SelfElicit: Your Language Model Secretly Knows Where is the Relevant Evidence
This addresses the issue of LMs failing to focus on relevant evidence in real-world scenarios with noise, offering a computationally efficient solution without additional training.
The paper tackles the problem of language models struggling to utilize key evidence in noisy contexts by proposing SelfElicit, an inference-time method that uses attention scores to highlight relevant evidence, resulting in consistent and significant improvements on multiple evidence-based QA tasks for various LM families.
Providing Language Models (LMs) with relevant evidence in the context (either via retrieval or user-provided) can significantly improve their ability to provide better-grounded responses. However, recent studies have found that LMs often struggle to fully comprehend and utilize key evidence from the context, especially when it contains noise and irrelevant information, an issue common in real-world scenarios. To address this, we propose SelfElicit, an inference-time approach that helps LMs focus on key contextual evidence through self-guided explicit highlighting. By leveraging the inherent evidence-finding capabilities of LMs using the attention scores of deeper layers, our method automatically identifies and emphasizes key evidence within the input context, facilitating more accurate and grounded responses without additional training or iterative prompting. We demonstrate that SelfElicit brings consistent and significant improvement on multiple evidence-based QA tasks for various LM families while maintaining computational efficiency. Our code and documentation are available at https://github.com/ZhiningLiu1998/SelfElicit.