Birds of the Feather and Affinity Group Events

Sunday, October 25, 2026

Time: 11:00–12:30 (Session 2)
Location: G9-10, G11
Organizer: Han Xiao

Description:
Embeddings and rerankers are the backbone of agentic search. These models are smaller and more discriminative than their large generative counterparts, but getting them to perform well still takes real expertise. After our Embeddings BoF at EMNLP 2023, 2024 and 2025, we’ve watched omnimodal embeddings, rerankers and general-purpose discriminative models like Jev take off. Come hear invited talks on what’s new, debate where the research goes next, and talk through advance capabilities of those small models.

This is our fourth BoF on the same topic. Singapore, Miami, Suzhou: full house every time. With the hype around Jev, a general-purpose discriminative model, we expect even bigger turnout this year, and a wider conversation covering embeddings, rerankers and uses well beyond traditional search.

Culturally Grounded and Trustworthy NLP: Perspectives from Islamic Media Ethics

Time: 16:00–17:30 (Session 4)
Location: G9-10
Organizer: Firoj Alam

Description:
How should NLP systems respond when reliable information, cultural context, and ethical judgment intersect? This Birds-of-a-Feather session brings together researchers and practitioners interested in culturally grounded NLP, agentic AI, misinformation, trustworthy question answering, and responsible AI, drawing on perspectives from Islamic media ethics. The discussion will examine how agentic approaches, including retrieval-augmented generation and tool-assisted verification, can support evidence-grounded responses. We will explore their opportunities and limitations, particularly in handling insufficient evidence, conflicting sources, source attribution, and human oversight. We will also consider how ethical principles, including truthfulness, responsibility, justice, and avoidance of harm, can inform dataset design, model development, and evaluation. Discussions will distinguish factual verification from ethical and interpretive judgments, while recognizing diversity within and across religious and cultural traditions. Participants will explore three questions: When do agentic approaches improve the reliability of culturally grounded responses? How should systems represent uncertainty and differing interpretations? How can researchers, domain scholars, and communities jointly develop meaningful benchmarks and evaluation practices? Our goal is to identify shared research priorities, resource needs, and opportunities for interdisciplinary collaboration. Participants from all backgrounds and all career stages are welcome.

This session aligns with EMNLP 2026’s theme, “New Missions for NLP Research,” by connecting culturally grounded evaluation, agentic AI, and responsible NLP. Drawing on Islamic media ethics, it explores evidence grounding, uncertainty, and diverse interpretations while encouraging collaboration among NLP researchers, domain scholars, and communities.

Doing NLP Research in Japan

Time: 16:00–17:30 (Session 4)
Location: G12
Organizer: Benjamin Heinzerling

Description:
Researchers currently working in Japan, researchers interested in working or studying in Japan, people collaborating with Japanese institutions, and anyone interested in strengthening connections between the Japanese and international NLP communities.

The EMNLP 2026 theme asks what NLP research should focus on now that strong general-purpose language models are widely available. Japan offers an interesting case: it has a large NLP community and substantial investment in language models, but is not currently home to a leading frontier-model developer. If building ever more capable general-purpose models is no longer the obvious mission of NLP, what should researchers in Japan focus on instead? This BoF will use Japan as a starting point for discussing new missions for NLP research.

Monday, October 26, 2026

Trustworthy Language Model Agents for High-Stakes Industry Applications

Time: 09:00–10:30 (Session 5)
Location: G9-10
Organizer: Yuqicheng Zhu

Description:
Language model agents are moving from demos into settings where a wrong action can be costly: suggesting a clinical intervention, modifying an industrial process, triggering an enterprise workflow, or making a high-impact recommendation. But what would it actually take to trust such an agent? The session will focus on two closely connected questions. First, what capabilities and safeguards are needed for trustworthy deployment in high-stakes applications? Second, which technical approaches are most promising for achieving these properties beyond controlled benchmark settings and in real-world deployment? We expect the session to foster an open exchange between researchers and practitioners from academia and industry, bringing together different perspectives on trustworthy language model agents in high-stakes applications. Beyond the session itself, we hope to identify shared challenges, stimulate new collaborations, and help build a community around trustworthy deployment of language model agents in real-world settings.

This session directly connects to the EMNLP 2026 theme, New Missions for NLP Research, by focusing on the transition from language models as benchmarked components to agentic systems deployed in consequential real-world settings. As language model agents increasingly interact with tools, environments, and human decision-making processes, progress can no longer be measured only through static benchmark performance. The session therefore emphasizes trustworthiness, robustness, and real-world deployment, and asks what new capabilities, safeguards, and technical approaches are needed for NLP systems operating in high-stakes industry applications.

ARR Editors-in-Chief Meet’n’Greet

Time: 11:00–12:30 (Session 6)
Location: G9-10
Organizer: Asad Sayeed

Description:
Informally meet some members of the Editor-in-Chief team at ACL Rolling Review and get a personal view of how the process is run along with the challenges the EiC team faces.

It’s about how papers actually got to the conference in the first place. I don’t know if 30 is a good estimate of attendance, I assume it’ll be pretty popular.

