Create an automated tool to generate human-readable natural language descriptions for attribution graphs in interpretability research. This replaces manual inspection with model-driven insight.