Papers by Kai-Wei Chang
Representation Learning for Resource-Constrained Keyphrase Generation (2022.findings-emnlp)
Copied to clipboard
| Challenge: | State-of-the-art keyphrase generation methods depend on large annotated datasets, limiting their performance in domains with limited annotation data. |
| Approach: | They propose a method that first identifies salient information using retrieval-based corpus-level statistics and then learns a task-specific intermediate representation based on a pre-trained language model. |
| Outcome: | The proposed method improves keyphrase generation and zero-shot domain adaptation on multiple keyphrase benchmarks. |
Codec-SUPERB: An In-Depth Analysis of Sound Codec Models (2024.findings-acl)
Copied to clipboard
Haibin Wu, Ho-Lam Chung, Yi-Cheng Lin, Yuan-Kuei Wu, Xuanjun Chen, Yu-Chi Pai, Hsiu-Hsuan Wang, Kai-Wei Chang, Alexander Liu, Hung-yi Lee
| Challenge: | Researchers have developed a sound codec that can be used as tokenizers for preserving audio data and minimizing data transmission latency. |
| Approach: | They propose to use codec-SUPERB to assess codec models across representative sound applications and signal-level metrics rooted in sound domain knowledge. |
| Outcome: | The proposed codec-SUPERB model is evaluated on selected experimental settings. |
Searching for an Effective Defender: Benchmarking Defense against Adversarial Word Substitution (2021.emnlp-main)
Copied to clipboard
Zongyi Li, Jianhan Xu, Jiehang Zeng, Linyang Li, Xiaoqing Zheng, Qi Zhang, Kai-Wei Chang, Cho-Jui Hsieh
| Challenge: | Existing methods to defend against adversarial word-substitution attacks have not been evaluated or compared in a systematic manner. |
| Approach: | They propose to compare different defense methods under representative adversarial attacks . they propose a method that improves the robustness of neural text classifiers against such attacks a . |
| Outcome: | The proposed method improves robustness of neural text classifiers against such attacks by a significant margin. |
Mitigating Gender Bias Amplification in Distribution by Posterior Regularization (2020.acl-main)
Copied to clipboard
| Challenge: | Recent studies show that data-driven machine learning models carry societal biases in the dataset they trained on. |
| Approach: | They propose to calibrate top predictions of a model by injecting corpus-level constraints to ensure that the gender disparity is not amplified. |
| Outcome: | The proposed method can almost remove bias amplification in the distribution with little loss of performance. |
Text encoders bottleneck compositionality in contrastive vision-language models (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing multimodal models are often unable to reason about simple spatial relations or attribute attachments. |
| Approach: | They first curate CompPrompts, a set of increasingly compositional image captions that VL models should be able to capture . then train text-only recovery probes that aim to reconstruct captions from single-vector text representations produced by several VL model. |
| Outcome: | The proposed model can reconstruct captions from single-vector text representations produced by several models on a broader range of scenes compared to previous models. |
Where Fact Ends and Fairness Begins: Redefining AI Bias Evaluation through Cognitive Biases (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing benchmarks conflate factual correctness and normative fairness . a model may generate responses that are factually accurate but socially unfair . |
| Approach: | They propose a benchmark to examine the boundary between fact and fair . they draw on representativeness bias, attribution bias and ingroup–outgroup bias to explain why models often misalign fact and faireness. |
| Outcome: | The proposed model is based on ten frontier models and is available on github . it is compared with a standard model that generates people of color in Nazi-era uniforms . |
DEGREE: A Data-Efficient Generation-Based Event Extraction Model (2022.naacl-main)
Copied to clipboard
I-Hung Hsu, Kuan-Hao Huang, Elizabeth Boschee, Scott Miller, Prem Natarajan, Kai-Wei Chang, Nanyun Peng
| Challenge: | Existing models for event extraction require expensive human annotations. |
| Approach: | They propose a data-efficient event extraction model that formulates event extraction as a conditional generation problem. |
| Outcome: | The proposed model can be trained with only a few labeled examples. |
MetaVL: Transferring In-Context Learning Ability From Language Models to Vision-Language Models (2023.acl-short)
Copied to clipboard
| Challenge: | Large-scale pre-trained vision-language models do not possess the ability to conduct in-context learning. |
| Approach: | They propose to meta-train a language model to perform in-context learning on NLP tasks and then transfer this model to VL tasks by attaching a visual encoder. |
| Outcome: | The proposed model outperforms the baseline model on VQA, OK-VQA, and GQA while having 20 times fewer parameters. |
V-ALPHASOCIAL: Benchmark and Self-Reflective Chain-of-Thought Generation for Visual Social Commonsense Reasoning (2025.findings-acl)
Copied to clipboard
Zongyu Lin, Zhikun Xu, Xiaohan Song, Yixin Wan, Xingcheng Yao, Tsung-Han Lin, Selina Song, Pranav Subbaraman, Ben Zhou, Kai-Wei Chang, Yizhou Sun
| Challenge: | Social commonsense reasoning is a multimodal task that requires both textual and visual cues. |
| Approach: | They propose a method that integrates visual cues into social commonsense reasoning tasks. |
| Outcome: | The proposed method improves social commonsense reasoning on a multimodal foundation model. |
Mitigating Bias for Question Answering Models by Tracking Bias Influence (2024.naacl-long)
Copied to clipboard
Mingyu Ma, Jiun-Yu Kao, Arpit Gupta, Yu-Hsiang Lin, Wenbo Zhao, Tagyoung Chung, Wei Wang, Kai-Wei Chang, Nanyun Peng
| Challenge: | Existing literature observes bias in question answering (QA) models, but there is no method to mitigate it. |
| Approach: | They propose an approach to mitigate the bias of question answering models by observing the influence of a query instance on another instance. |
| Outcome: | The proposed method reduces bias level in all 9 bias categories while maintaining comparable QA accuracy. |
LUME: LLM Unlearning with Multitask Evaluations (2025.findings-emnlp)
Copied to clipboard
Anil Ramakrishna, Yixin Wan, Xiaomeng Jin, Kai-Wei Chang, Zhiqi Bu, Bhanukiran Vinzamuri, Volkan Cevher, Mingyi Hong, Rahul Gupta
| Challenge: | Unlearning aims to remove copyrighted, sensitive, or private content from large language models without a full retraining. |
| Approach: | They propose a multi-task unlearning benchmark LUME that unlearns short novels, biographies and public biographie . |
| Outcome: | The proposed benchmark unlearns short novels, biographies and public biographie . it also releases fine-tuned models with 1B and 7B parameter sizes as targets . |
Contextual Label Projection for Cross-Lingual Structured Prediction (2024.naacl-long)
Copied to clipboard
| Challenge: | Prior work favors simplified label translation or relying on word-level alignments for label projection. |
| Approach: | They propose a novel approach CLaP which translates text to target language and performs *contextual translation* on the labels using the translated text as the context. |
| Outcome: | The proposed approach improves translation accuracy on two prediction tasks and shows 2.4 F1 improvement for EAE and 1.4 F1 for named entity recognition. |
AVATAR: A Parallel Corpus for Java-Python Program Translation (2023.findings-acl)
Copied to clipboard
| Challenge: | Program translation is a time-consuming and costly process that requires expertise in both the source and target languages. |
| Approach: | They present a collection of 9,515 programming problems and their solutions written in Java and Python. |
| Outcome: | The proposed model lacks in generating functionally accurate code. |
Magnet: Multi-turn Tool-use Data Synthesis and Distillation via Graph Translation (2025.acl-long)
Copied to clipboard
Fan Yin, Zifeng Wang, I-Hung Hsu, Jun Yan, Ke Jiang, Yanfei Chen, Jindong Gu, Long Le, Kai-Wei Chang, Chen-Yu Lee, Hamid Palangi, Tomas Pfister