Finding Your Path: Honest Conversations on Early Career Paths in NLP

Time: 09:00–10:30 (Session 5)
Location: G12
Organizers: Esra Dönmez, Marlene Lutz, Matthias Orlikowski

Description:
We plan to host a BoF session on working in NLP that’s different from standard career counseling, focusing on impacts on a personal level and on our lives outside work. The session will use participatory discussion formats and a panel discussion with more senior members of the NLP community. At the end of the session, we expect all participants to have a better idea of each other’s struggles and strategies to navigate various trade-offs, fostering community around common challenges.

This session connects to the conference theme (New Missions for NLP Research) by creating space to reflect not only on where NLP research is headed, but also on how careers in the field are evolving with it. Honest conversations about changing roles, expectations, opportunities, and career paths can help the community think more broadly about who will shape the next chapter of NLP and how.

South East Asian NLP

Time: 11:00–12:30 (Session 6)
Location: G11
Organizers: SIGSEA (Samuel Cahyawijaya; Peerat)

Description:
Reshaping the Future of South East Asian NLP Research

Our initiative aims to reshaping the mission and vision of NLP research in South East Asian regions beyond the existing paradigms on addressing linguistics and cultural gaps in multilingual representation in order to achieve an equitable access to AI capabilities across region which aligns perfectly with EMNLP 2026’s theme on “What are the missions of NLP research now, and what kinds of contributions most advance that mission?”

NLP Against Human Trafficking: How the Research Community Can Help Close the Technology Gap

Time: 11:00–12:30 (Session 6)
Location: G12
Organizer: Hongliu Cao

Description:
Human trafficking is intensifying while traffickers rapidly exploit technology. Drawing on our recent discussion with UN, NGOs, Europol and related industry, this BoF brings researchers together to discuss urgent challenges, explore promising directions for NLP research and foster collaboration toward stronger prevention, earlier intervention and better protection, turning research into meaningful social impact.

Aligned with “New Missions for NLP Research,” this BoF connects practitioner challenges with responsible NLP research to bridge fragmented evidence, strengthen institutional collaboration and support earlier intervention.

Linguistic Style across the NLP pipeline

Time: 16:45–18:15 (Session 9)
Location: G12
Organizers: Anna Wegmann, Hongyu Chen, Sarubi Thillainathan, JongKyung Shin, Christoph Leiter, Steffen Eger, Adrian de Wynter

Description:
Style plays a central role across the full NLP pipeline: the stylistic diversity of training data influences model robustness, users expect generated texts in a helpful or personalized style, evaluation on texts with limited stylistic variation can obfuscate a model’s ability to function in the wild, and applications like writing assistance, AI-generated text detection, anonymization and authorship verification rely on style features. With this BoF session we want to connect researchers that are interested in how style can be modeled, measured and applied across the NLP pipeline. We will open the session with a 15-minute introduction to the topic and open questions, and will then encourage informal discussions around topics participants are interested in. They might cover style in language modeling and generation (personalization, style steering/transfer), measuring and explaining style, applications (authorship attribution and verification, AI-text detection, anonymization, forensic and literary analysis), and linguistic perspectives on style.

Style is central to New Missions for NLP Research: Language models encounter diverse registers, dialects and individual writing styles, and generate texts with stylistic prints of their own. Current benchmarks may capture only a narrow range of stylistic features or definitions of style, leaving open how to model and evaluate style across domains, tasks and social contexts. These questions are also relevant to applications such as writing assistance, personalization and chatbots, where style also depends on humans’ subjective preferences. Studying linguistic style thus contributes both to understanding human language and to developing language technologies that support diverse needs.

Data to Output: Diversity Across the NLP Pipeline

Time: 9:00–10:30 (Session 5)
Location: G11
Organizer: Cantao Su

Description:
Diversity is increasingly discussed across the NLP pipeline, yet many fundamental questions remain poorly understood. What do we mean by diversity, and how should we measure diversity in datasets and LLM outputs? How can we collect or select training data and in-context examples that capture greater diversity? When does diversity actually help and which kinds? At generation time, how can we encourage outputs that better reflect the variability of human language? When can synthetic data enhance diversity, and when might it undermine it? Despite growing interest in these questions, our understanding of the role of diversity in NLP is still in its beginning stages. Our main goal is to provide a platform for researchers interested in these topics to get to know each other and share insights and motivations. We plan to open the session with a 15-minute introduction to the topic and why it is important, ending with open questions. The introduction is followed by an informal discussion of these open questions. Participants are welcome to bring their own discussion topics. The session is organized by the DataDivers team (including Dong Nguyen, Anna Wegmann, Esther Ploeger and Cantao Su).

The New Missions for NLP Research theme asks, among others, how to measure progress beyond benchmark scores. The theme description mentions robustness and model deployment in diverse settings as potential new evaluation settings. The theme also directly asks about the consequences of synthetic data feedback loops. Diversity is central to such questions: The diversity of training data can affect how robust models are across settings, the diversity of model outputs determines the diversity of synthetic data used to train future models, and only diverse evaluation data can show whether models work across settings.

