Attention with Dependency Parsing Augmentation for Fine-Grained Attribution
This work addresses the need for efficient validation of RAG-generated content by improving fine-grained attribution mechanisms, though it is incremental as it builds on existing model-internals-based methods.
The paper tackled the problem of fine-grained attribution for RAG-generated content by proposing a method that aggregates token-wise evidence and integrates dependency parsing, which consistently outperformed all prior works.
To assist humans in efficiently validating RAG-generated content, developing a fine-grained attribution mechanism that provides supporting evidence from retrieved documents for every answer span is essential. Existing fine-grained attribution methods rely on model-internal similarity metrics between responses and documents, such as saliency scores and hidden state similarity. However, these approaches suffer from either high computational complexity or coarse-grained representations. Additionally, a common problem shared by the previous works is their reliance on decoder-only Transformers, limiting their ability to incorporate contextual information after the target span. To address the above problems, we propose two techniques applicable to all model-internals-based methods. First, we aggregate token-wise evidence through set union operations, preserving the granularity of representations. Second, we enhance the attributor by integrating dependency parsing to enrich the semantic completeness of target spans. For practical implementation, our approach employs attention weights as the similarity metric. Experimental results demonstrate that the proposed method consistently outperforms all prior works.