Papers by Liwei Jiang
Guardrails and Security for LLMs: Safe, Secure and Controllable Steering of LLM Applications (2025.acl-tutorials)
Copied to clipboard
Traian Rebedea, Leon Derczynski, Shaona Ghosh, Makesh Narsimhan Sreedhar, Faeze Brahman, Liwei Jiang, Bo Li, Yulia Tsvetkov, Christopher Parisien, Yejin Choi
| Challenge: | Pretrained generative models provide novel ways for users to interact with computers. |
| Approach: | This tutorial provides an overview of key guardrail mechanisms developed for LLMs along with evaluation methodologies and a detailed security assessment protocol. |
| Outcome: | This tutorial provides an overview of key guardrail mechanisms developed for LLMs, along with evaluation methodologies and a detailed security assessment protocol. |
Inference-Time Policy Adapters (IPA): Tailoring Extreme-Scale LMs without Fine-tuning (2023.emnlp-main)
Copied to clipboard
Ximing Lu, Faeze Brahman, Peter West, Jaehun Jung, Khyathi Chandu, Abhilasha Ravichander, Prithviraj Ammanabrolu, Liwei Jiang, Sahana Ramnath, Nouha Dziri, Jillian Fisher, Bill Lin, Skyler Hallinan, Lianhui Qin, Xiang Ren, Sean Welleck, Yejin Choi
| Challenge: | Extreme-scale language models have shown exceptional performance on a variety of language tasks, but the degree of control offered by these models through pure prompting is limited. |
| Approach: | They propose an inference-time policy adapter which tailors a large base model without fine-tuning it. |
| Outcome: | The proposed model outperforms baseline methods on five challenging text generation tasks and even over GPT-4. |
Impossible Distillation for Paraphrasing and Summarization: How to Make High-quality Lemonade out of Small, Low-quality Model (2024.naacl-long)
Copied to clipboard
Jaehun Jung, Peter West, Liwei Jiang, Faeze Brahman, Ximing Lu, Jillian Fisher, Taylor Sorensen, Yejin Choi
| Challenge: | Impossible Distillation is a framework for paraphrasing and sentence summarization that can be trained from a low-quality teacher model. |
| Approach: | They propose a framework that distills a high-quality dataset from a low-quality teacher . they hypothesize and verify the paraphrastic proximity intrinsic to pre-trained LMs . |
| Outcome: | The proposed framework outperforms baseline models on unconstrained paraphrase generation and sentence summarization benchmarks. |
“I’m Not Mad”: Commonsense Implications of Negation and Contradiction (2021.naacl-main)
Copied to clipboard
| Challenge: | a new commonsense knowledge graph for negated and contradicted events is developed to help humans reason about their underlying causes and effects. |
| Approach: | They propose a new commonsense knowledge graph with 624K if-then rules focusing on negated and contradictory events. |
| Outcome: | The proposed model can be used to analyze negated and contradicted statements in natural language. |
What Makes it Ok to Set a Fire? Iterative Self-distillation of Contexts and Rationales for Disambiguating Defeasible Social and Moral Situations (2023.findings-emnlp)
Copied to clipboard
Kavel Rao, Liwei Jiang, Valentina Pyatkin, Yuling Gu, Niket Tandon, Nouha Dziri, Faeze Brahman, Yejin Choi
| Challenge: | Moral or ethical judgments rely heavily on the contexts in which they occur . a student model that produces defeasible contexts with improved validity, diversity, and defasibility is superior to intermediate student models . |
| Approach: | a new study uses a student model to provide contextualizations that make an action morally acceptable . the model is based on a dataset of 115K defeasible moral actions rated highly by human annotators . |
| Outcome: | The proposed model outperforms all intermediate models in a high-quality dataset . the model is based on 1.2M entries of contextualizations and rationales for 115K moral actions . |
WESR: A Benchmark and Strong Baseline for Word-level Event-Speech Recognition (2026.findings-acl)
Copied to clipboard
Chenchen Yang, Kexin Huang, Liwei Fan, Qian Tu, Botian Jiang, Dong Zhang, Linqi Yin, Shimin Li, Zhaoye Fei, Qinyuan Cheng, Xipeng Qiu
| Challenge: | aaron carroll: the precise localization of non-verbal vocal events remains a critical yet under-explored challenge. carroll says current methods suffer from insufficient task definitions with limited category coverage. carrol: knowing exactly where an event occurred is not enough; knowing exactly what it happened is. |
| Approach: | They propose a taxonomy of 21 vocal events with a new categorization into discrete versus continuous types. |
| Outcome: | The proposed model disentangles ASR errors from event detection while maintaining ASR quality. |
