The 1st Workshop on Customizable NLP: Progress and Challenges in Customizing NLP for a Domain, Application, Group, or Individual will be held at EMNLP 2024.
For NLP models to be usable in practice, particularly in emerging scenarios with widely varying use cases, situations, and user expectations, there is a need to develop models that can be tailored to different consumers (individuals, groups, or organizations) and easily controlled by them; models that can reason about their users’ (often private) knowledge and context to provide personalized responses. The topics of this workshop include (but not limited to):
1. Data: Data collection, processing, analysis, and annotation efforts to increase representation and aid customization; discussion and analysis of data sources not publicly available, and associated issues of privacy and copyright.
2. Modeling: New pretraining, fine-tuning, inference methods for customizing NLP models; customizing reward models and model alignment to diverse consumers. New modeling paradigms aimed at customization such as model ensembles, model averaging, federated learning, nonparametric models, etc.; customizing models at inference time via prompting, in-context learning, chain-of-thought prompting, etc.
3. Evaluation: Evaluation of existing generalist, non-customized models, identifying their shortcomings for varied use-cases; evaluation of customization techniques and customized models; interpretability and analysis of customization patterns across different kinds of consumers.
4. Open Science: Best practices for open and reproducible science concerning customizable NLP: dataset release and licensing, open-sourcing models, related privacy, copyright, and policy issues.
5. Applications: e.g., information seeking on sensitive data comprising legal, medical, or financial information; NLP models for communities reflecting sociolects, dialects, or other language varieties; personalized AI assistants, etc.
6. Ethical Issues: privacy and copyright; personalization, intrusiveness, unintended biases; invisibility versus hypervisibility.
Important Dates
Submission Deadline: Aug 15, 2024 (11:59pm, anywhere on earth)
Notification of Acceptance: Sep 30, 2024
Camera-ready papers due: Oct 7 (11:59pm, anywhere on earth)
Workshop Date: co-located with EMNLP 2024 in November, Miami (exact date to be announced)
More Details: https://customnlp4u-24.github.io/
Organization:
Sachin Kumar, Ohio State University, Allen Institute for AI
Chan Young Park, Carnegie Mellon University
Vidhisha Balachandran, Microsoft Research
Weijia Shi, University of Washington
Shirley Anugrah Hayati, University of Minnesota, Twin Cities
Yulia Tsvetkov, University of Washington
Noah A. Smith, University of Washington
Hannaneh Hajishirzi, University of Washington
Dongyeop Kang, University of Minnesota, Twin Cities
David Jurgens, University of Michigan