Swabha Swayamdipta

Career

Swayamdipta received a PhD from Carnegie Mellon University in 2019, advised by Noah A. Smith and Chris Dyer, and was then a postdoctoral researcher at the Allen Institute for AI and the University of Washington, working with Yejin Choi. Since fall 2022 Swayamdipta has been an assistant professor of computer science at USC, where Swayamdipta leads the Datasets, Interpretability, Language and Learning (DILL) Lab.⁴

Research

Swayamdipta's work has focused on estimating dataset quality and finding and reducing undesirable biases in data and models. Papers in this line include "Annotation Artifacts in Natural Language Inference Data" (2018), Dataset Cartography (2020) and "Understanding Dataset Difficulty with V-Usable Information" (2021); Swayamdipta also co-authored "Don't Stop Pretraining" (2020) and MAUVE (2021), a measure of the gap between machine-written and human text.⁴⁵²⁶⁷⁸

Current directions listed on Swayamdipta's site are robust evaluation of language models' generative capabilities, the mechanisms and geometry of language models, and private, safe human-AI collaboration in high-stakes settings. Recent papers include an ICML 2026 spotlight on the relationship between Transformer weight geometry and performance and an ACL 2026 study of language-model assistants for public health data annotation.¹

Work and contributions

• Dataset Cartography: Mapping and Diagnosing Datasets with Training Dynamics Open-source code

  First author of the 2020 Dataset Cartography paper; the allenai/cartography repository, released under the Apache License 2.0, provides the code and cites the paper with Swayamdipta as first author.²³

Timeline

1. Mar 6, 2018Co-authored "Annotation Artifacts in Natural Language Inference Data".⁵
2. 2019Completed a PhD at Carnegie Mellon University, advised by Noah A. Smith and Chris Dyer.⁴
3. Apr 23, 2020Co-authored "Don't Stop Pretraining - Adapt Language Models to Domains and Tasks".⁷
4. Sep 22, 2020Released Dataset Cartography as first author.²
5. Feb 2, 2021Co-authored MAUVE, a measure of the gap between neural and human text.⁸
6. Oct 16, 2021Co-authored "Understanding Dataset Difficulty with V-Usable Information".⁶
7. 2022Joined USC as a Gabilan Assistant Professor of computer science in fall 2022.⁴
8. Dec 8, 2025Received an NIH AIM-Ahead Hub-Specific Project Award with Eric Rice.¹

Other documented roles

Roles the sources document that are not confirmed as current.

• Postdoctoral researcher working with Yejin Choi, Allen Institute for AI (Ai2) and University of Washington⁴

Honors and recognition

• ICML Outstanding Paper Award · International Conference on Machine Learning, 2022⁴
• NeurIPS Outstanding Paper Award · Neural Information Processing Systems Foundation, 2021⁴
• ACL Best Paper Honorable Mention · Association for Computational Linguistics, 2020⁴

Where they work

• University of Southern California 2 people in the wiki
• Amazon 39 people in the wiki

Related people

• Dani YogatamaAlso at University of Southern California
• Leo GaoLanguage models · Evaluation and benchmarks
• Jesse DodgeEvaluation and benchmarks · Datasets and data
• Yacine JerniteDatasets and data · Language models
• Kyle LoDatasets and data · Language models
• Marina DanilevskyEvaluation and benchmarks · Language models
• Rohan TaoriLanguage models · Evaluation and benchmarks
• Ruchir PuriLanguage models · Datasets and data
• Pradeep DasigiLanguage models · Evaluation and benchmarks
• Eugene YanLanguage models · Evaluation and benchmarks