Analogical Similarity

A 323-item BIG-bench task that classifies how two short event sentences share objects, relations, and structure, using Gentner-style similarity labels.

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 seven-way analogical similarity of sentence episodes (323 items)
Page statusunknown
Metricmultiple_choice_grade
Directionhigher_is_better
Unit%
Dataset size323
Dataset licenceApache-2.0
PublisherGoogle (BIG-bench collaboration)

What it measures

analogical_similarity gives two short English "episodes" (nested cause/effect event sentences) and asks which of seven similarity classes best describes the pair: literal similarity, analogy, cross mapping, surface similarity, false analogy, only-objects similarity, or no similarity. Authors Cameron Diao and Denis Kleyko built the items from Plate's holographic-reduced- representation scheme so that object attributes, first-order relations, and higher-order structure can be present or absent independently. The skill is structured analogy, not SAT word analogies.

Task format

Seven-option multiple choice, preferred metric multiple_choice_grade. task_prefix explains the rating scheme and gives a literal-similarity example. append_choices_to_input is false. Keywords include many-shot. Canary GUID embedded.

Models reporting this benchmark

No model card in ModelSpec reports this benchmark yet.

Data

This page as JSON · Edit on GitHub