EMNLP 2026 edition · every accepted paper, counted and compared
EMNLP grew 59% to 5,540 papers in its main, Findings, industry and demo tracks. Some topics grew much faster.
We grouped every accepted EMNLP 2026 paper, and every EMNLP 2025 paper from the ACL Anthology, into the same 91 topics and compared their share of the conference. The area that gained most was agents and tool use (187 → 627 papers); the single topic that gained most was agent benchmarks and environments: 51 papers in 2025, 180 in 2026. Next came agent memory and long-horizon tasks (0.1% → 1.9%) and RL for reasoning (RLVR, GRPO) (1.0% → 2.4%). Shares, not raw counts, are compared throughout, because the whole conference grew.
91% of this year's papers build on, train, prompt or study large language models, up from 88% in 2025. 3,499 of the 5,540 papers (63%) already have a public version on arXiv; those are linked throughout this edition.
AI-assisted. Topics, languages and contribution types were assigned by an AI model from each paper's title and, where available, its arXiv abstract. They are not checked by hand and can be wrong; read the paper before citing it.
First edition. Built from the official presenter schedule and arXiv (abstracts and full texts) before the ACL Anthology proceedings are out. A revised edition covering the papers without a preprint follows once the proceedings are published.
Free for signed-in readers
Eight more pages: the topics that rose and fell since 2025, words in the titles, which languages EMNLP studies, every oral by session, all 91 topics with their papers, and a Budapest guide. Sign in with a free scime account to keep reading.
Already signed in? Reload this page. Accounts are free; we only use your email to sign you in and, if you agree, to tell you about new editions.
How many papers. EMNLP announced 5,252 accepted main-conference and Findings papers. The presenter schedule lists 2,705 main and 2,484 Findings papers after withdrawn ones are removed, plus 265 industry, 86 demo and 63 journal papers; this edition counts what the schedule lists.
Sources. EMNLP 2026: the conference's public presenter schedule (titles, tracks, presentation types, sessions) and arXiv (abstracts and links). EMNLP 2025: the ACL Anthology. Conference details: 2026.emnlp.org.
Method, in short. An AI model placed each paper in one of 91 topics in 16 areas, and noted the languages it studies, whether it works with large language models and its main contribution. The same topics are used for both years so shares compare fairly. Topics, languages and contributions are AI-assigned and not checked by hand; they can be wrong.
Revised edition. When the ACL Anthology publishes the EMNLP 2026 proceedings, we will relabel the papers that have no preprint yet from their abstracts and add their full texts. Corrections: contact@latenticlabs.com
Independent. Published by Latentic Labs with scime atlas. Not affiliated with or endorsed by EMNLP, the ACL or SIGDAT.
← All editions · © 2026 Latentic Labs