Context Definition Alignment

A programmatic BIG-bench task that matches masked SemCor contexts to WordNet definitions across 3,184 CoDA synset groups.

Also known as: CoDA, BIG-bench context_definition_alignment

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 programmatic alignment of masked SemCor contexts with WordNet definitions (3,184 groups)
Page statusunknown
Metricalignment_score
Directionhigher_is_better
Dataset size3184
Dataset licenceApache-2.0
PublisherGoogle (BIG-bench collaboration)

What it measures

context_definition_alignment hides a noun or verb behind a nonce word (default bkatuhla) in a SemCor sentence and asks the model to pair that context with the matching WordNet 3.0 definition, one-to-one inside a synset group of size 5-10. Authors: Lütfi Kerem Şenel, Timo Schick, and Hinrich Schütze. English text. Programmatic task.py, not a JSON item list. Not ordinary word-sense disambiguation with the lemma given.

Task format

For each group the harness scores every context-definition pair by conditional log-probability of the definition after a pattern, then searches permutations for the best total. Preferred metric alignment_score (fraction of definitions matched). Four CoDA JSON files times nouns and verbs yield eight subtasks. Canary GUID embedded.

Models reporting this benchmark

No model card in ModelSpec reports this benchmark yet.

Data

This page as JSON · Edit on GitHub