EMNLP 2026 BoF on Embeddings, Rerankers and Small LMs for Better Search

Our Birds of a Feather on embeddings, rerankers and small LMs is back for a 4th year. Sunday, October 25, 2026, 11:00–12:30, Room G9-10 & G11, HUNGEXPO Budapest, at EMNLP 2026. Same format as before: a 90-minute mini-workshop of 10-minute lightning talks from people who actually build and train retrieval models, then an open floor. Open to every EMNLP attendee, no registration needed to walk in. If you want a talk slot or a seat on the panel, fill in the form below.

Event Details

• Date: Sunday, October 25, 2026
• Time: 11:00–12:30 (Budapest time, CET. Note: Hungary leaves summer time that same morning)
• Location: Room G9-10 & G11, HUNGEXPO Budapest
• Format: In-person, 10-minute lightning talks, hard stop, then open discussion
• Organizers: EMNLP 2026 and Jina AI by Elastic

The slot is tentative until the EMNLP 2026 program on Underline is final. If it moves, we'll update this page and email everyone who registered. Add to Google Calendar.

Register

Attending only? Registration is optional, but it tells us how many chairs to fight for. Want to speak? Tell us your talk title and a link (paper, arXiv, poster or repo). We pick by fit and freshness and confirm by email before the session.

Form not loading? Open it in a new tab.

Topics

Anything that makes search better with models. This year we especially want to hear about:

• Embedding and reranker architectures: encoder vs. decoder, LLM-based
• Small language models for search and document understanding
• Omni and multimodal retrieval: text, image, audio, video
• Visual document retrieval and OCR-free pipelines
• Single vs. multi-vector, late interaction
• Multilingual and low-resource retrieval
• Reasoning-intensive retrieval and reasoning rerankers
• Agentic search, deep research, retrieval for agents
• Test-time compute for search
• Synthetic data, hard negatives, distillation
• Matryoshka, quantization, binary embeddings
• Sparse and hybrid retrieval, on-device retrieval models
• Benchmarks and evaluation: MTEB, RTEB, BRIGHT and friends

Three Years, Three Full Rooms

Singapore 2023. The first one. We expected a small round table and got 80+ people, five lightning talks and a panel discussion.

Miami 2024. We added rerankers and small LMs to the title and moved into the Miami Lecture Hall. So many talk submissions that the panel never happened: the lightning talks ate the whole 90 minutes.

Suzhou 2025. Full room again, right after Jina AI joined Elastic. Talks went from jina-embeddings-v4 to interpretable text embeddings to cross-lingual hard negatives.

Budapest is round four. Bring your newest results, especially the ones that didn't work. See you on October 25.