Talk Title: Reference Aware Delexicalization (RAD) Framework: Theory Driven Artificial Intelligence Modeling for Domain Generalization

Abstract:

The ability to generalize is critical for machine learning and natural language processing models to perform effectively across a wide range of domains. However, state-of-the-art neural models often struggle to maintain performance when tested out-of-domain (OOD), emphasizing their inability to generalize beyond the data distribution used for training. This challenge of oversensitivity to spurious biases in the training data remains an open research problem across various natural language processing (NLP) task areas. Although prior work has introduced techniques to improve domain generalization capabilities of neural networks, existing methods are constrained in their ability to identify and mitigate complex latent data biases. In response to these limitations, we propose the novel Reference Aware Delexicalization (RAD) data augmentation framework designed to improve generalization for inference-based NLP tasks. The RAD framework uses attention weighting to detect biases and extract bias-prone lexical concepts. Grounded in the theory of reference, RAD's delexicalization method utilizes the principles of reference fixing and borrowing to generate context-aware placeholder mappings to reduce data oversensitivity. We conduct rigorous benchmark evaluations using RAD-augmented transformer architectures (e.g., BERT, RoBERTa) on several natural language inference, recognizing textual entailment, and fact verification datasets. Our findings demonstrate consistent improvements in OOD performance, indicating RAD's ability to improve model generalization for key natural language inference (NLI) and recognizing textual entailment (RTE) tasks. As advancing fundamental NLI and RTE capabilities remains crucial for many downstream NLP applications, this work highlights RAD's potential for positive impact across areas where inference and OOD robustness are highly valued. 

Recording:

Loading player for https://video.vt.edu/media/t/1_n36777e9...

About Dr. Sue Brown:

Susan (Sue) Brown is the Stevie Eller Professor and Department Head of Management Information Systems in the Eller College of Management at the University of Arizona. She received a PhD from the University of Minnesota and an MBA from Syracuse University. Her research interests include individual motivations for and consequences of IT use, diffusion of misinformation, and research methods. She has received funding for her research from the National Science Foundation, and other public and private organizations. Her work has appeared in leading journals in information systems and management including MIS Quarterly, Information Systems Research, Organizational Behavior and Human Decision Processes, and Journal of Management Information Systems. She has served as an Associate Editor at MIS Quarterly, Information Systems Research, Journal of the Association for Information Systems, and Decision Sciences and as a Senior Editor at MIS Quarterly and Information Systems Research. She is currently coeditor-in-chief of AIS Transactions on Replication Research. Sue has been active in the information systems community, serving as a faculty mentor for numerous doctoral, junior faculty, and mid-career faculty consortia. She currently co-chairs the doctoral consortium at HICSS and served as co-chair of the 2022 AMCIS in Minneapolis. She has received awards for her teaching, research, and service activities. In 2016, she received the Sandra Slaughter Service award, in 2017 she was named a fellow of the AIS (Association for Information Systems), and in 2022 received a Woman of Impact Award at the University of Arizona. In 2024, she became the 14th Editor-in-Chief of MIS Quarterly.