Tuesday, October 27, 2026

Efficient and Explainable Authorship Verification

Time: 09:00–10:30 (Session 10)
Location: G11
Organizers: Christoph Leiter, Steffen Eger

Description:
Authorship verification, the task of determining whether two texts were written by the same author, is relevant to applications such as forensic analysis, plagiarism detection, and the study of online and historical texts. Recent approaches increasingly rely on large language models, raising concerns around computational cost, privacy, and transparency. In this BoF, we want to discuss challenges and possible solutions towards more efficient and explainable authorship verification, while connecting researchers interested in the field. The minute session will include a short introduction, group discussions, and a final plenary discussion to identify common themes, open questions, and opportunities for collaboration.

The BoF connects with the EMNLP theme by reconsidering how progress in authorship verification should be measured beyond benchmark performance. By focusing on efficiency and explainability, its aim is to foster discussion around novel approaches and evaluation criteria for authorship verification.

Symbolic and Trustworthy Reasoning Across Language and Vision

Time: 09:00–10:30 (Session 10)
Location: G12
Organizer: Shreya Rajpal

Description:
AI models can produce correct answers even when the reasoning behind them is incomplete, unsupported, or unreliable. As these models are increasingly used in real-world and sensitive applications, an important question is not only whether an answer is correct, but whether it is supported by reasoning that can be trusted and verified. Symbolic and structured reasoning methods can help make this reasoning easier to inspect and verify. These methods include rules, constraints, programs, graphs, knowledge bases, and other explicit representations. This Birds-of-a-Feather session will explore how such approaches can support trustworthy reasoning across language, vision, and multimodal systems. It will also examine how their role and implications differ across modalities. We plan to invite senior researchers working in these areas to participate in a panel discussion and help identify open challenges and promising research directions. Through this discussion, we hope participants will gain concrete ideas for how symbolic and structured methods can be used to make reasoning more explicit, inspectable, and verifiable in their own work. We welcome researchers working on vision and language reasoning, neuro-symbolic methods, structured reasoning, trustworthiness, robustness, and related areas. We also welcome work in application domains such as spatial reasoning, legal reasoning, and scientific reasoning. The session will focus on common failure modes, evaluation challenges, and opportunities for building reasoning systems that are more reliable, interpretable, and verifiable.

The session connects directly to the EMNLP 2026 theme, “New Missions for NLP Research.” NLP is rapidly moving toward more capable models, with advances being made in scaling data, training, post-training, and reasoning methods. At the same time, increasingly complex systems use language model judgments to call tools, coordinate agents, and combine multiple models. Even evaluation is becoming increasingly automated through approaches such as LLM-as-a-Judge. As these systems are also being deployed in high-stakes and sensitive domains such as scientific and legal applications, it is important to ask not only how we can improve and scale their capabilities, but also how we can verify the reasoning behind their outputs and make it more trustworthy and reliable. One important direction for addressing these questions is the use of symbolic and structured reasoning methods to make model reasoning more explicit, inspectable, and verifiable. These methods provide mechanisms for representing constraints, structuring intermediate reasoning, and supporting verification. This session will bring together researchers to discuss how these approaches can complement current language and multimodal systems and support trustworthy reasoning. We hope the discussion will help participants identify concrete ways to use symbolic and structured methods in their own work. The session will also explore how these approaches can be evaluated across modalities and application domains such as spatial, legal, and scientific reasoning.

ArabicNLP

Time: 11:00–12:30 (Session 11)
Location: G9-10
Organizers: Salam Khalifa, Samar Magdy

Description:
The ArabicNLP BoF is for anyone who is interested in Arabic-centered NLP and related topics and applications. The session will discuss the latest trends and challenges, as well as networking opportunities with the larger ArabicNLP community.

We will make EMNLP 2026’s theme a central question of the BoF: as general-purpose models become widely available, what should Arabic NLP research prioritize, from dialect coverage and responsible data creation to evaluation that reflects the needs of Arabic-speaking communities

Queer in AI

Event: Affinity group lunch (Day 3)
Time: 12:30–14:00
Location: G12
Organizer: Dennis Ulmer

Description:
Queer in AI’s mission is to raise awareness of queer issues in AI/ML, foster a community of queer researchers and celebrate the work of queer scientists. We use “queer” as an umbrella term for people with diverse non-normative sexual orientations, romantic orientations, and/or genders, corresponding to acronyms like LGBTQIA+. We also explicitly include those questioning their identities. It was established by queer scientists in AI with the mission to make the AI community a safe and inclusive place that welcomes, supports, and values LGBTQIA+ people. Queer in AI’s demographic survey reveals that most queer scientists in our community do not feel completely welcome in conferences or their work environments, with the main reasons being a lack of queer community and role models. At EMNLP 2026, we will host a networking lunch. After an initial presentation with more information on Queer in AI, you invite you to socialize with other researchers. Please make sure to bring your own lunch and join the corresponding group in the conference app for more information on the event and a potential Queer in AI after-hours social during one of the evenings of the conference.