Research & publications

Learning, evaluating, and making decisions under uncertainty.

Figure 1: safety-alignment design space
ICML 2026

Configurable Reward Model for Balanced Safety Alignment

Zhengping Jiang, Mehran Khodabandeh, Akash Bharadwaj, Manik Bhandari, Mayur Srungarapu, Anqi Liu, Benjamin Van Durme, Li Chen

Train a calibrated reward model that follows changing safety guidelines and guides the balance between helpfulness and safety.

Reward ModelingPost-training & AlignmentSafetyCalibration & Risk-control
Paper
Figure 1: critique-to-claim grounding
EMNLP 2025 Findings

ClaimCheck: How Grounded are LLM Critiques of Scientific Papers?

Jiefu Ou, William Gantt Walden, Kate Sanders, Zhengping Jiang, Kaiser Sun, Jeffrey Cheng, William Jurayj, Miriam Wanner, Shaobo Liang, Candice Morgan, Seunghoon Han, Weiqi Wang, Chandler May, Hannah Recknor, Daniel Khashabi, Benjamin Van Durme

Ground scientific critiques in the specific claims they challenge, then evaluate their validity and specificity.

EvaluationSemantics
Paper Dataset
Figure 1, method panels: leakage detection and counterfactual editing
ACL2024

RORA: Robust Free-Text Rationale Evaluation

Zhengping Jiang*, Yining Lu*, Hanjie Chen, Daniel Khashabi, Benjamin Van Durme, Anqi Liu

Evaluate what a rationale explains beyond the answer itself, while resisting shortcuts caused by label leakage.

Evaluation
Paper Code
Figure 1: modes of claim decomposition
*SEM2024

A Closer Look at Claim Decomposition

Miriam Wanner*, Seth Ebner*, Zhengping Jiang, Mark Dredze, Benjamin Van Durme

Separate errors in claim decomposition from errors in factuality, and assess whether subclaims faithfully represent the original text.

EvaluationSemantics
Paper Code
Figure 1: multilingual collection pipeline
2023

MegaWika: Millions of reports and their sources across 50 diverse languages

Samuel Barham, Orion Weller, Michelle Yuan, Kenton Murray, Mahsa Yarmohammadi, Zhengping Jiang, Siddharth Vashishtha, Alexander Martin, Anqi Liu, Aaron Steven White, Jordan Boyd-Graber, Benjamin Van Durme

Link Wikipedia passages to their sources across 50 languages to support evidence-grounded retrieval and report generation.

MultilingualEvaluation
Paper Dataset
Figure 2: synthetic segmentation example
EACL2021

Segmenting Subtitles for Correcting ASR Segmentation Errors

David Wan, Chris Kedzie, Faisal Ladhak, Elsbeth Turcan, Petra Galuščáková, Elena Zotkina, Zhengping Jiang, Peter Bell, Kathleen McKeown

Use subtitles and simulated boundary errors to train models that repair speech-recognition segmentation before translation.

Multilingual
Paper
Figure 1: scalar probability distributions across entailment categories
ACL2020

Uncertain natural language inference

Tongfei Chen*, Zhengping Jiang*, Adam Poliak, Keisuke Sakaguchi, and Benjamin Van Durme

Replace categorical entailment labels with human probability judgments about whether a hypothesis follows from a premise.

Probabilistic InferenceSemanticsEvaluation
Paper Code