BlackboxNLP 2026

The Ninth Workshop on Analyzing and Interpreting Neural Networks for NLP

Co-located with EMNLP 2026 in Budapest, Hungary on October 28th, 2026

Call for Papers

Many recent performance improvements in NLP have come at the cost of understanding of the systems. How do we assess what representations and computations models learn? How do we formalize desirable properties of interpretable models, and measure the extent to which existing models achieve them? How can we build models that better encode these properties? What can new or existing tools tell us about these systems’ inductive biases?

The goal of this workshop is to bring together researchers focused on interpreting and explaining NLP models by taking inspiration from fields such as machine learning, psychology, linguistics, and neuroscience. We hope the workshop will serve as an interdisciplinary meetup that allows for cross-collaboration.

📣 News

No news yet. Check back later!

🗓️ Important dates

October 28th, 2026 - Workshop date.

All deadlines are 11:59PM UTC-12:00 (“Anywhere on Earth”).

Program

Program under construction, stay tuned!

FAQ

Can't find your answer? Send an email and we will get back to you.

Virtual poster sessions will be held on Gather.town. The links to the virtual poster sessions will be available on the Underline page of the BlackboxNLP workshop.
Yes, dual submissions are allowed if the other venue also allows dual submissions. If your submission is accepted to the archival track of BlackboxNLP, you will have to withdraw your submission to the other venue. Dual submissions to BlackboxNLP and EMNLP (including ARR) are not allowed.
Submissions to the Circuit Localization track should evaluate at least two models and two tasks, while there are no restrictions on the number of models and tasks for the Causal Variable Localization track. However, for any submission (regardless of the track) to be considered for our final ranking, it must be evaluated at least on the three less resource-intensive settings in MIB, namely: (1) GPT-2 on IOI (2) Qwen-2.5 on IOI; and (3) Qwen-2.5 on MCQA.
Shared task submissions will be evaluated by the workshop organizers and MIB creators based on the novelty and effectiveness of the proposed method. In practice, including more model-task combinations in the evaluation will strengthen high-scoring submissions by demonstrating the generality of the proposed method's effectiveness. Novelty will be evaluated in light of currently established methods for each one of the tracks.
Yes, submissions to the shared task will be considered archival, and will be published in the BlackboxNLP 2025 workshop proceedings on the ACL Anthology.
Shared task reports will be evaluated by the workshop organizers to assess whether they meet the requirements for archival publication.