Text this: Symbolic Rule Extraction From Attention-Guided Sparse Representations in Vision Transformers.