| Challenge: | Large language models have been shown to be effective in multi-turn interactions . however, their performance may be limited in complex, multi-turned interactions involving users and multiple tools. |
| Approach: | They propose a framework for synthesizing high-quality training trajectories to enhance the function calling capability of large language model agents in multi-turn conversations with humans. |
| Outcome: | The proposed model outperforms the teacher model by 68.01 on BFCL-v3 and 73.30 on ToolQuery. |
BRIEF: Bridging Retrieval and Inference for Multi-hop Reasoning via Compression (2025.findings-naacl)
Copied to clipboard
| Challenge: | Existing approaches to augment language models with external knowledge but they are limited by static nature of pre-training data. |
| Approach: | They propose a lightweight approach that compresses retrieved documents into highly dense textual summaries to integrate into in-context RAG. |
| Outcome: | The proposed approach reduces latency and costs while achieving high performance in open-domain questions. |
Data Advisor: Dynamic Data Curation for Safety Alignment of Large Language Models (2024.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies have explored using LLMs for efficient data collection. |
| Approach: | They propose a method that takes into account the characteristics of the desired dataset and monitors the status of the generated data. |
| Outcome: | The proposed method improves safety and quality of three representative large language models against safety issues without sacrificing model utility. |
Improving Zero-Shot Cross-Lingual Transfer Learning via Robust Training (2021.emnlp-main)
Copied to clipboard
| Challenge: | Pre-trained multilingual language encoders do not precisely align words and phrases across languages. |
| Approach: | They propose a learning strategy for training robust models by drawing connections between adversarial examples and failure cases of zero-shot cross-lingual transfer. |
| Outcome: | The proposed model can achieve good performance even if representations of different languages are not aligned well. |
Generating Syntactically Controlled Paraphrases without Using Annotated Parallel Pairs (2021.eacl-main)
Copied to clipboard
| Challenge: | Paraphrase generation requires many annotated paraphrase pairs, which are expensive to obtain. |
| Approach: | They propose a model that learns to disentangle the semantics and syntax of a sentence from unannotated texts. |
| Outcome: | The proposed model learns to disentangle the semantics and syntax of a sentence from a collection of unannotated texts. |
Retrieval Enhanced Data Augmentation for Question Answering on Privacy Policies (2023.eacl-main)
Copied to clipboard
| Challenge: | Existing labeled datasets are heavily imbalanced, limiting the QA performance in this domain. |
| Approach: | They propose a question answering task that captures relevant text segments from unlabeled policy documents and expands the positive examples in the training set. |
| Outcome: | The proposed framework elevates the baseline by a large margin (10% F1) and achieves a new state-of-the-art F1 score of 50%. |
GeoMLAMA: Geo-Diverse Commonsense Probing on Multilingual Pre-Trained Language Models (2022.emnlp-main)
Copied to clipboard
| Challenge: | Recent work shows that Pre-trained Language Models store relational knowledge and utilize it for performing downstream tasks. |
| Approach: | They propose a benchmark dataset for probing the diversity of relational knowledge in multilingual PLMs. |
| Outcome: | The proposed dataset contains 3125 prompts in English, Chinese, Hindi, Persian, and Swahili . larger multilingual PLMs variants do not store geo-diverse concepts better than its smaller variant . |
MM-PoisonRAG: Disrupting Multimodal RAG with Local and Global Knowledge Poisoning Attacks (2026.acl-long)
Copied to clipboard
Hyeonjeong Ha, Qiusi Zhan, Jeonghwan Kim, Dimitrios Bralios, Saikrishna Sanniboina, Nanyun Peng, Kai-Wei Chang, Daniel Kang, Heng Ji
| Challenge: | Existing research exposes multimodal large language models to knowledge poisoning attacks . localized poisoning attack achieves up to 56% success rate even under restricted access . globalized poison attack completely disrupts model generation to 0% accuracy with just one poisoned content. |
| Approach: | They propose a framework to study the vulnerability of multimodal RAG under knowledge poisoning attacks. |
| Outcome: | The proposed framework exploits two new attack strategies on multimodal RAGs under knowledge poisoning. |
On the Robustness of Language Encoders against Grammatical Errors (2020.acl-main)
Copied to clipboard
| Challenge: | Pre-trained language encoders are effective in facilitating downstream natural language processing tasks, but they often assume training and test corpora are clean and it is unclear how the models behave when confronted with noisy input. |
| Approach: | They conduct adversarial attacks to simulate grammatical errors on clean text data. |
| Outcome: | The proposed model performs better when confronted with natural grammatical errors than when faced with noisy input. |
LLM-A*: Large Language Model Enhanced Incremental Heuristic Search on Path Planning (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing path planning algorithms suffer from significant computational and memory inefficiencies as the state space grows . large language models excel in environmental analysis but fall short in detailed spatial and temporal reasoning . |
| Approach: | They propose a new path planning method that synergistically combines A* and LLMs to improve pathfinding efficiency. |
| Outcome: | The proposed method improves pathfinding efficiency while maintaining integrity of path validity in large-scale scenarios. |
Conditional Supervised Contrastive Learning for Fair Text Classification (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Recent advances in natural language processing have demonstrated societal bias in existing NLP models. |
| Approach: | They propose to use contrastive learning to learn fair representations for text classification . they conduct experiments on two text datasets to demonstrate their methods are stable . |
| Outcome: | The proposed methods balancing task performance and bias mitigation are stable in different hyperparameter settings. |
Full-Duplex-Bench-v2: A Multi-Turn Evaluation Framework for Duplex Dialogue Systems with an Automated Examiner (2026.acl-short)
Copied to clipboard
Guan-Ting Lin, Shih-Yun Shan Kuan, Jiatong Shi, Kai-Wei Chang, Siddhant Arora, Shinji Watanabe, Hung-yi Lee
| Challenge: | Full-duplex speech agents are often half-duplice, alternating turns between user and system. |
| Approach: | They propose a streaming framework that integrates with an examiner that enforces staged goals under two pacing setups. |
| Outcome: | The framework reports fluency, multi-turn instruction following, and task-specific competence. |
Beyond Facts- Benchmarking Distributional Reading Comprehension in Large Language Models (2026.findings-acl)
Copied to clipboard
Pei-Fu Guo, Ya An Tsai, Chun-Chia Hsu, Kai-Xin Chen, Yun-Da Tsai, Kai-Wei Chang, Nanyun Peng, Mi-Yen Yeh, Shou-De Lin
| Challenge: | Existing reading comprehension benchmarks focus on factual information, but many real-world tasks require distributional knowledge expressed across text. |
| Approach: | They propose a reading comprehension benchmark for LLMs to evaluate their ability to infer distributional knowledge from natural language. |
| Outcome: | Experiments with multiple LLMs show that the model outperforms baselines, but performance varies widely across distribution types and characteristics. |
Training LLMs for Divide-and-Conquer Reasoning Elevates Test-Time Scalability (2026.acl-long)
Copied to clipboard
Xiao Liang, Zhong-Zhi Li, Zhenghao Lin, Eric Hanchen Jiang, Hengyuan Zhang, Yelong Shen, Kai-Wei Chang, Ying Nian Wu, Yeyun Gong, Weizhu Chen
| Challenge: | Large language models have demonstrated strong reasoning capabilities through step-by-step chain-of-thought (CoT) reasoning, but their strictly sequential nature constrains test-time scalability. |
