SyntaxShap: Syntax-aware Explainability Method for Text Generation

arXiv:2402.09259v1 Announce Type: new
Abstract: To harness the power of large language models in safety-critical domains we need to ensure the explainability of their predictions. However, despite the significant attention to model interpretability, there remains an unexplored domain in explaining sequence-to-sequence tasks using methods tailored for textual data. This paper introduces SyntaxShap, a local, model-agnostic explainability method for text generation that takes into consideration the syntax in the text data. The presented work extends Shapley values to account for parsing-based …

abstract arxiv attention cs.ai cs.cl data domain domains explainability harness interpretability language language models large language large language models model interpretability paper power predictions safety safety-critical syntax tasks text text generation textual type