The 1st Late Interaction Workshop (LIR) @ ECIR 2026
A forum for researchers and practitioners to discuss late interaction (a.k.a. multi-vector, or ColBERT) retrieval methods, as well as their limitations and applications.
About the Workshop
Creating a highly-interactive environment for researchers and practitioners to discuss all aspects of late interaction retrieval
Late interaction retrieval methods, pioneered by ColBERT, have emerged as a powerful alternative to single-vector neural IR. By leveraging fine-grained, token-level representations, they have been demonstrated to deliver strong generalisation and robustness, particularly in out-of-domain settings. They have recently been shown to be particularly well-suited for novel use cases, such as reasoning-based or cross-modality retrieval.
At the same time, these models pose significant challenges of efficiency, usability, and integrations into fully fledged systems; as well as the natural difficulties encountered while researching novel application domains. Recent years have seen rapid advances across many of these areas, but research efforts remain fragmented across communities and frequently exclude practitioners.
We intend to make LIR the venue where all actors of the late-interaction research ecosystem can freely gather and share ideas. To this aim, we have planned for the workshop to be highly interactive and are encouraging a large variety of publication formats, from short, early technical reports on specific phenomenon to full-fledged research papers.
Workshop Details
Date: April 02 2026 (ECIR 2026 Workshop Day)
Location: Delft, the Netherlands
Format: Half-day workshop (09:00-13:00) with keynote, paper sessions, demos, and roundtable discussion
Mode: Hybrid for all sessions except roundtable discussion
Keynote Speaker
We're honored to have the creator of ColBERT as our keynote speaker
Omar Khattab
Assistant Professor, Massachusetts Institute of Technology (MIT)
Talk Title
TBD
Abstract
TBD
Bio
Omar Khattab is an Assistant Professor at MIT whose research interests span information retrieval and the creation of end-to-end AI systems centered on Large Language Models via the DSPy project. Omar's research pioneered modern multi-vector retrieval with the original ColBERT approach, ColBERTv2, and the de facto standard PLAID indexing method.
Workshop Schedule (Tentative)
Call for Papers
We invite submissions on all aspects of late interaction retrieval and multi-vector methods
Topics of Interest
Late-Interaction Training Recipes
Studies on the impact of different scheduling, data mixes, training losses, and how they specifically impact late-interaction models.
Theoretical Understanding of Late Interaction
Work building on understanding Chamfer similarity, which the MaxSim operator effectively calculates, and designing approximations with strong guarantees.
Analysis of Late-Interaction-Specific Mechanisms
Understanding tricks that have empirically improved performance but whose actual mechanisms are not fully understood, such as [MASK] token usage for query augmentation.
Multi-Modal Late Interaction
Exploring late-interaction models for multi-modal retrieval, including work on ColPali, VideoColBERT, and other modality-specific approaches.
Alleviating Efficiency Concerns
Addressing efficiency issues stemming from storing considerable numbers of vectors. We encourage contributions highlighting limits of current methods and novel extensions.
Usability
Improving compatibility with software and indexing stacks, including work on toolkits like PyTerrier, PyLate, RAGatouille, and novel scoring/indexing methods.
Nascent Applications
Exploring long-context retrieval, Agentic Search, Deep Research tasks, and reasoning-based retrieval where models use reasoning LLM's thinking traces for fine-grained retrieval.
Others
Anything else related to late interaction, even if not listed above.
Submission Guidelines
All papers should use the CEUR-WS template. The template exists on Overleaf, as well as in offline LaTeX format and LibreOffice format.
We invite submissions in multiple forms: fully-fledged research papers, position papers, demo or technical reports, and opinion papers. We also welcome ongoing work and strongly encourage sharing both early and negative results.
We accept three submission formats:
- •Notes (1-to-3 pages): Short, hyper-focused technical reports describing narrow results, positive or negative, or presenting ongoing work. All notes will be presented as posters.
- •Papers (short: up to 6 pages; full-length: up to 12 pages): To be presented either orally or as a poster, as recommended by the program committee.
- •Tech Reports (2-12 pages): Describing model, systems or software, their design system and presenting insights from the development. To be presented as demos or posters.
All submissions will be reviewed by program committee members for relevance to the workshop and the community at large. Papers will be submitted for archival by default, with proceedings published in CEUR-WS. Authors may request non-archival.
Proceedings shall be submitted to CEUR-WS.org for online publication.
Important Dates
Workshop Organizers
Benjamin Clavié
Researcher at Mixedbread & PhD student at NII (Japan)
Xianming Li
Researcher at Mixedbread & PhD student at Hong Kong PolyU
Antoine Chaffin
Researcher at LightOn
Omar Khattab
Assistant Professor at Massachusetts Institute of Technology (MIT)
Manuel Faysse
PhD Student at CentraleSupélec
Tom Aarsen
Research Engineer at Hugging Face
Jing Li
Associate Professor at Hong Kong Polytechnic University
Program Committee
TBD