| Approach: | They propose an end-to-end reinforcement learning framework to enhance LLMs' DAC-style reasoning capacity by decomposing a problem into subproblems and solving them sequentially. |
| Outcome: | The proposed model surpasses CoT by 8.6% and 6.3% on competition-level benchmarks and is available at the [github.com/MasterVito/DAC-RL]. |
LiveCLKTBench: Towards Reliable Evaluation of Cross-Lingual Knowledge Transfer in Multilingual LLMs (2026.acl-long)
Copied to clipboard
Pei-Fu Guo, Yun-Da Tsai, Chun-Chia Hsu, Kai-Xin Chen, Ya An Tsai, Kai-Wei Chang, Nanyun Peng, Mi-Yen Yeh, Shou-De Lin
| Challenge: | Evaluating cross-lingual knowledge transfer in large language models is challenging, as correct answers in a target language may arise either from genuine transfer or from prior exposure during pre-training. |
| Approach: | They propose a pipeline to isolate and measure cross-lingual knowledge transfer by identifying self-contained, time-sensitive knowledge entities from real-world domains and generating factual questions. |
| Outcome: | The proposed pipeline analyzes multiple LLMs across five languages and shows that cross-lingual transfer is strongly influenced by linguistic distance and often asymmetric across language directions. |
Mitigating Gender Bias in Natural Language Processing: Literature Review (P19-1)
Copied to clipboard
Tony Sun, Andrew Gaut, Shirlyn Tang, Yuxin Huang, Mai ElSherief, Jieyu Zhao, Diba Mirza, Elizabeth Belding, Kai-Wei Chang, William Yang Wang
| Challenge: | NLP models propagate and may even amplify gender bias found in text corpora . methods to mitigate gender bias in NLP are relatively nascent . |
| Approach: | They propose to analyze gender bias based on four forms of representation bias and discuss the advantages and drawbacks of existing gender debiasing methods. |
| Outcome: | The proposed methods are based on four forms of representation bias and have advantages and drawbacks. |
Visualizing Trends of Key Roles in News Articles (D19-3)
Copied to clipboard
| Challenge: | a demonstration system visualizes news trend of key roles based on natural language processing techniques . semantic role labelling and word embeddings can help users understand news topics . |
| Approach: | They propose a system that visualizes the news trend of key roles based on natural language processing techniques. |
| Outcome: | The proposed system analyzes the news trend of key roles using semantic role labelling . it also analyzes how similarities between key roles and news topics change over time . |
Gender Bias in Contextualized Word Embeddings (N19-1)
Copied to clipboard
| Challenge: | Existing studies show that training word embeddings in large corpora could lead to encoding societal biases present in these human-produced data. |
| Approach: | They conduct several intrinsic analyses to quantify, analyze and mitigate gender bias exhibited in ELMo’s contextualized word vectors. |
| Outcome: | The proposed method mitigates gender bias on WinoBias probing corpus and demonstrates that it can be implemented in other systems. |
IdealGPT: Iteratively Decomposing Vision and Language Reasoning via Large Language Models (2023.findings-emnlp)
Copied to clipboard
Haoxuan You, Rui Sun, Zhecan Wang, Long Chen, Gengyu Wang, Hammad Ayyubi, Kai-Wei Chang, Shih-Fu Chang
| Challenge: | Existing approaches to decompose VL reasoning rely on domain-specific sub-question decomposing models. |
| Approach: | They propose a framework that iteratively decomposes VL reasoning using large language models. |
| Outcome: | The proposed framework outperforms existing models on multiple VL reasoning tasks. |
SPEED++: A Multilingual Event Extraction Framework for Epidemic Prediction and Preparedness (2024.emnlp-main)
Copied to clipboard
Tanmay Parekh, Jeffrey Kwan, Jiarui Yu, Sparsh Johri, Hyosang Ahn, Sreya Muppalla, Kai-Wei Chang, Wei Wang, Nanyun Peng
| Challenge: | Prior studies focused on English posts to provide early warnings for epidemic prediction, but these work focused on non-English posts. |
| Approach: | They propose a multilingual event extraction framework for extracting epidemic event information for any disease and language using 5.1K tweets in four languages. |
| Outcome: | The proposed framework can provide epidemic warnings for COVID-19 in its earliest stages in Dec 2019 (3 weeks before global discussions) and aggregate community epidemic discussions like symptoms and cure measures, aiding misinformation detection and public attention monitoring. |
Efficient Shapley Values Estimation by Amortization for Text Classification (2023.acl-long)
Copied to clipboard
| Challenge: | Shapley Values are often estimated with a small number of stochastic model evaluations, but this can only be mitigated by aggregating thousands of model evaluation. |
| Approach: | They propose to combine a model with thousands of model evaluations to estimate Shapley Values without additional model evaluation. |
| Outcome: | The proposed model estimates Shapley Values accurately with up to 60 times speedup compared to traditional methods and does not suffer from stability issues as inference is deterministic. |
Bias and Fairness in Natural Language Processing (D19-2)
Copied to clipboard
| Challenge: | a tutorial will review the history of bias and fairness studies in machine learning and language processing . |
| Approach: | This tutorial reviews the history of bias and fairness studies in machine learning and language processing . it presents recent community effort to quantify and mitigat bias in natural language processing models . |
| Outcome: | This tutorial reviews the history of bias and fairness studies in machine learning and language processing . it aims to quantify and mitigate bias in natural language processing models for a wide spectrum of tasks . |
On Leveraging Encoder-only Pre-trained Language Models for Effective Keyphrase Generation (2024.lrec-main)
Copied to clipboard
| Challenge: | a new study examines the use of encoder-only pre-trained language models in keyphrase generation (KPG) keyphrases are phrases that condense salient information of a document. |
| Approach: | They propose to use encoder-only pre-trained language models in keyphrase generation . they also examine optimal architectural decisions for employing encoder only PLMs in KPG . |
| Outcome: | The proposed model outperforms general-domain seq2seq models in keyphrase generation. |
Understanding and Mitigating Spurious Correlations in Text Classification with Neighborhood Analysis (2024.findings-eacl)
Copied to clipboard
| Challenge: | Recent research has revealed that machine learning models have a tendency to leverage spurious correlations that exist in the training set but may not hold true in general circumstances. |
| Approach: | They propose a metric to detect spurious tokens and a family of regularization methods to mitigate spurious correlations in text classification. |
| Outcome: | The proposed method prevents spurious clusters and significantly improves the robustness of classifiers without auxiliary data. |
VDebugger: Harnessing Execution Feedback for Debugging Visual Programs (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Visual programs are executable code generated by large language models to address visual reasoning problems. |
| Approach: | They propose a critic-refiner framework that localizes and debugs visual programs by tracking execution step by step. |
| Outcome: | The proposed framework detects and corrects program errors leveraging detailed execution feedback, improving interpretability and accuracy. |
Empowering Language Models with Knowledge Graph Reasoning for Open-Domain Question Answering (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing Language Models lack the power to store all required knowledge, resulting in a lack of ability to infer out-of-context knowledge. |
| Approach: | They propose a Knowledge Interaction Layer that can be flexibly plugged into existing Transformer-based LMs to interact with a differentiable Knowledge Graph Reasoning module collaboratively. |
| Outcome: | The proposed model can be plugged into existing Transformer-based LMs to interact with a differentiable Knowledge Graph Reasoning module collaboratively. |
