Analytic Entailment

A 70-item BIG-bench task that asks whether a second English sentence follows from the first by meaning alone, not by world facts.

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-way analytic vs non-analytic sentence entailment (70 items)
Page statusunknown
Metricmultiple_choice_grade
Directionhigher_is_better
Unit%
Dataset size70
Dataset licenceApache-2.0
PublisherGoogle (BIG-bench collaboration)

What it measures

analytic_entailment presents two English sentences and asks whether the second is analytically entailed by the first: it must be true given what the words mean, not given extra empirical facts. Authors Rachel Etta Rudolph, Ethan Jerzak, and Alexander W. Kocurek wrote the pairs to separate meaning from regularity (a king is a man; a US president need not be). Logical words such as negation appear, so dictionary lookup of one noun is not enough. It is not NLI from MultiNLI or [anli](anli.md).

Task format

Two-option multiple choice (entailment vs no-entailment), preferred metric multiple_choice_grade. task_prefix asks whether the pair embodies an entailment relation. example_input_prefix is "Sentences:"; example_output_prefix is "Relation:". 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