About SenseBench

Purpose

SenseBench evaluates English word sense disambiguation with auditable LLM run artifacts.

Each model receives a target word in context and candidate WordNet senses, then returns the chosen sense index.

Design

The leaderboard is static, reproducible, and rebuilt from verified submissions.

Scores are recomputed from predictions and checked against the registered dataset and prompt.

Licensing

Code Apache-2.0; lexEN data CC BY-NC 4.0 (inherited from Maru et al. 2022), research and non-commercial evaluation only.

The software and the dataset are licensed separately. The SenseBench package, runner, verifier and site builder are Apache-2.0 and may be used commercially.

The lexEN dataset derives from the Maru et al. 2022 benchmark, which is distributed under CC BY-NC 4.0, so lexEN inherits that restriction and is not ours to relicense. Submitted run artifacts are covered by the repository's Apache-2.0 licence.