Control Large Language Models via Divide and Conquer (2024.emnlp-main)
Copied to clipboard
| Challenge: | Lexically Constrained Generation (LCG) is a crucial task of text generation. |
| Approach: | They propose a Divide and Conquer Generation strategy to enhance LLMs' performance in Lexically Constrained Generation with prompt-based controlling. |
| Outcome: | The proposed strategy shows 90% improvement on the most challenging LCG task. |
Learning to Discriminate Perturbations for Blocking Adversarial Attacks in Text Classification (D19-1)
Copied to clipboard
| Challenge: | Existing studies on adversarial attacks on deep learning models focus on generation of adversarials and defense against adversarial attacks. |
| Approach: | They propose a framework to identify and adjust malicious perturbations and block adversarial attacks for machine learning models. |
| Outcome: | The proposed framework outperforms baseline methods in blocking adversarial attacks for text classification models. |
InsideOut: Measuring and Mitigating Insider–Outsider Bias in Interview Script Generation (2026.acl-long)
Copied to clipboard
| Challenge: | Recent research has raised concerns about culture-related fairness issues in LLM-generated content. |
| Approach: | They propose to use 4,000 generation prompts and three evaluation metrics to quantify LLMs' **insider-outsider bias** . |
| Outcome: | The proposed method reduces bias in Llama model by 89.70% and mitigates bias on Qwen by 82.54% on cultural alignment gap metric. |
Harms of Gender Exclusivity and Challenges in Non-Binary Representation in Language Technologies (2021.emnlp-main)
Copied to clipboard
| Challenge: | Recent work analyzes, quantifies, and mitigates language model biases such as gender, race or religion-related stereotypes in static word embeddings and contextual representations. |
| Approach: | They explain the complexity of gender and language around it and examine how current representations perpetuate harms associated with binary gender. |
| Outcome: | The proposed model and dataset biases perpetuate harms associated with the treatment of gender as binary in English language technologies. |
Syntax-augmented Multilingual BERT for Cross-lingual Transfer (2021.acl-long)
Copied to clipboard
| Challenge: | Existing studies show that pre-trained multilingual text encoders capture language syntax, helping cross-lingual transfer. |
| Approach: | They provide language syntax and train mBERT to encode universal dependency tree structure. |
| Outcome: | The proposed model improves cross-lingual transfer on PAWS-X and MLQA benchmarks by 1.4 and 1.6 points on average across all languages. |
ParaAMR: A Large-Scale Syntactically Diverse Paraphrase Dataset by AMR Back-Translation (2023.acl-long)
Copied to clipboard
| Challenge: | Paraphrase generation is a long-standing task in natural language processing (NLP). |
| Approach: | They propose to generate large-scale syntactically diverse paraphrase datasets by abstract meaning representation back-translation. |
| Outcome: | The proposed dataset is syntactically more diverse than existing datasets while maintaining good semantic similarity. |
SNaRe: Domain-aware Data Generation for Low-Resource Event Detection (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for ED struggle with label noise and domain drift when applied to specialized domains. |
| Approach: | They propose a domain-aware synthetic data generation framework composed of three components: Scout, Narrator, and Refiner. |
| Outcome: | The proposed framework outperforms baseline approaches on three diverse domain ED datasets and achieves average F1 gains of 3-7% in the zero-shot/few-shot settings and 4-20% improvement for multilingual generation. |
PIP: Parse-Instructed Prefix for Syntactically Controlled Paraphrase Generation (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing fine-tuning methods for this task are costly and require updating the parameters of the entire model to adapt to the newly included syntax information. |
| Approach: | They propose a method to instruct model’s encoder prefix to capture syntax-related knowledge by direct initiation and indirect optimization. |
| Outcome: | The proposed methods are 10 times more efficient and learnable than existing methods. |
Towards Understanding Gender Bias in Relation Extraction (2020.acl-main)
Copied to clipboard
Andrew Gaut, Tony Sun, Shirlyn Tang, Yuxin Huang, Jing Qian, Mai ElSherief, Jieyu Zhao, Diba Mirza, Elizabeth Belding, Kai-Wei Chang, William Yang Wang
| Challenge: | Existing bias mitigation techniques have a negative effect on NRE, a study finds . |
| Approach: | They create a dataset to analyze gender bias in relation extraction systems . they find that existing bias mitigation techniques have a negative effect on NRE . |
| Outcome: | The proposed dataset analyzes gender bias in relation extraction systems using a 10% human annotated test set. |
DiCoRe: Enhancing Zero-shot Event Detection via Divergent-Convergent LLM Reasoning (2025.emnlp-main)
Copied to clipboard
| Challenge: | Understanding the complex event ontology, extracting domain-specific triggers from the passage, and structuring them appropriately overloads and limits the utility of Large Language Models (LLMs). |
| Approach: | They propose a divergent-convergent reasoning framework that decouples the task of ED using Dreamer and Grounder. |
| Outcome: | The proposed framework outperforms baselines on six datasets across five domains and nine LLMs, achieving 4–7% average gains over the best baseline. |
SYNTHIA: Novel Concept Design with Affordance Composition (2025.acl-long)
Copied to clipboard
Hyeonjeong Ha, Xiaomeng Jin, Jeonghwan Kim, Jiateng Liu, Zhenhailong Wang, Khanh Duy Nguyen, Ansel Blume, Nanyun Peng, Kai-Wei Chang, Heng Ji
| Challenge: | Existing studies on concept design using text-to-image models have enabled rapid ideation of novel visual concepts. |
| Approach: | They propose a framework for generating novel, functionally coherent designs based on desired affordances by decomposing concepts into parts and affordance . they also develop a curriculum learning scheme that fine-tunes T2I models to progressively learn affordance composition while maintaining visual novelty. |
| Outcome: | The proposed framework outperforms state-of-the-art models for novelty and functional coherence in human evaluation. |
GENEVA: Benchmarking Generalizability for Event Argument Extraction with Hundreds of Event Types and Argument Roles (2023.acl-long)
Copied to clipboard
| Challenge: | Existing benchmarking datasets for Event Argument Extraction (EAE) cover less than 40 event types and 25 entity-centric argument roles. |
| Approach: | They propose to use a large and diverse EAE ontology to create a semantic role labeling dataset for EAE that incorporates 115 events and 220 argument roles. |
| Outcome: | The proposed ontology concludes with 115 events and 220 argument roles, with a significant portion of roles not being entities. |
What’s “up” with vision-language models? Investigating their struggle with spatial reasoning (2023.emnlp-main)
Copied to clipboard
| Challenge: | Recent work has re-surfaced a concern that has long plagued vision-language models: poor performance on simple tasks like attribute attachment, counting, etc. |
| Approach: | They evaluate 18 vision-language models and find they perform poorly on VQAv2 . they find that popular vision-linguistic pretraining corpora lack reliable data for learning spatial relationships . |
| Outcome: | The new models are compared with existing datasets on what'sup and visual-language models . they achieve 56% accuracy on the new benchmarks compared to 99% for humans . |
Active Instruction Tuning: Improving Cross-Task Generalization by Training on Prompt Sensitive Tasks (2023.emnlp-main)
Copied to clipboard
| Challenge: | Instruction tuning (IT) achieves impressive zero-shot generalization results by training large language models on diverse tasks with instructions. |
| Approach: | They propose a framework to identify informative tasks and then actively tune models on selected tasks. |
| Outcome: | The proposed method outperforms baseline strategies for task selection on NIV2 and Self-Instruct datasets. |
