Skip to main content Skip to navigation

ICLEF: In-Context Learning with Expert Feedback for Explainable Style Transfer

Project Overview

The document explores the application of generative AI in education through the ICLEF framework, which merges in-context learning with expert feedback to improve explainable style transfer tasks. This innovative approach generates high-quality datasets tailored for both formal and informal writing, while also emphasizing the importance of model explainability and enhancing the quality of text generation through expert annotations. Notably, the findings reveal that student models can surpass teacher models in certain tasks, highlighting the potential of generative AI to transform educational practices. Overall, the integration of generative AI within this framework not only advances the capabilities of text generation but also fosters a deeper understanding of model behavior, ultimately contributing to more effective learning outcomes in educational settings.

Key Applications

ICLEF - In-Context Learning with Expert Feedback for Explainable Style Transfer

Context: Educational context focusing on style transfer tasks for students and researchers in computational linguistics.

Implementation: Utilizes large language models (LLMs) to generate datasets enhanced by expert feedback, improving the quality of style transfer outputs.

Outcomes: Student models trained on the generated datasets outperform generalist teacher models, achieving higher accuracy in style transfer tasks.

Challenges: Gathering expert feedback is costly and limited, which can affect the quality of the training data.

Implementation Barriers

Resource Limitation

Expert feedback is expensive and difficult to obtain, limiting the dataset quality.

Proposed Solutions: Incorporate in-context learning to maximize the utility of limited expert feedback.

Model Limitations

Large language models may generate inaccurate or biased outputs that require correction.

Proposed Solutions: Implement self-critique mechanisms and expert annotations to refine model outputs.

Project Team

Arkadiy Saakyan

Researcher

Smaranda Muresan

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Arkadiy Saakyan, Smaranda Muresan

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