Skip to main content Skip to navigation

The ADAIO System at the BEA-2023 Shared Task on Generating AI Teacher Responses in Educational Dialogues

Project Overview

The document explores the application of generative AI in education through the ADAIO system's involvement in the BEA 2023 Shared Task, which emphasizes the generation of AI-driven teacher responses during educational dialogues. Utilizing OpenAI's GPT-3 model, specifically the text-davinci-003 variant, the system demonstrates an ability to produce meaningful and contextually appropriate responses in teacher-student interactions, highlighting the potential of generative AI to enhance educational experiences. The findings suggest that such AI models can effectively emulate teacher-like responses, offering promising implications for personalized learning and support. However, the document also addresses notable challenges, including the need for robust data acquisition, comprehension of diverse student learning styles, and the evaluation of the quality of generated responses, indicating that while generative AI holds significant promise in education, careful consideration and ongoing refinement are essential for its effective implementation and integration into educational practices.

Key Applications

ADAIO System using OpenAI GPT-3 for generating AI teacher responses

Context: Used in educational dialogues to enhance teacher-student interactions, particularly for ESL learners.

Implementation: The system implemented various approaches like few-shot and fine-tuning using the TSCC dataset to generate responses.

Outcomes: Achieved second place in the BEA 2023 Shared Task, demonstrating effective dialogue generation with AI teachers.

Challenges: Challenges include acquiring appropriate training data, understanding long-range dependencies in conversations, and evaluating the quality of responses.

Implementation Barriers

Data Acquisition

Difficulty in acquiring appropriate real teacher-student interaction data for training models.

Proposed Solutions: Develop strategies for collecting diverse educational dialogue data.

Model Development

Creating models that can effectively understand and adapt to different student learning styles and long-range conversational contexts.

Proposed Solutions: Research and develop models that capture varied learning styles and maintain context within dialogues.

Response Evaluation

Evaluating the quality of AI-generated responses to ensure they are helpful and contextually relevant.

Proposed Solutions: Implement comprehensive evaluation methods, including human assessments alongside automated metrics.

Project Team

Adaeze Adigwe

Researcher

Zheng Yuan

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Adaeze Adigwe, Zheng Yuan

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