From Narrow Unlearning to Emergent Misalignment in LLMs (2026.acl-short)
Copied to clipboard
Erum Mushtaq, Anil Ramakrishna, Satyapriya Krishna, Sattvik Sahai, Prasoon Goyal, Kai-Wei Chang, Tao Zhang, Rahul Gupta
| Challenge: | Recent work shows that fine-tuning on insecure code data can trigger an emergent misalignment (EMA) phenomenon . |
| Approach: | They extend their study by demonstrating that EMA can arise from narrow refusal unlearning . they perform refusal unLearning on Cybersecurity and Safety concept and evaluate EMA . |
| Outcome: | The proposed model can generate malicious responses even to unrelated prompts . the proposed model is able to restore alignment across the affected domains while having lower refusal rates. |
On Localizing and Deleting Toxic Memories in Large Language Models (2025.findings-naacl)
Copied to clipboard
Anubrata Das, Manoj Kumar, Ninareh Mehrabi, Anil Ramakrishna, Anna Rumshisky, Kai-Wei Chang, Aram Galstyan, Morteza Ziyadi, Rahul Gupta
| Challenge: | Existing methods to reduce toxic generation in large language models are not fully understood. |
| Approach: | They propose to understand the mechanisms that drive toxic generation in large language models by using memory localization to reduce toxic generation. |
| Outcome: | The proposed method reduces toxic generation from 62.86% to 28.61%, but it also improves generation quality. |
Learning Gender-Neutral Word Embeddings (D18-1)
Copied to clipboard
| Challenge: | Word embeddings trained on human-generated corpora inherit strong gender stereotypes . prior studies show such embeddables exhibit social biases, such as gender stereotype . |
| Approach: | They propose a method to preserve gender information in certain dimensions of word vectors . they propose GN-GloVe, which is a gender-neutral variant of the word embedding model . |
| Outcome: | The proposed method preserves gender information in certain dimensions of word vectors while compelling other dimensions to be free of gender influence. |
Unlearning as multi-task optimization: A normalized gradient difference approach with an adaptive learning rate (2025.naacl-long)
Copied to clipboard
Xiaomeng Jin, Zhiqi Bu, Bhanukiran Vinzamuri, Anil Ramakrishna, Kai-Wei Chang, Volkan Cevher, Mingyi Hong
| Challenge: | Existing methods to remove unwanted knowledge from large language models are formulated as minimizing memorization through the loss of the model. |
| Approach: | They propose a normalized gradient difference algorithm that optimizes a forgetting objective and an automatic learning rate scheduler that allows for better control over the trade-off between the objectives. |
| Outcome: | The proposed method improves on TOFU and MUSE datasets while exhibiting stable training. |
CASA: Causality-driven Argument Sufficiency Assessment (2024.naacl-long)
Copied to clipboard
| Challenge: | Existing methods to assess the sufficiency of arguments are laborious and inconsistent due to subjective criteria. |
| Approach: | They propose a causality-driven argument sufficiency assessment framework that uses the probability of sufficience to estimate the probability that a premise event would lead to a conclusion when both premise and conclusion events are absent. |
| Outcome: | The proposed framework identifies insufficient arguments and improves them in a writing aid application. |
ADDMU: Detection of Far-Boundary Adversarial Examples with Data and Model Uncertainty Estimation (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods show poor performance under Far Boundary (FB) adversarial examples. |
| Approach: | They propose to use a new technique to detect adversarial examples based on data and model uncertainty to outperform existing methods. |
| Outcome: | The proposed method outperforms existing methods by 3.6 and 6.0 AUC points under each scenario. |
Rethinking Model Selection and Decoding for Keyphrase Generation with Pre-trained Sequence-to-Sequence Models (2023.emnlp-main)
Copied to clipboard
| Challenge: | Keyphrase generation is a longstanding task in NLP with widespread applications. |
| Approach: | They propose a likelihood-based decode-select algorithm for seq2seq PLMs that improves greedy search by an average of 4.7% semantic F1 across five datasets. |
| Outcome: | The proposed algorithm improves greedy search by an average of 4.7% semantic F1 across five datasets. |
On the Intrinsic and Extrinsic Fairness Evaluation Metrics for Contextualized Language Representations (2022.acl-short)
Copied to clipboard
Yang Trista Cao, Yada Pruksachatkun, Kai-Wei Chang, Rahul Gupta, Varun Kumar, Jwala Dhamala, Aram Galstyan
| Challenge: | Recent natural language processing systems use large language models as the backbone . however, societal biases are encoded in these models and transferred to downstream applications . |
| Approach: | They propose to use two categories to measure fairness in natural language processing tasks . they find intrinsic and extrinsic metrics do not correlate in their original setting . |
| Outcome: | The proposed metrics do not correlate in their original setting, the authors show . they find that they are not accurate when correcting for metric misalignments and noise . |
Cross-Lingual Dependency Parsing by POS-Guided Word Reordering (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to cross-lingual dependency parsing rely on large corpus size and cost. |
| Approach: | They propose a cross-lingual dependency parsing approach based on word reordering . they propose to train a model that transfers knowledge learned in one or multiple languages to target languages . |
| Outcome: | The proposed approach outperforms the baseline approach in Hindi and Latin by 15.3% and 6.7%. |
“Kelly is a Warm Person, Joseph is a Role Model”: Gender Biases in LLM-Generated Reference Letters (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are an effective tool to assist individuals in writing documents. |
| Approach: | They examine gender biases in large language models (LLMs)-generated reference letters . they find that models are biased because they are hallucinated . |
| Outcome: | The proposed model-generated reference letters are evaluated on 2 popular LLMs- ChatGPT and Alpaca. |
Agent Lumos: Unified and Modular Training for Open-Source Language Agents (2024.acl-long)
Copied to clipboard
Da Yin, Faeze Brahman, Abhilasha Ravichander, Khyathi Chandu, Kai-Wei Chang, Yejin Choi, Bill Yuchen Lin
| Challenge: | Lumos is a framework for training open-source agents on complex interactive tasks. |
| Approach: | They propose a framework for training open-source LLM-based agents called Lumos . Lumos features a learnable, unified and modular architecture with a planning module that learns high-level subgoal generation and a grounding module trained to translate these into the actions using various tools in the execution module. |
| Outcome: | The framework outperforms open-source agents on QA and web tasks. |
MACAROON: Training Vision-Language Models To Be Your Engaged Partners (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Large vision-language models (LVLMs) generate detailed responses even when questions are ambiguous or unanswerable, leading to hallucinations and bias issues. |
| Approach: | They propose a three-tiered hierarchy for questions of invalid, ambiguous, and personalizable nature to measure the proactive engagement capabilities of LVLMs. |
| Outcome: | The proposed model generates contrastive response pairs for unlabeled questions, achieving 0.84 AAR, while maintaining comparable performance on general tasks. |
Target Language-Aware Constrained Inference for Cross-lingual Dependency Parsing (D19-1)
Copied to clipboard
| Challenge: | Existing work on cross-lingual dependency parsing focuses on capturing commonalities between source and target languages and overlooking the potential to leverage the linguistic properties of the target languages to facilitate the transfer. |
| Approach: | They propose to use Lagrangian relaxation and posterior regularization techniques to conduct inference with corpus-statistics constraints to capture commonalities between source and target languages. |
