Drift Inspector

Measuring and exploring scientific drift with Atomic Contribution Claims

An LLM distills every abstract into atomic contribution claims — what a paper actually adds, stripped of motivation and boilerplate. Claims are embedded, clustered across years, and served as an interactive map where every field-level trend can be traced down to the individual claims and papers behind it.

Live demo

*ACL main conferences 2018–2026

Six major venues in one joint claim space — ACL, EMNLP, NAACL, EACL, COLING, AACL — with cross-venue comparison.

69,950 claims · 88 clusters · 15,457 papers · 6 venues
Paper case study

EMNLP 2020–2025

The single-venue, year-balanced corpus analysed in depth in the EMNLP 2026 demo paper.

16,576 claims · 80 clusters · 4,488 papers
Dataset · Hugging Face

Full ACL Anthology

Every abstract in the Anthology, 1963–2026, extracted into atomic claims. Released as a dataset on Hugging Face.

346,010 claims · 80,144 papers · 423 venues