervera, University of California - Irvine

• Julian Levine, University of California - Irvine

• Ying Xu, Harvard University

• Penelope Collins, University of California - Irvine

• Mark Warschauer, University of California - Irvine