Clarifying Insertions from Revision Edits (CLAIRE)

Dataset

Talita Anthonio, Anna Sauer, Michael Roth

Dataset for SemEval 2022 Task 7: Identifying Plausible Clarifications of Implicit and Underspecified Phrases in Instructional Texts

The goal of the SemEval task was to distinguish between plausible and implausible revisions for unclear passages in instructions.

The dataset contains sentences from instructions published on wikiHow,  a platform for collaboratively edited how-to guides. Each sentence is associated to five possible revisions that can potentially add information that was previously implicit or underspecified and can thus serve as clarifications for the text.

Each of the five possible revisions has been annotated with how plausible the respective insertion is in the given context. These annotations are provided both as a continuous score on a 5-point Likert scale and as a class label (implausible, neutral or plausible).

https://github.com/acidAnn/claire