commonsense-part-whole

Dataset for the EMNLP 2019 paper

Do Nuclear Submarines Have Nuclear Captains? A Challenge Dataset for Commonsense Reasoning over Adjectives and Objects

James Mullenbach, Jonathan Gordon, Nanyun Peng and Jonathan May

The dataset consists of part-whole relations along with an adjective describing the whole. The task is to infer the relationship between the adjectives and the parts.

• data/annotated contains the dataset in lexical format (i.e. just the whole, part, and adjective words), and in NLI format (with retrieved premises and templated hypotheses).
• mturk/ contains the prompt templates and scripts used to create the AMT tasks.