| Outcome: | The proposed algorithms improve on 15 and 17 out of 19 target languages. |
LACMA: Language-Aligning Contrastive Learning with Meta-Actions for Embodied Instruction Following (2023.emnlp-main)
Copied to clipboard
| Challenge: | Embodied Instruction Following has shown an impressive success rate when the environment has been seen in training, but when deployed in an unseen environment, it tends to struggle when deployed with an unsightly environment. |
| Approach: | They propose to explicitly align the agent’s hidden states with the instructions via contrastive learning to bridge the semantic gap between high-level language instructions and the agent's low-level action space. |
| Outcome: | The proposed meta-actions achieve a 4.5% success rate in unseen environments compared to a strong multi-modal Transformer baseline . |
The steerability of large language models toward data-driven personas (2024.naacl-long)
Copied to clipboard
Junyi Li, Charith Peris, Ninareh Mehrabi, Palash Goyal, Kai-Wei Chang, Aram Galstyan, Richard Zemel, Rahul Gupta
| Challenge: | Large language models generate biased responses where opinions of certain groups and populations are underrepresented. |
| Approach: | They propose a data-driven notion of persona that allows for a more nuanced understanding of different (latent) social groups present in the population. |
| Outcome: | The proposed method improves model steerability by 57% over baselines. |
On the Transferability of Adversarial Attacks against Neural Text Classifier (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies show that deep neural networks are vulnerable to adversarial examples . a small perturbation to an input alters the model prediction . |
| Approach: | They propose a genetic algorithm to find models that can induce adversarial examples to fool models . they propose word replacement rules that can be used for model diagnostics from these examples . |
| Outcome: | The proposed model can fool almost all existing models, while ignoring the data bias in the training set. |
Unified Pre-training for Program Understanding and Generation (2021.naacl-main)
Copied to clipboard
| Challenge: | PLUG is a programming language that is used for programming and language understanding and generation tasks. |
| Approach: | They propose a sequence-to-sequence model that performs a broad spectrum of program and language understanding and generation tasks. |
| Outcome: | The proposed model outperforms or rivals state-of-the-art models on code summarization, code generation, and code translation tasks in seven programming languages. |
PLUE: Language Understanding Evaluation Benchmark for Privacy Policies in English (2023.acl-short)
Copied to clipboard
| Challenge: | Existing efforts to understand privacy policies are limited by processing the language in a way exclusive to a single task focusing on certain privacy practices. |
| Approach: | They propose a privacy policy language understanding evaluation benchmark to evaluate the understanding of privacy policies across multiple tasks. |
| Outcome: | The proposed framework improves the understanding of privacy policies across multiple tasks. |
BRIEF-Pro: Universal Context Compression with Short-to-Long Synthesis for Fast and Accurate Multi-Hop Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | Experiments show that BRIEF-Pro generates more concise and relevant summaries, enhancing performance across small, large, and proprietary language models. |
| Approach: | They propose a universal, lightweight compressor that distills relevant evidence from retrieved documents into a concise summary for seamless integration into in-context RAG. |
| Outcome: | Experiments on four open-domain multi-hop question-answering datasets show that BRIEF-Pro generates more concise and relevant summaries, enhancing performance across small, large, and proprietary language models. |
Vulnerability of LLMs to Vertically Aligned Text Manipulations (2025.acl-long)
Copied to clipboard
| Challenge: | Recent research shows that vertical text input significantly degrades the accuracy of large language models (LLMs) in text classification tasks. |
| Approach: | They investigate the impact of vertical text input on the performance of LLMs . they find that chain of thought reasoning does not help LLM recognize vertical input . |
| Outcome: | The proposed model can significantly mislead models, posing a risk of bypassing detection in real-world scenarios involving harmful or sensitive information. |
Can Small Language Models Help Large Language Models Reason Better?: LM-Guided Chain-of-Thought (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing frameworks for guiding a language model in reasoning tasks are limited by their tendency to generate low-quality rationales that are repetitive and vacuous. |
| Approach: | They propose a framework that leverages a lightweight language model for guiding a black-box large LM in reasoning tasks. |
| Outcome: | The proposed framework outperforms baselines in answer prediction accuracy. |
On Measures of Biases and Harms in NLP (2022.findings-aacl)
Copied to clipboard
Sunipa Dev, Emily Sheng, Jieyu Zhao, Aubrie Amstutz, Jiao Sun, Yu Hou, Mattie Sanseverino, Jiin Kim, Akihiro Nishi, Nanyun Peng, Kai-Wei Chang
| Challenge: | Recent studies show that natural language processing (NLP) technologies propagate societal biases about demographic groups associated with attributes such as gender, race, and nationality. |
| Approach: | They propose a framework for harms and questions to help practitioners understand biases . they propose measurable measures to detect and mitigate biased groups . |
| Outcome: | The proposed framework provides a framework for harms and questions for practitioners to answer to guide the development of bias measures. |
Find Someone Who: Visual Commonsense Understanding in Human-Centric Grounding (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Visual scenes often involve multiple people and humans can distinguish between them based on context descriptions about what happened before, their mental/physical states, and intentions. |
| Approach: | They propose a task that tests human-centric commonsense grounding models' ability to distinguish individuals given context descriptions about what happened before and their mental/physical states or intentions. |
| Outcome: | The proposed model outperforms pre-trained and non-pretrained models on 130k commonsense descriptions annotated on 67k images. |
Adapting Coreference Resolution for Processing Violent Death Narratives (2021.naacl-main)
Copied to clipboard
| Challenge: | Existing coreference models suffer from poortransferability due to domain gaps . existing models are not robust enough to handle text data about LGBT individuals . |
| Approach: | They propose to use a dataaugmentation rule to improve coreference resolution in an administrative database written in English to better handle LGBT data. |
| Outcome: | The proposed model improves perfor-mance and accuracy of coreference resolution in a violent death nar-rative from the Centers for Disease Control's (CDC) national Violent Death Re-porting System. |
Dynamic Generation of Multi LLM Agents Communication Topologies with Graph Diffusion Models (2026.acl-long)
Copied to clipboard
Eric Hanchen Jiang, Levina Li, Frank Wan, Xiao Liang, Sophia Yin, Yuchen Wu, Xinfeng Li, Yizhou Sun, Wei Wang, Kai-Wei Chang, Ying Nian Wu
| Challenge: | Existing frameworks rely on static or rule-based topologies that fail to adapt to task requirements. |
| Approach: | They propose a generative framework that generates highly task-adaptive topologies . they validated the framework on multiple benchmarks and validated it on multiple platforms . |
| Outcome: | The proposed framework outperforms existing frameworks in task-adaptive communication topologies. |
Mitigating Over-Refusal in Aligned Large Language Models via Inference-Time Activation Energy (2026.acl-long)
Copied to clipboard
Eric Hanchen Jiang, Weixuan Ou, Run Liu, Shengyuan Pang, Guancheng Wan, Ranjie Duan, Wei Dong, Kai-Wei Chang, XiaoFeng Wang, Ying Nian Wu, Xinfeng Li
| Challenge: | Existing safety alignment techniques prioritize mitigating harmful responses at the expense of overcautious behavior, leading models to incorrectly refuse benign requests. |