NovaCOMET: Open Commonsense Foundation Models with Symbolic Knowledge Distillation (2023.findings-emnlp)
Copied to clipboard
Peter West, Ronan Bras, Taylor Sorensen, Bill Lin, Liwei Jiang, Ximing Lu, Khyathi Chandu, Jack Hessel, Ashutosh Baheti, Chandra Bhagavatula, Yejin Choi
| Challenge: | a new commonsense knowledge model, NovaCOMET, combines knowledge and general task models. |
| Approach: | They propose an open commonsense knowledge model that combines knowledge and general task models. |
| Outcome: | The proposed model matches or exceeds existing knowledge models on commonsense reasoning tasks. |
Aligning to Social Norms and Values in Interactive Narratives (2022.naacl-main)
Copied to clipboard
| Challenge: | Social value alignment is the ability to create agents that act in alignment with socially beneficial norms and values in interactive narratives or text-based games. |
| Approach: | They introduce a game-value ALignment agent that uses social commonsense to restrict its action space to actions that are aligned with socially beneficial values. |
| Outcome: | The proposed agent improves state-of-the-art task performance by 4% while reducing the frequency of socially harmful behaviors by 25% compared to strong contemporary value alignment approaches. |
CulturalBench: A Robust, Diverse and Challenging Benchmark for Measuring LMs’ Cultural Knowledge Through Human-AI Red-Teaming (2025.acl-long)
Copied to clipboard
Yu Ying Chiu, Liwei Jiang, Bill Yuchen Lin, Chan Young Park, Shuyue Stella Li, Sahithya Ravi, Mehar Bhatia, Maria Antoniak, Yulia Tsvetkov, Vered Shwartz, Yejin Choi
| Challenge: | CulturalBench is a set of 1,696 human-written and human-verified questions to assess LMs’ cultural knowledge covering 45 global regions including underrepresented ones like Bangladesh, Zimbabwe, and Peru. |
| Approach: | They construct a set of 1,696 human-written and human-verified questions to assess LMs' cultural knowledge, covering 45 global regions including underrepresented ones like Bangladesh, Zimbabwe, and Peru. |
| Outcome: | The proposed model outperforms other models across cultures, while underperforming on questions related to North Africa, South America and Middle East. |
JAMDEC: Unsupervised Authorship Obfuscation using Constrained Decoding over Small Language Models (2024.naacl-long)
Copied to clipboard
| Challenge: | Existing methods to protect the identity and privacy of online authorship are lacking supervision data for diverse authorship and domains. |
| Approach: | They propose an unsupervised inference-time approach to authorship obfuscation that uses a user-controlled, inference time algorithm to oblige the authorship. |
| Outcome: | The proposed method outperforms state-of-the-art methods while performing competitively against a propriety model two orders of magnitudes larger. |
Can Language Models Reason about Individualistic Human Values and Preferences? (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods and evaluation frameworks for achieving pluralistic alignment are limited by the diversity of people, which is pre-specified and coarsely categorized, papering over individuality. |
| Approach: | They propose to use a dataset transformed from the influential World Values Survey to study language models on the specific challenge of individualistic value reasoning. |
| Outcome: | The proposed model can predict individualistic values with accuracies between 55% and 65%, while a precise description of individualistic value judgments cannot be approximated only via demographic information. |
SODA: Million-scale Dialogue Distillation with Social Commonsense Contextualization (2023.emnlp-main)
Copied to clipboard
Hyunwoo Kim, Jack Hessel, Liwei Jiang, Peter West, Ximing Lu, Youngjae Yu, Pei Zhou, Ronan Bras, Malihe Alikhani, Gunhee Kim, Maarten Sap, Yejin Choi
| Challenge: | a dataset of 1.5 million conversations distilled from everyday spoken situations is limited in scale due to its associated costs. |
| Approach: | They propose to make SODA a publicly available, million-scale high-quality social dialogue dataset . they contextualize social commonsense knowledge from a knowledge graph to distill broad spectrum of social interactions . |
| Outcome: | The proposed dataset is the first publicly available, million-scale high-quality social dialogue dataset. |
ProsocialDialog: A Prosocial Backbone for Conversational Agents (2022.emnlp-main)
Copied to clipboard
Hyunwoo Kim, Youngjae Yu, Liwei Jiang, Ximing Lu, Daniel Khashabi, Gunhee Kim, Yejin Choi, Maarten Sap
| Challenge: | Existing dialogue systems fail to respond properly to potentially unsafe user utterances . existing systems either ignore or passively agree with unsafe content . |
