SyntaxShap datasets

This project contains the three datasets used in the paper "Syntax-aware Explainability Method for Text Generation".- Inconsistent Negation Dataset: contains disjoint sentence pairs, i.e., a sentence and its negated version, whereby the last token in the sentence is masked with the [MASK] token;- Generics KB: contains generic sentences, which are high-quality, semantically complete statements;- ROCStories Winter2017: contains a collection of five-sentence everyday life stories. More information on the preprocessing and filtering of the data can be found in the paper.