| Approach: | They propose a fine-tuning free framework to improve safety and reduce false refusals by dynamic, inference-time intervention. |
| Outcome: | The proposed framework raises compliance on the ORB-H benchmark from 57.3% to 82.6% while maintaining the baseline safety performance. |
SentiBERT: A Transferable Transformer-Based Architecture for Compositional Sentiment Semantics (2020.acl-main)
Copied to clipboard
| Challenge: | Existing models that capture compositional sentiment semantics are difficult to handle when the sentences are complex. |
| Approach: | They propose a variant of BERT that captures compositional sentiment semantics . they demonstrate that SentiBERT can be applied to other sentiment analysis tasks . |
| Outcome: | The proposed model is better than baseline approaches in capturing negation and contrast . it can be applied to other sentiment analysis tasks and emotion classification tasks . |
Attribute Controlled Fine-tuning for Large Language Models: A Case Study on Detoxification (2024.findings-emnlp)
Copied to clipboard
Tao Meng, Ninareh Mehrabi, Palash Goyal, Anil Ramakrishna, Aram Galstyan, Richard Zemel, Kai-Wei Chang, Rahul Gupta, Charith Peris
| Challenge: | Using a sequence-level constraint, we regularize the LLMtraining by penalizing the KL divergence between the desired output distribution and the LRM’s posterior. |
| Approach: | They propose a constraint learning schema forfine-tuning Large Language Models with attribute control by penalizing the KL divergence be-tween the desired output distribution and the LLM's posterior. |
| Outcome: | The proposed approach improves the performance of large language models while enhancing their utility and generation quality. |
Weight Perturbation as Defense against Adversarial Word Substitutions (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existence and pervasiveness of textual adversarial examples have raised serious concerns to security-critical applications. |
| Approach: | They propose to perform weight perturbations in the parameter space rather than the input feature space to improve adversarial robustness of NLP models. |
| Outcome: | The proposed method improves adversarial robustness of models by performing weight perturbations in the parameter space rather than the input feature space. |
Broaden the Vision: Geo-Diverse Visual Commonsense Reasoning (2021.emnlp-main)
Copied to clipboard
| Challenge: | Generally, commonsense knowledge is correlated with culture and geographic locations and is only shared locally. |
| Approach: | They construct a Geo-Diverse Visual Commonsense Reasoning dataset to test vision-and-language models’ ability to understand cultural and geo-location-specific commonsense. |
| Outcome: | The proposed models perform better in non-Western regions including East Asia, South Asia, and Africa than in the Western regions. |
Not Every Token Needs Forgetting: Selective Unlearning Balancing Forgetting and Utility in Large Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Conventional unlearning approaches forget all tokens in a target document, including common tokens that carry general knowledge. |
| Approach: | They propose a method that identifies a critical subset of tokens within the forgetting set that is relevant to the unwanted information and unlearns only those tokens. |
| Outcome: | Experiments on two benchmarks and six baseline unlearning algorithms show that selective unlearning achieves effective unlearning on the targeted forget data. |
Unsupervised Vision-and-Language Pre-training Without Parallel Images and Captions (2021.naacl-main)
Copied to clipboard
| Challenge: | Existing models require large amounts of image-caption data for pre-training . existing models require expensive data collection and curation . |
| Approach: | They propose to conduct "mask-and-predict" pre-training on text-only and image-only corpora and introduce the object tags detected by an object recognition model as anchor points to bridge two modalities. |
| Outcome: | The proposed approach achieves performance close to a model pre-trained with aligned data, on four English benchmarks. |
FLIRT: Feedback Loop In-context Red Teaming (2024.emnlp-main)
Copied to clipboard
Ninareh Mehrabi, Palash Goyal, Christophe Dupuy, Qian Hu, Shalini Ghosh, Richard Zemel, Kai-Wei Chang, Aram Galstyan, Rahul Gupta
| Challenge: | Recent work has evaluated the vulnerabilities of large generative models, such as DALL-E, ChatGPT, and GPT-4. |
| Approach: | They propose an automatic red teaming framework that evaluates a given black-box model and exposes its vulnerabilities against unsafe and inappropriate content generation. |
| Outcome: | The proposed framework evaluates a given black-box model and exposes its vulnerabilities against unsafe and inappropriate content generation. |
QUDSELECT: Selective Decoding for Questions Under Discussion Parsing (2024.emnlp-main)
Copied to clipboard
| Challenge: | Question Under Discussion (QUD) uses implicit questions to reveal discourse relationships between sentences. |
| Approach: | They propose a framework that selectively decodes the QUD dependency structures considering the QUC criteria. |
| Outcome: | The proposed framework outperforms the state-of-the-art baseline models by 9% in human evaluation and 4% in automatic evaluation. |
Learning Word Embeddings for Low-Resource Languages by PU Learning (N18-1)
Copied to clipboard
| Challenge: | Existing approaches to learn word embedding on a corpus with only a few million tokens are limited to low-resource languages. |
| Approach: | They propose to use a sparse co-occurrence matrix to factorize the co-existence matrix and validate the proposed approaches in four different languages. |
| Outcome: | The proposed model is validated in four different languages. |
Gender Bias in Coreference Resolution: Evaluation and Debiasing Methods (N18-2)
Copied to clipboard
| Challenge: | Existing methods for co-reference resolution focus on gender bias. |
| Approach: | They propose a new benchmark for co-reference resolution focused on gender bias, WinoBias. |
| Outcome: | The proposed system removes the bias without significantly affecting performance on existing datasets. |
Investigating Ensemble Methods for Model Robustness Improvement of Text Classifiers (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods to reduce model's reliance on bias features ignore the learnability of these features. |
| Approach: | They propose to reduce models' reliance on bias features by first training models with fixed low-capacity models which ignore the learnability of the bias features. |
| Outcome: | The proposed models can perform better on out-of-distribution datasets than baseline models with a more sophisticated model design. |
Intent Classification and Slot Filling for Privacy Policies (2021.acl-long)
Copied to clipboard
| Challenge: | Sentences written in privacy policies explain privacy practices and the constituent text spans convey further specific information. |
| Approach: | They propose an English corpus of 5,250 intent and 11,788 slot annotations . they propose two alternative neural approaches to model the corpus as a sequence-to-sequence learning task. |
| Outcome: | The proposed corpus predicts intent classification and slot filling, while the sequence tagging method outperforms slot filler by a large margin. |
Tokenization Matters: Navigating Data-Scarce Tokenization for Gender Inclusive Language Technologies (2024.findings-naacl)
Copied to clipboard
Anaelia Ovalle, Ninareh Mehrabi, Palash Goyal, Jwala Dhamala, Kai-Wei Chang, Richard Zemel, Aram Galstyan, Yuval Pinter, Rahul Gupta
| Challenge: | a recent study documented the harmful limitations of gender binary-centric large language models . data scarcity is a known culprit, but the precise mechanisms through which scarcity affects this behavior remain underexplored. |
| Approach: | They propose to use BPE tokenization to enforce consistent tokenization across gendered pronouns to improve neopronoun proficiency. |
| Outcome: | The proposed methods outperform finetuning with standard BPE, and improve neopronoun proficiency. |
