Cause and Effect

A 51-pair BIG-bench task that asks which of two English events caused the other, scored as two-way multiple choice under three prompt formats.

Also known as: BIG-bench cause_and_effect

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 two-event causal commonsense (51 pairs, three promptings)
Page statusunknown
Metricmultiple_choice_grade
Directionhigher_is_better
Unit%
Dataset size153
Dataset licenceApache-2.0
PublisherGoogle (BIG-bench collaboration)

What it measures

cause_and_effect gives two short English events and asks which one caused the other. Guy Gur-Ari and Neta Krakover wrote about fifty such pairs (51 in each JSON file). Unlike SuperGLUE COPA, there is no separate premise: the model sees only the two events. The skill is everyday causal direction, not COPA's three-sentence plausibility setup and not BBH causal_judgement.

Task format

Two-option multiple choice, preferred metric multiple_choice_grade. Three subtasks share the same pairs: two_sentences (pick the causing event), one_sentence (pick the more likely "because" sentence), and one_sentence_no_prompt (compare the two "because" strings with an empty context; append_choices_to_input is false). Canary GUID embedded.

Models reporting this benchmark

No model card in ModelSpec reports this benchmark yet.

Data

This page as JSON · Edit on GitHub