Skip to main content Skip to navigation

Inspecting Spoken Language Understanding from Kids for Basic Math Learning at Home

Project Overview

The document explores the application of generative AI in early childhood math education through a novel multimodal dialogue system aimed at enhancing learning experiences at home. By utilizing conversational AI techniques, the system engages children in interactive, play-based math activities that bolster their spoken language understanding. It underscores the critical need for effective STEM education and illustrates how AI can foster engaging educational environments. While the study acknowledges challenges such as speech recognition accuracy and the difficulty of understanding children's speech in noisy home contexts, it also emphasizes the potential of this AI-driven approach to significantly improve math skills among young learners. Overall, the findings suggest that generative AI can play a transformative role in addressing educational challenges and enriching early childhood learning experiences.

Key Applications

Multimodal Dialogue System for Math Learning

Context: At-home learning for children aged 5-8, focusing on basic math concepts through interactive games.

Implementation: The system utilizes a pipeline of Automatic Speech Recognition (ASR) and Natural Language Understanding (NLU) components to facilitate interactions during math games.

Outcomes: Improved engagement and understanding of basic math concepts among young learners, leveraging play-based learning strategies.

Challenges: Recognition accuracy of children's speech in noisy environments, and the complexity of understanding diverse utterances.

Implementation Barriers

Technical Barrier

Challenges in automatic speech recognition (ASR) for children's speech, leading to higher error rates compared to adults.

Proposed Solutions: Utilization of advanced ASR models like Whisper, tuning models on children's speech data, and incorporating multimodal inputs (visual and physical interactions).

Socioeconomic Barrier

Limited access to technology and resources for low-income families to effectively utilize the at-home learning system.

Proposed Solutions: Simplifying setup requirements for home use, focusing on cost-effective solutions, and potential community support initiatives.

Data Scarcity

Small dataset size collected from a limited number of children, affecting the robustness of the system.

Proposed Solutions: Implementing transfer learning techniques and data augmentation strategies to enhance model training.

Project Team

Eda Okur

Researcher

Roddy Fuentes Alba

Researcher

Saurav Sahay

Researcher

Lama Nachman

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Eda Okur, Roddy Fuentes Alba, Saurav Sahay, Lama Nachman

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies