April 15–16, 2026 — Urbana-Champaign, Illinois
Midwest Speech and Language Days (MSLD) is a two-day meeting that continues the tradition of Illinois Speech Day and the Midwest Computational Linguistics Colloquium. Researchers from across the Midwest are invited to share work, hear keynotes, and foster new collaborations.
About the Program
MSLD’26 features two full days of invited keynotes, paper presentations, posters, and community meetups that highlight the breadth of speech, language, and conversational AI research happening across the region.
Sponsorship InformationSchedule Highlights
Below is the scheduled activities for each day. Please visit this link for more details on the Poster Sessions.
Day 1 — Wednesday, April 15
- 08:00 – 08:45Registration & Breakfast
- 08:45 – 09:00Opening Remarks
- 09:00 – 10:00Keynote: Rada Mihalcea (Session Chair: Gokhan Tur)
- 10:00 – 10:30Coffee Break
- 10:30 – 11:30Keynote: Dan Roth (Session Chair: Manling Li)
- 11:30 – 13:30Poster Session 1 & Lunch
- 13:30 – 14:30Keynote: Julia Hirschberg (Session Chair: Dilek Hakkani-Tür)
- 14:30 – 15:00Coffee Break
- 15:00 – 16:00Best Paper Talks (3×20 mins) (Session Chairs: Minje Kim, Hao Peng, Manling Li)
- 16:00 – 18:00Poster Session 2 & Refreshments
Day 2 — Thursday, April 16
- 08:00 – 09:00Breakfast
- 09:00 – 10:00Keynote: Nanyun (Violet) Peng (Session Chair: Heng Ji)
- 10:00 – 10:30Coffee Break
- 10:30 – 11:30Keynote: Shinji Watanabe (Session Chair: Minje Kim)
- 11:30 – 13:30Poster Session 3 & Lunch
- 13:30 – 14:30Keynote: Yizhe Zhang (Session Chair: Hao Peng)
- 14:30 – 15:00Coffee Break
- 15:00 – 15:45Panel (Panelists: Dan Roth, Ruhi Sarikaya, Karen Livescu, Heng Ji, Manling Li, Kevin Small) (Chairs: Hao Peng)
- 15:45 – 17:45Poster Session 4 & Refreshments
- 17:45 – 18:00Closing Remarks
Invited Speakers
Julia Hirschberg
Columbia Unversity
Code-Switching in Dialogue and Dialogue Systems
Abstract
For people who speak more than one language, code-switching (CSW) is a common phenomenon. However, spoken language recognition systems, including voice assistants, find it difficult to understand and appropriately reacting to this multilingual speech. We are studying how spoken and written CSW interacts with other aspects of communication, including the production of named entities and dialogue acts and the influence of entrainment, empathy, prosody, formality, and information load. Our goals are to improve prediction of when, why, and to what effect CSW occurs as well as how to produce appropriate code-switched responses to inform further development of voice assistants and their ability to successfully interact with multilingual users. We have studied many aspects of CSW, e.g.: Does the degree of formality of a conversation influence the degree of CSW in it? 2) What is the role of information load in predicting and explaining CSW? 3) Do speakers entrain on strategies of CSW in speech? 4) Is there a quantifiable relationship between CSW and empathy in speech? We are current examining: 5) Which dialogue acts tend to be produced most often in CSW? 6) Does the presence of named entities prime CSW? 7) How do speakers produce intonational contours (ToBI) when they perform CSW -- Do these match either of their languages or are they different from both? We are testing all these topics on speech and lexical features of: Standard American English with Spanish, Mandarin Chinese, and Hindi.
Bio
Julia Hirschberg is a Columbia CS Percy K. and Vida L.W. Hudson Professor and was previously at Bell Laboratories/AT&TLabs working on TTS. She currently studies spoken language: code-switching; conversational entrainment, emotion and empathy; identified risk for hospitalization in healthcare and producing safe dialogue agents for medical care; and identifying dialogue acts. She served on the ACL, CRA, IEEE SLTC, NAACL, ISCA (president 2005-7) Executive Boards, and AAAI Council, was editor of Computational Linguistics and Speech Communication, and is an AAAI, ISCA, ACL, ACM, and IEEE fellow and a member of the National Academy of Engineering, the American Academy of Arts and Sciences, and the American Philosophical Society member, and received the IEEE Flanagan Award and the ISCA Medal for Scientific Achievement. She received the National Academy of Artificial Intelligence Artificial Intelligence Exploration Award and the ACL Distinguished Service Medal in 2025.She has 5 PhD students and many research project students in Columbia Computer Science.
Rada Mihalcea
University of Michigan, Ann Arbor
All Users Are Equal, but Some Are More Equal Than Others
Abstract
Large Language Models are increasingly deployed as general-purpose systems intended to equally serve all their users. Yet in practice, we often see that these systems do not represent all users equally, leading to the unintended effect that some users end up benefiting from AI tools more than others. In this talk, I will describe recent research showing the uneven treatment of users across several dimensions, including age, gender, socioeconomic status, and political ideology. I will also discuss strategies to address these gaps, with the goal of preventing AI systems from amplifying existing inequalities and reinforcing a “rich get richer” dynamic.
Bio
Rada Mihalcea is the Janice M. Jenkins Professor of Computer Science and Engineering at the University of Michigan and the Director of the Michigan Artificial Intelligence Lab. Her research spans natural language processing, large language and language-vision models, and computational social science, with a focus on cross-cultural and cross-lingual models, fairness, interpretability, and responsible AI deployment. She leads multiple interdisciplinary initiatives at the intersection of AI, health, education, and social impact, and has developed widely adopted datasets and computational methods. She served as Program Co-Chair for EMNLP 2009 and ACL 2011, General Chair for NAACL 2015 and *SEM 2019, and co-founded the ACL Mentorship Program, the TextGraphs workshop series, and the NLP for Positive Impact workshop series. She is an ACM Fellow, AAAI Fellow, and ACL Fellow, and a former President of the ACL. She is a recipient of the Presidential Early Career Award for Scientists and Engineers awarded by President Obama.
Nanyun (Violet) Peng
University of California, Los Angeles
David and Goliath: Compute-Efficient Strategies for LLM Steering
Abstract
Large language models (LLMs) have achieved remarkable capabilities through massive scaling and extensive preference alignment. Yet, the prohibitive computational cost of training, aligning, and controlling these models remains a fundamental challenge. Overcoming this barrier is essential to democratize LLM development, enabling the broader research community to ensure robust safety and factual reliability without relying on resource-heavy retraining pipelines. In this talk, I present our explorations into lightweight, compute-efficient interventions across distinct architectural levels. We demonstrate that because alignment training often induces only small, directional parameter shifts, we can mathematically extrapolate these weights to boost alignment performance without additional training steps. Complementing this parameter-level insight, we investigate the LLM representation space, showing how directed optimization can nudge query representations toward safe, "higher-refusal" directions to prevent harmful compliance. We extend this mechanistic control to the internal routing of Mixture-of-Experts (MoE) architectures, illustrating how inference-time expert activation and deactivation can reliably dictate high-level behaviors like factual faithfulness and safety. Taken together, these directions sketch a roadmap toward highly steerable language models—systems that can be efficiently adapted, safeguarded, and refined within constrained computational budgets.
Bio
Nanyun (Violet) Peng is an Associate Professor of Computer Science at the University of California, Los Angeles, currently on sabbatical, and a Senior Staff Research Scientist at Google. Her research focuses on controllable and creative language generation, multilingual and multimodal models, and the development of automatic evaluation metrics, with a strong commitment to advancing robust and trustworthy artificial intelligence (AI). Her work has been recognized with multiple paper awards, including an Outstanding Paper Award at NAACL 2022, three Outstanding Paper Awards at EMNLP 2024, Oral Papers at NeurIPS 2022 and ICML 2023, as well as several Best Paper Awards at workshops. Her research has received support from the NSF CAREER Award, NIH R01, DARPA, IARPA, and multiple industrial research awards. She served as Program Chair for ICLR 2025 and EMNLP 2025, and as a board member of NAACL.
Dan Roth
University of Pennsylvania / Oracle
On Reasoning & Retrieving LLMs: Myths, Merits, and How to Move Forward
Abstract
"The rapid progress made over the last few years in generating linguistically coherent natural language has blurred, in the mind of many, the difference between natural language generation, understanding, knowledge retrieval and use, and the ability to reason with respect to the world. Nevertheless, reliably and consistently supporting high-level decisions that depend on natural language understanding and heterogenous information retrieval is still difficult for fundamental reasons that range from computational complexity to data organization in the wild (and are here to stay). I will discuss some of the challenges underlying reasoning and information access, argue that we should exploit what LLMs do well while delegating responsibility to special purpose models and solvers for decision making, and present some of our work in this space. I hope to collectively acknowledge some of the key GenAI myths and their consequences, think about their underlying causes, and discuss ways to move forward."
Bio
Dan Roth is the Eduardo D. Glandt Distinguished Professor at the University of Pennsylvania and Chief AI Scientist at Oracle. Previously Dan was a VP/Distinguished Scientist at AWS AI where he led the scientific effort behind Amazon's first-generation GenAI products, including Titan Models, Amazon Q, and Amazon Bedrock. Dan is a Fellow of the AAAS, ACM, AAAI, and ACL, and a recipient of the IJCAI John McCarthy Award “for major conceptual and theoretical advances in the modeling of natural language understanding, machine learning, and reasoning.” He has published broadly in natural language processing, machine learning, knowledge representation and reasoning, and learning theory, was the Editor-in-Chief of the Journal of Artificial Intelligence Research (JAIR) and has served as a Program Chair and Conference Chair for the major conferences in his research areas. Roth has been involved in several ML/NLP/GenAI startups in domains that range from legal and compliance to health care. Dan received his B.A Summa cum laude in Mathematics from the Technion, Israel and his Ph.D. in Computer Science from Harvard University in 1995.
Shinji Watanabe
Carnegie Mellon University
Reproducing Large Speech Foundation Models for Open Science
Abstract
Speech foundation models are transforming the field by unifying diverse speech-processing tasks through large-scale data, increased model capacity, and task diversity. This paradigm shift has led to a growing division of research roles: large technology companies primarily develop foundational models, while academic and smaller research groups focus on adaptation, analysis, and downstream applications. This separation raises concerns about transparency, reproducibility, and explainability. To address these challenges, our group at Carnegie Mellon University has developed Open Whisper-style Speech Models (OWSM), reproducing OpenAI Whisper-style training using only publicly available data and the open-source toolkit ESPnet as an effort toward open science. Because the entire training pipeline is transparent, OWSM enables reproducible research and more explainable model behaviors. In addition, we will present our recent efforts on pre-training large-scale multimodal speech–text language models and discuss the research challenges they raise. This presentation highlights both the technical advances and the open challenges in reproducing large speech foundation models, emphasizing the role of openness and transparency in advancing accessible, interpretable speech and audio technologies.
Bio
Shinji Watanabe is an Associate Professor at Carnegie Mellon University, Pittsburgh, PA. He received his B.S., M.S., and Ph.D. (Dr. Eng.) degrees from Waseda University, Tokyo, Japan. He was a research scientist at NTT Communication Science Laboratories, Kyoto, Japan, from 2001 to 2011, a visiting scholar at Georgia Institute of Technology, Atlanta, GA, in 2009, and a senior principal research scientist at Mitsubishi Electric Research Laboratories (MERL), Cambridge, MA USA from 2012 to 2017. Before Carnegie Mellon University, he was an associate research professor at Johns Hopkins University, Baltimore, MD, USA, from 2017 to 2020. His research interests include automatic speech recognition, speech enhancement, spoken language understanding, and machine learning for speech and language processing. He has published over 500 papers in peer-reviewed journals and conferences and received several awards, including the best paper award from ISCA Interspeech in 2024. He is a Senior Area Editor of the IEEE Transactions on Audio Speech and Language Processing. He was/has been a member of several technical committees, including the APSIPA Speech, Language, and Audio Technical Committee (SLA), IEEE Signal Processing Society Speech and Language Technical Committee (SLTC), and Machine Learning for Signal Processing Technical Committee (MLSP). He is an IEEE and ISCA Fellow.
Yizhe Zhang
Apple
Towards Understanding and Building Intuition for Language Model
Abstract
Intuition in language models yields short cut to the outcome and represents the implicit ability to anticipate, reason, and plan beyond next-token prediction. It may lead toward more efficient search over solution space and better generalization ability. In this talk I hope to understand and explore how to build intuition for language model, through several diffusion-based and hybrid models demonstrating stronger global planning, reasoning, and code generation capabilities. I will also discuss how future-aware data creation and agent-based training can further strengthen such intuition, allowing models to anticipate outcomes, refine reasoning paths, and act more efficiently. Finally, I will share some early insights on understanding when and how language models intuit—highlighting the balance between intuitive shortcuts and generalizable reasoning.
Bio
Dr. Yizhe Zhang is a Staff Research Scientist at Apple Machine Learning Research, focusing on natural language processing and machine learning. His recent work includes developing text diffusion model, LLM reasoning and code agent. Previously, he worked at Meta AI and Microsoft Research. He holds a Ph.D. from Duke University. Dr. Zhang has served as an Area Chair for top ML and NLP conferences, including NeurIPS, ICML, ICLR, ACL and EMNLP. He is also an Action Editor for Transactions on Machine Learning Research and the ACL Rolling Review. His recent research interests lie in bidirectional text models, code generation, implicit reasoning and lookahead training for large language models.
Organizing Team
General Chair
Gokhan Tur
Dilek Hakkani-Tür
Heng Ji
Technical Chair
Minje Kim
Hao Peng
Manling Li
Student Organizers
Vardhan Dongre
Ishika Agarwal
Beyza Bozdag
Shuhaib Mehri
Jaesung Bae
Jinu Lee
Dylan Zhang
Chi Han
Sofia Stoica
Aditi Tiwari
Kunlun Zhu
Takyoung Kim
Yuji Zhang
Cameron Churchwell
Emre Can Acikgoz
Nguyen Hoang
Yi-Jyun Sun
Parisa Rabbani
Frequently Asked Questions
General
What is MSLD?
Midwest Speech and Language Days (MSLD) is a two-day, non-archival gathering that brings together researchers in speech, language, dialogue, and AI. It continues the tradition of Illinois Speech Day and the Midwest Computational Linguistics Colloquium. Think of it as an informal networking event rather than a formal conference.
Is there a Slack for all participants?
Yes, please join this: https://join.slack.com/t/msld-2026--uiuc/shared_invite/zt-3uxv2ps5f-nBjxFfVBP_8d0vEWZy~fYw
When and where is MSLD 2026?
April 15–16, 2026, at the National Center for Supercomputing Applications (first floor auditorium), University of Illinois Urbana-Champaign.
Who can attend?
Everyone is welcome! MSLD especially encourages participation from students and early-career researchers across the Midwest, but attendees from any institution or region are welcome.
Is there a registration fee?
No, registration is completely free for everyone. Meals during the event are also provided at no cost.
Is there a registration deadline?
No, there is no registration deadline. However, space may be limited, so we encourage you to register early to secure your spot. The deadline for abstract submissions is March 2, 2026.
Can I attend only one of the two days?
Yes.
Is there a virtual or remote attendance option?
No, MSLD is an in-person-only event.
Will talks be recorded?
No.
Submissions
What is the submission format?
Abstracts should be one page plus references, describing published work, ongoing projects, or new ideas related to speech, language, and agentic AI. Submit via OpenReview.
Can I submit work that has already been accepted or published at another venue?
Yes! MSLD is non-archival, so you are welcome to present work that has been accepted or published elsewhere.
Can I submit work that was rejected from another venue?
Absolutely. We welcome submissions at all stages, including work-in-progress and papers you are currently refining for future submission.
Will my submission be published in proceedings?
No. MSLD is non-archival and does not publish formal proceedings. PDFs of accepted abstracts may be posted on the workshop website with author permission.
Can I submit more than one abstract?
Yes.
Do I need to be from a Midwest institution to submit?
No, submissions from any institution are welcome.
What are the important dates?
Submission deadline: March 2, 2026. Notification of acceptance: March 9, 2026. Conference dates: April 15–16, 2026.
What is the poster dimension?
They should be 3'4" by 2'6", but it is rotatable. You will have to print your own poster, but UIUC will provide the poster board/board pins to hang up the poster.
Presenting
What is the presentation format?
Accepted abstracts will be presented as posters. A subset will be selected for short oral presentations. The program also includes keynotes, panels, and informal networking sessions.
Travel & Logistics
🚨 Discounted Hotel Booking (Limited Availability)
A limited block of discounted rooms has been reserved at the TownePlace Suites by Marriott for MSLD attendees. Rooms are available on a first-come, first-served basis.
Deadline: April 1, 2026 (may be extended if rooms remain).
Funded students have separate accommodations and should not use this link. If you requested a funded room and have not received confirmation from the organizers, please book and pay for your own room using the link above.
How do I get to Champaign-Urbana?
You can fly into Willard Airport (CMI) directly, or fly into one of the Chicago-area airports (O'Hare or Midway), which have shuttle and bus service to campus. Peoria Charter offers a bus service from O'Hare to campus.
Is travel support available?
Limited travel support may be available, with priority given to student presenters. Some student accommodations may also be provided. Details will be announced on the website.
Are there recommended hotels nearby?
Most people stay at TownePlace Suites or Hampton Inn.
Will parking information be provided?
There is ample paid parking available near the venue.
During the Event
Will meals or coffee be provided?
Yes, meals are provided free of charge for all attendees.
Is there Wi-Fi access for campus visitors?
Yes, Illinois Guest Wi-Fi is free for everybody.
Is the venue wheelchair accessible?
Yes, the NCSA is fully accessible.
Will there be a social event or dinner?
No.
Contact
How can I contact the organizers?
Submit your questions through our contact form.
