Question Selection (BIG-bench)

BIG-bench task: given an English short answer and its paragraph, choose which paraphrased question that answer actually satisfies.

Also known as: question selection

unassessed

This page is a discovery lead. Nobody has yet assessed it against the catalogue contract, so it carries no disposition. Absence of evidence here is not evidence of staleness.
Categoryreasoning
SubcategoryBIG-bench multiple-choice: pick the paraphrased SQuAD question that matches a short answer in context
Page statusunknown
Metricmultiple_choice_grade
Directionhigher_is_better
Unit%
Dataset size1590
Dataset licenceApache-2.0
PublisherBIG-bench collaboration (Google); task author Kaustubh Dhole

What it measures

question_selection gives a short English answer and the paragraph it came from, then asks which of several paraphrased questions that answer belongs to. Gold questions are paraphrases of SQuAD evaluation items. Negatives are other questions from the same paragraph that share question type and do not overlap in answer span. The skill is matching an answer to an interrogative, not span extraction. It is not [squad](squad.md) and is not in [bbh](bbh.md).

Task format

JSON multiple choice. Preferred metric multiple_choice_grade. task_prefix asks the model to pick the question whose answer is the given short answer. Choice count varies (2–10). Canary GUID embedded. Dummy-model header: 1582 multiple-choice queries.

Models reporting this benchmark

No model card in ModelSpec reports this benchmark yet.

Data

This page as JSON · Edit on GitHub