Papers by Zhilin Wang
Enhancing Language Representation with Constructional Information for Natural Language Understanding (2023.acl-long)
Copied to clipboard
| Challenge: | Recent advances in natural language processing focus on acquiring lexico-semantic information. |
| Approach: | They propose a construction grammar which highlights the pairings of form and meaning to enrich language representation. |
| Outcome: | The proposed model is superior to existing models on a variety of NLU tasks. |
Just Like a Human Would, Direct Access to Sarcasm Augmented with Potential Result and Reaction (2023.acl-long)
Copied to clipboard
| Challenge: | sarcasm is a form of irony conveying mockery and contempt . social media has become increasingly popular for identifying sarcasm . |
| Approach: | They develop a method to detect sarcasm from social media using augmented potentials. |
| Outcome: | The proposed method outperforms baselines on benchmark datasets. |
Simple and Effective Graph-to-Graph Annotation Conversion (2022.coling-1)
Copied to clipboard
| Challenge: | Existing work on graph-structured annotation conversions has focused on feature-based models which are not easily applicable to new conversions. |
| Approach: | They propose two graph-to-graph conversion approaches which use pseudo data and inherit parameters to guide conversions respectively. |
| Outcome: | The proposed approaches outperform strong baselines with higher conversion score on a graph-structured dataset and other datasets. |
HelpSteer3: Human-Annotated Feedback and Edit Data to Empower Inference-Time Scaling in Open-Ended General-Domain Tasks (2025.acl-long)
Copied to clipboard
Zhilin Wang, Jiaqi Zeng, Olivier Delalleau, Daniel Egert, Ellie Evans, Hoo-Chang Shin, Felipe Soares, Yi Dong, Oleksii Kuchaiev
| Challenge: | Inference-Time Scaling is critical to the success of recent models such as OpenAI o1 and DeepSeek R1 . however, many techniques require tasks to have answers that can be verified . |
| Approach: | They use data to train dedicated Feedback and Edit Models capable of inference-time scaling for open-ended tasks. |
| Outcome: | The proposed model can reach SoTA performance on Arena Hard at 92.7 as of 5 Mar 2025. |
Lost in Literalism: How Supervised Training Shapes Translationese in LLMs (2025.acl-long)
Copied to clipboard
| Challenge: | Large language models exhibit translationese errors and generate unexpected unnatural translations . Neural machine translation (NMT) has become the dominant method in machine translation research . |
| Approach: | They evaluate the prevalence of translationese in LLM-generated translations and investigate its roots during supervised fine-tuning. |
| Outcome: | The proposed methods reduce translationese while improving translation naturalness . the proposed methods are validated by human evaluations and automatic metrics . |
SEE: Continual Fine-tuning with Sequential Ensemble of Experts (2025.findings-acl)
Copied to clipboard
| Challenge: | Continual fine-tuning of large language models suffers from catastrophic forgetting . some approaches use routers to assign tasks to experts, but continual learning often requires retraining . |
| Approach: | They propose a framework that integrates routing and response mechanisms within each expert . it eliminates the need for an additional router and allows each expert to decide whether a query should be handled . |
| Outcome: | The proposed framework outperforms previous approaches in continual fine-tuning . it can handle learning tasks and out-of-distribution instances, paving the way for distributed model ensembling. |
No, you’re not alone: A better way to find people with similar experiences on Reddit (D19-55)
Copied to clipboard
| Challenge: | a probabilistic clustering algorithm can help users find posts that discuss experiences similar to their own . a recent study shows that probabilistic Clustering can yield a better performance than baseline clustering methods . |
| Approach: | They propose a probabilistic clustering algorithm that can help Reddit users find posts that discuss experiences similar to their own. |
| Outcome: | The proposed algorithm can find posts that discuss experiences similar to their own . it performs better than baseline clustering methods due to high runtime overhead . |
Spotting AI’s Touch: Identifying LLM-Paraphrased Spans in Text (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing work focuses on detecting (partially) AI-generated texts, but paraphrasing is commonly employed in various application scenarios for text refinement and diversity. |
