Skip to main content Skip to navigation

Data Augmentation with Paraphrase Generation and Entity Extraction for Multimodal Dialogue System

Project Overview

The document explores the implementation of generative AI in education, specifically through the development of a multimodal dialogue system that supports young learners, aged 5-8, in grasping fundamental math concepts via interactive, game-based learning. Emphasizing the enhancement of the Natural Language Understanding (NLU) component, the system employs data augmentation strategies, such as paraphrase generation, to improve responsiveness and engagement. By integrating physical elements and multimodal inputs, the approach aims to create a more immersive and effective educational experience. The findings suggest that such interactive systems can significantly bolster children's understanding of math, making learning both enjoyable and effective while showcasing the potential of AI technologies in transforming educational methodologies.

Key Applications

Multimodal Dialogue System for early childhood education (Kid Space project)

Context: Game-based learning for children aged 5-8, focusing on basic math concepts.

Implementation: The system uses spoken dialogue systems (SDS) that process user utterances and multimodal inputs (visual, audio, gestures) to facilitate interactions.

Outcomes: Improved NLU performance for intent recognition from user utterances; engagement and interactive learning experiences for children.

Challenges: Limited datasets for training NLU models; need for robust interaction handling in a multimodal environment.

Implementation Barriers

Technical Barrier

Limited datasets for training the Natural Language Understanding models. Challenges in ensuring accurate understanding of multimodal inputs (e.g., gestures, speech) for effective dialogue management.

Proposed Solutions: Data augmentation techniques such as paraphrase generation to create more training data. Iterative training and evaluation of NLU models; integrating various sensing technologies for better input recognition.

Project Team

Eda Okur

Researcher

Saurav Sahay

Researcher

Lama Nachman

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Eda Okur, 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