What does "Rationale Quality" mean?
Table of Contents
Rationale quality refers to how well explanations can support decisions made by models. When models provide clear and accurate reasons for their choices, it helps people understand their actions better.
Importance of Readability
The way these explanations are written is crucial. If they are too complex, people may not grasp the points being made. Therefore, controlling the readability level is important. This means tailoring explanations to suit different audiences, such as children or college students. However, sometimes the difficulty of the text does not match the intended audience.
Small Models Supporting Large Models
Using smaller language models can help larger models improve their reasoning. A lightweight model can create simple and clear reasons for a decision. Then, a bigger model can use these reasons to make its predictions. This approach saves resources, as it focuses on training the smaller model.
Evaluation of Rationale Quality
When evaluating rationale quality, it's important to check how well these explanations work across different complexity levels. Feedback shows that explanations generally receive a good response, especially at a high school reading level. Higher-quality explanations lead to better answers in question-and-answer tasks.