| Approach: | They propose a framework for paraphrased text span detection that takes in the full text and assigns each sentence with a score indicating the paraphrasing degree. |
| Outcome: | The proposed framework can detect paraphrased text spans within a text . it takes in the full text and assigns each sentence with a score indicating the paraphrasing degree. |
FVQA 2.0: Introducing Adversarial Samples into Fact-based Visual Question Answering (2023.findings-eacl)
Copied to clipboard
| Challenge: | Fact-based Visual Question Answering (FVQA) is a visual question answering task that requires information retrieval using common sense knowledge graphs to answer. |
| Approach: | They propose a new test question with adversarial variants to address this imbalance by using a KB-VQA dataset that is small and contains only one answer per question. |
| Outcome: | The proposed version reduces the vulnerability of the original FVQA dataset without human annotations. |
Learning to Detect Noisy Labels Using Model-Based Features (2022.findings-emnlp)
Copied to clipboard
Zhihao Wang, Zongyu Lin, Junjie Wen, Xianxin Chen, Peiqi Liu, Guidong Zheng, Yujun Chen, Zhilin Yang
| Challenge: | Existing approaches to reduce label noise rely on heuristics and sample losses. |
| Approach: | They propose a method that transfers the noise distribution to a clean set and trains a model to distinguish noisy labels from clean ones using model-based features. |
| Outcome: | Empirically, the proposed approach improves over strong baselines on a wide range of tasks including text classification and speech recognition. |
A Closer Look into the Robustness of Neural Dependency Parsers Using Better Adversarial Examples (2021.findings-acl)
Copied to clipboard
| Challenge: | Neural network-based models have been successful in a wide range of NLP tasks, but their performance is undermined by adversarial examples that would pose no confusion for humans. |
| Approach: | They propose a method to generate high-quality adversarial examples with a higher number of candidate generators and stricter filters and then verify their quality using automatic and human evaluations. |
| Outcome: | The proposed method improves the robustness of English parsing models by relying on adversarial training and model ensembling. |
Data, Data Everywhere: A Guide for Pretraining Dataset Construction (2024.emnlp-main)
Copied to clipboard
Jupinder Parmar, Shrimai Prabhumoye, Joseph Jennings, Bo Liu, Aastha Jhunjhunwala, Zhilin Wang, Mostofa Patwary, Mohammad Shoeybi, Bryan Catanzaro
| Challenge: | Recent language models have impressive capabilities on a number of evaluation areas. |
| Approach: | They conduct systematic analysis of pretraining set construction to identify which methods yield the greatest gains in model accuracy. |
| Outcome: | The proposed method can be used to refine and improve a pretraining set. |
HelpSteer: Multi-attribute Helpfulness Dataset for SteerLM (2024.naacl-long)
Copied to clipboard
Zhilin Wang, Yi Dong, Jiaqi Zeng, Virginia Adams, Makesh Narsimhan Sreedhar, Daniel Egert, Olivier Delalleau, Jane Scowcroft, Neel Kant, Aidan Swope, Oleksii Kuchaiev
| Challenge: | Existing helpfulness preference datasets do not specify what makes some responses more helpful and others less helpful. |
| Approach: | They use a dataset that has annotated for correctness, coherence, complexity, and verbosity. |
| Outcome: | The dataset has annotations for correctness, coherence, complexity, and verbosity in addition to overall helpfulness of responses. |
MAGE: Machine-generated Text Detection in the Wild (2024.acl-long)
Copied to clipboard
Yafu Li, Qintong Li, Leyang Cui, Wei Bi, Zhilin Wang, Longyue Wang, Linyi Yang, Shuming Shi, Yue Zhang
| Challenge: | Existing research has focused on evaluating detection methods for specific domains or language models. |
| Approach: | They build a testbed to detect texts from diverse human writings and LLMs using different detection methods. |
| Outcome: | Empirical results show that the top performing detector can identify 84.12% out-of-domain texts generated by a new LLM, indicating the feasibility for application scenarios. |
Humanoid Agents: Platform for Simulating Human-like Generative Agents (2023.emnlp-demo)
Copied to clipboard
| Challenge: | Humanoid Agents aims to guide Generative Agents to behave more like humans using System 1 processing . we introduce three elements of System 1 that can influence their behavior . |
| Approach: | They propose a system that guides Generative Agents to behave more like humans . they introduce three elements of System 1 processing that can influence their behavior . humanoid agents can use these dynamic elements to adapt their daily activities and conversations . |
| Outcome: | The proposed system guides Generative Agents to behave more like humans . it incorporates three elements of System 1 processing that can influence their behavior . humanoid agents can adapt their daily activities and conversations with other agents . |
Unveiling Attractor Cycles in Large Language Models: A Dynamical Systems View of Successive Paraphrasing (2025.acl-long)
Copied to clipboard
| Challenge: | Dynamical systems theory provides a framework for understanding iterative processes and evolution over time. |
| Approach: | They propose to apply this perspective to large language models which iteratively map input text to output text and re-express meaning with linguistic variation. |
| Outcome: | The proposed model reveals that paraphrases re-express meaning with linguistic variation limiting linguistic diversity . |
CoELM: Construction-Enhanced Language Modeling (2024.acl-long)
Copied to clipboard
| Challenge: | Recent studies show that integrating constructional information can improve the performance of pre-trained language models. |
| Approach: | They propose a construction-Enhanced language model that embeds constructional semantics into language models for natural language generation. |
| Outcome: | The proposed model outperforms existing models on various benchmarks. |
GPS: Genetic Prompt Search for Efficient Few-Shot Learning (2022.emnlp-main)
Copied to clipboard
| Challenge: | Pretrained language models are often finetuned for downstream tasks, which has been shown to improve performance over non-pretrained models. |
| Approach: | They propose a genetic algorithm to automatically search for the best prompt for few-shot learning with pretrained language models by gradient-free algorithm. |
| Outcome: | Experiments on diverse datasets show that the proposed method outperforms manual prompts by 2.6 points. |
The Best of Both Worlds: Combining Parallel and Sequential Inference Scaling via Aggregation Fine-Tuning (2026.findings-acl)
Copied to clipboard
| Challenge: | Empirical results show that AFT-trained models achieve substantial gains with test-time scaling. |
| Approach: | They introduce a supervised fine-tuning paradigm where models synthesize multiple draft responses into a single, refined answer. |
| Outcome: | Empirical results show that AFT-trained models outperform baseline models while eliminating external guidance. |
SteerLM: Attribute Conditioned SFT as an (User-Steerable) Alternative to RLHF (2023.findings-emnlp)
Copied to clipboard
| Challenge: | supervised fine-tuning and reinforcement learning from human feedback (RLHF) are not effective in generating useful and high-quality responses. |
| Approach: | They propose a supervised fine-tuning method that empowers end-users to control responses during inference. |
| Outcome: | Experiments show that supervised fine-tuning and reinforcement learning from human feedback (RLHF) can generate helpful and high-quality responses while maintaining customizability. |
ZeroPrompt: Scaling Prompt-Based Pretraining to 1,000 Tasks Improves Zero-Shot Generalization (2022.findings-emnlp)
Copied to clipboard
| Challenge: | a recent study shows that task scaling can be an efficient alternative to model scaling. |
| Approach: | They propose a multitask pretraining approach ZeroPrompt for zero-shot generalization . they focus on task scaling and zero-shooting to improve model performance . |
| Outcome: | The proposed approach improves zero-shot generalization efficiency by 30 times with task scaling. |
Partial Order-centered Hyperbolic Representation Learning for Few-shot Relation Extraction (2025.coling-main)
Copied to clipboard
| Challenge: | Existing methods for few-shot relation extraction are limited to labeled instances and rely on data labeling. |
| Approach: | They propose a partial order-centered hyperbolic representation learning framework which imposes constraints on relations on instances by modeling partial order in hyperbolical space. |
| Outcome: | The proposed framework outperforms baseline methods on three benchmark datasets on 1-shot settings lacking relation descriptions. |