When Two Conferences Are One

Published September 19, 2026 · paper counts are current, not as-published

A conference is usually treated as a container: a paper is in one, and counting papers means counting containers. That breaks in a specific and recurring way - some conferences run as a single event in some years, and their proceedings are one set of papers filed under two names.

The Same Work, Filed Twice

6,745 papers in this corpus are marked as a second filing of a paper already counted - 2.9% of everything indexed. They are not scattered: they concentrate in 11 venue pairs, and the largest is ACL and IJCNLP with 2,246 shared papers across 2009, 2015, 2021.

The years are the tell. If this were an indexing fault the overlaps would be spread thinly across every year a pair of venues both ran. Instead each pair names particular years, which is what a jointly held edition looks like from the outside: one call for papers, one programme, two names on the proceedings.

Why Every Figure Here Says "Canonical"

This is the reason the counts on this site are taken over canonical papers rather than rows. Counting rows would inflate the affected venues, and it would inflate them unevenly - a venue that has never held a joint edition is unaffected, so the error would not even be a constant factor you could correct for. Every number on the conferences directory and in the Insights articles excludes second filings.

What This Does Not Show

• Not every joint edition. A pair is visible here only when the same paper was indexed under both names. An event held jointly but published under one banner leaves no trace in this data.
• Not which filing is "the" paper. One row is marked as the duplicate of the other, and that choice is an artefact of ingestion order, not a statement about where the work belongs.
• Not a judgement about the venues. Holding a joint edition is ordinary practice. The only problem it creates is for whoever is counting.

The numbers here are read from the database when you load the page.