| Approach: | They introduce a dataset to teach conversational agents to respond to problematic content following social norms. |
| Outcome: | The proposed dataset shows that ProsocialDialog generates more socially acceptable dialogues than existing models. |
Reading Books is Great, But Not if You Are Driving! Visually Grounded Reasoning about Defeasible Commonsense Norms (2023.emnlp-main)
Copied to clipboard
Seungju Han, Junhyeok Kim, Jack Hessel, Liwei Jiang, Jiwan Chung, Yejin Son, Yejin Choi, Youngjae Yu
| Challenge: | NormLens is a visual-grounded framework for understanding commonsense norms . state-of-the-art models are not well-aligned with human annotation, we show . |
| Approach: | They propose a visual-grounded framework to study commonsense norms by NormLens . they find that models are not well-aligned with human annotation . |
| Outcome: | The proposed model judgments and explanations are not well-aligned with human annotations. |
Following the Autoregressive Nature of LLM Embeddings via Compression and Alignment (2025.emnlp-main)
Copied to clipboard
Jingcheng Deng, Zhongtao Jiang, Liang Pang, Zihao Wei, Liwei Chen, Kun Xu, Yang Song, Huawei Shen, Xueqi Cheng
| Challenge: | Experimental results demonstrate that our method significantly outperforms traditional contrastive learning approaches when using the same amount of data. |
| Approach: | They propose a new contrastive learning method built on embedding conditional probability distributions that integrates two tasks: information compression and conditional distribution alignment. |
| Outcome: | The proposed method outperforms traditional contrastive learning approaches and achieves comparable performance to state-of-the-art models when using the same amount of data. |
BiasX: “Thinking Slow” in Toxic Content Moderation with Explanations of Implied Social Biases (2023.emnlp-main)
Copied to clipboard
| Challenge: | Toxicity annotators and content moderators often default to mental shortcuts when making decisions, leading to subtle toxicity being missed and seemingly harmless content being over-detected. |
| Approach: | They propose a framework that provides AI-generated explanations of statements’ implied social biases to enhance content moderation setups. |
| Outcome: | The proposed framework significantly improves content moderation setups by enabling users to think more thoroughly about their decisions. |
ClarifyDelphi: Reinforced Clarification Questions with Defeasibility Rewards for Social and Moral Situations (2023.acl-long)
Copied to clipboard
Valentina Pyatkin, Jena D. Hwang, Vivek Srikumar, Ximing Lu, Liwei Jiang, Yejin Choi, Chandra Bhagavatula
| Challenge: | Changing contexts can flip the moral judgment of an action. |
| Approach: | They propose an interactive system that learns to ask clarification questions to elicit salient contexts of a social or moral situation. |
| Outcome: | The proposed system generates more relevant, informative and defeasible questions compared to baselines. |
Symbolic Knowledge Distillation: from General Language Models to Commonsense Models (2022.naacl-main)
Copied to clipboard
Peter West, Chandra Bhagavatula, Jack Hessel, Jena Hwang, Liwei Jiang, Ronan Le Bras, Ximing Lu, Sean Welleck, Yejin Choi
| Challenge: | Prior studies suggested pre-trained language models possess limited understanding of commonsense knowledge despite otherwise stellar performance on leaderboards. |
| Approach: | They propose a framework that uses larger models to teach smaller models by distilling knowledge symbolically as text in addition to the neural model. |
| Outcome: | The proposed framework is based on a general language model teacher's commonsense knowledge graphs and a neural commonsensing model surpassing the teacher model's in all three criteria. |
NeuroLogic A*esque Decoding: Constrained Text Generation with Lookahead Heuristics (2022.naacl-main)
Copied to clipboard
Ximing Lu, Sean Welleck, Peter West, Liwei Jiang, Jungo Kasai, Daniel Khashabi, Ronan Le Bras, Lianhui Qin, Youngjae Yu, Rowan Zellers, Noah A. Smith, Yejin Choi
| Challenge: | Existing paradigms for text generation are left-to-right decoding from autoregressive language models. |
| Approach: | They propose a decoding algorithm that incorporates heuristic estimates of future cost that are efficient for large-scale language models. |
| Outcome: | The proposed method outperforms baselines on five generation tasks and achieves new state-of-the-art performance on table-to-text generation, constrained machine translation, and keyword-constrained generation. |