Linguistic Mappings

A BIG-bench task that maps lemmas and sentences through tense, plural, questions, negation, and pronouns from a handful of shots.

Also known as: BIG-bench linguistic_mappings

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 few-shot English/German morphological and syntactic mappings
Page statusunknown
Metricexact_str_match
Directionhigher_is_better
Dataset size15527
Dataset licenceApache-2.0
PublisherGoogle (BIG-bench collaboration)

What it measures

linguistic_mappings asks whether a model can apply a grammatical mapping after a few examples, rather than retrieve a memorised surface form. Morphology covers English and German past tense and English plurals, using CELEX-2 lemmas in Berko-style carrier sentences. Syntax covers English polar questions, main-verb negation, and pronoun or reflexive replacement, from a hand-written CFG. JSON subtasks mix items; programmatic subtasks also cross in-domain versus out-of-domain shots (regularity, number and tense, or name gender) at 0–32 shots (0–16 for plural).

Task format

Free-text generation. Preferred metric exact_str_match on every JSON task.json opened here. Parent task.json holds name and keywords only. Canary GUID embedded. Not multiple choice.

Models reporting this benchmark

No model card in ModelSpec reports this benchmark yet.

Data

This page as JSON · Edit on GitHub