MathKingdom: Teaching Children Mathematical Language Through Speaking at Home via a Voice-Guided Game

Mathemyths: Leveraging Large Language Models to Teach Mathematical Language through Child-AI Co-Creative Storytelling

“…Previous research has identified two prevalent types of interactions that children tend to have with CUIs [101]: voice assistants supporting open-domain conversations (e.g., Apple Siri, Amazon Alexa, Google Assistant) and voice-based apps that specifically designed for children for domain-specific conversations [8,9,56,62,63]. This type of specifically design apps have been found to support children's learning across a number of disciplines, including math [40,98], science [102], computational thinking [18], and literacy [101]. Though varied in their specific design, these applications are grounded in a common principle by utilizing conversational AI as a language partner to engage children in discussions related to the relevant topics.…”

“…It emphasizes the opportunities for children to not only be exposed to but also generate math talk themselves [2,83]. Interestingly, these studies have revealed that the children's own mathematical conversations play a crucial, if not more significant, role in shaping their mathematical competence, sometimes even outweighing the influence of the math talk provided by parents or teachers [90,98].…”

“…The study revealed that parents found the robot's questions inspiring and that these questions led to more meaningful conversations with their children. Another example is a voice-based game designed to teach children aged four to seven [98]. In this game, children interacted with visual representations of shapes in different colors and sizes, receiving prompts such as "turn the tallest triangle into a circle. "…”

ChatScratch: An AI-Augmented System Toward Autonomous Visual Programming Learning for Children Aged 6-12

“…Despite these ideals, previous AI assistants designed for children have often fallen short. Some systems [78] rely on digital sensors to perceive real world information but are limited to rudimentary interactions, and other approaches have relied on predefined rules to generate dialogue, which often led to logical inconsistencies [84]. Large language models (LLMs), however, present a compelling alternative.…”