On the Fallacy of Global Token Perplexity in Spoken Language Model Evaluation (2026.findings-acl)
Copied to clipboard
Chan-Jan Hsu, Liang-Hsuan Tseng, Yi-Cheng Lin, Yen-Chun Kuo, Ju-Chieh Chou, Kai-Wei Chang, Hung-yi Lee, Carlos Busso
| Challenge: | Generative spoken language models are often evaluated using global token perplexity, which overlooks fundamental differences between speech and text modalities. |
| Approach: | They propose a variety of likelihood- and generative-based evaluation methods that serve in place of naive global token perplexity. |
| Outcome: | The proposed evaluations more faithfully reflect perceived generation quality, as evidenced by stronger correlations with human-rated mean opinion scores (MOS). |
Re-ReST: Reflection-Reinforced Self-Training for Language Agents (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods to fine tune language agents with reasoning-action trajectories require high-quality model-generated samples, which are hard to obtain for challenging language agent tasks. |
| Approach: | They propose a method to employ reflection during inference without ground-truth feedback to improve agents more autonomously. |
| Outcome: | The proposed method improves self-training performance on open-source language agents by 7.6% and 14.1% respectively. |
The Factuality Tax of Diversity-Intervened Text-to-Image Generation: Benchmark and Fact-Augmented Intervention (2024.emnlp-main)
Copied to clipboard
| Challenge: | Prompt-based “diversity interventions” are commonly adopted to improve the diversity of Text-to-Image models depicting individuals with diverse racial or gender traits. |
| Approach: | They propose a benchmark to quantify the trade-off between using diversity interventions and preserving demographic factuality in T2I models. |
| Outcome: | The proposed model significantly improves the demographic factuality under diversity interventions while preserving diversity. |
Does Robustness Improve Fairness? Approaching Fairness with Word Substitution Robustness Methods for Text Classification (2021.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to reduce disparities in model outcomes have focused on data augmentation, debiasing model embeddings, or adding fairness-based optimization objectives during training. |
| Approach: | They propose to use certified word substitution robustness methods to improve equality of odds and equality of opportunity on multiple text classification tasks. |
| Outcome: | The proposed methods improve equality of odds and equality of opportunity on multiple text classification tasks. |
Gender Bias in Multilingual Embeddings and Cross-Lingual Transfer (2020.acl-main)
Copied to clipboard
| Challenge: | Multilingual word embeddings embed words from many languages into a single semantic space such that words with similar meanings are close to each other regardless of the language. |
| Approach: | They propose to use multilingual word embeddings to align embeddable words from multiple languages into a single semantic space so that words with similar meanings are close to each other regardless of the language. |
| Outcome: | The proposed model can be used to learn gender bias in multilingual representations and to improve transfer learning. |
Mitigating Gender Bias in Distilled Language Models via Counterfactual Role Reversal (2022.findings-acl)
Copied to clipboard
Umang Gupta, Jwala Dhamala, Varun Kumar, Apurv Verma, Yada Pruksachatkun, Satyapriya Krishna, Rahul Gupta, Kai-Wei Chang, Greg Ver Steeg, Aram Galstyan
| Challenge: | Language models excel at generating coherent text, but can be biased in multiple ways, including the unfounded association of male and female genders with gender-neutral professions. |
| Approach: | They propose to modify teacher probabilities and augment the training set to learn a fair model during knowledge distillation by modifying teacher probability and augmenting the training sets. |
| Outcome: | The proposed approach reduces gender disparity in open-ended text generated from the distilled and finetuned models with only a minor compromise in utility. |
Relation-Guided Pre-Training for Open-Domain Question Answering (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Existing QA datasets are imbalanced in some types of relations, which hurts generalization performance over long-tail questions. |
| Approach: | They propose a relation-guided pre-training framework to infer latent relations from a QA dataset . they then propose RGPT-QA to conduct extractive QA to get the target answer entity . |
| Outcome: | The proposed framework improves Exact Match accuracy on natural questions, TriviaQA, and WebQuestions. |
Double Perturbation: On the Robustness of Robustness and Counterfactual Bias Evaluation (2021.naacl-main)
Copied to clipboard
| Challenge: | Recent studies show that NLP models are vulnerable to adversarial perturbations such as synonym substitutions or syntax-guided paraphrasing. |
| Approach: | They propose a “double perturbation” framework to uncover model weaknesses beyond the test dataset. |
| Outcome: | The proposed attack achieves high success rates on both original and robustly trained CNNs and Transformers. |
PolicyQA: A Reading Comprehension Dataset for Privacy Policies (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Privacy policy documents are long and verbose. Hence, a question answering system can help users find the information that is relevant and important to them. |
| Approach: | They propose to provide users with a short text span from policy documents to search for answers from a long text segment. |
| Outcome: | The proposed question answering system can help users find information relevant to them. |
KPEval: Towards Fine-Grained Semantic-Based Keyphrase Evaluation (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing evaluation methods for keyphrase extraction and generation rely on exact matching with human references. |
| Approach: | They propose a framework for evaluation that includes four critical aspects: reference agreement, faithfulness, diversity, utility and semantic-based metrics. |
| Outcome: | The proposed evaluation framework correlates better with human preferences than previously proposed metrics. |
Dynosaur: A Dynamic Growth Paradigm for Instruction-Tuning Data Curation (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for instruction tuning do not include associating instructions with existing datasets. |
| Approach: | They propose a dynamic growth paradigm for the automatic curation of instruction-tuning data . they use existing datasets to automatically construct instruction-uning datasets . |
| Outcome: | The proposed model reduces the API cost for generating instructions and provides high-quality data. |
Defense against Synonym Substitution-based Adversarial Attacks via Dirichlet Neighborhood Ensemble (2021.acl-long)
Copied to clipboard
| Challenge: | Recent studies show vulnerability of deep neural networks to adversarial examples that intentionally fool the networks. |
| Approach: | They propose a method for training a robust model to defense synonym substitution-based attacks by sampling embedding vectors for each word in an input sentence and augmenting them with the training data. |
| Outcome: | The proposed method outperforms other proposed defense methods by a significant margin across different network architectures and multiple data sets. |
AutoSUIT Bench - Automated Security UnIt Test Benchmark for LLM Coding (2026.findings-acl)
Copied to clipboard
Samuel Osebe, Fan Yang, Junyi Li, Yue Gu, Yongxin Wang, Satyapriya Krishna, Kai-Wei Chang, Aram Galstyan, Rahul Gupta, Weitong Ruan
| Challenge: | Large Language Models (LLMs) are evolving rapidly on code generation tasks. |
| Approach: | They propose to automate the vulnerability code benchmark creation with iterative auto validation. |
| Outcome: | The proposed benchmark covers 232 CWE categories across C/C++, Java, and Python languages. |
Measuring Fairness of Text Classifiers via Prediction Sensitivity (2022.acl-long)
Copied to clipboard
| Challenge: | Existing fairness metrics are not yet available to measure the fairness of language processing systems. |
| Approach: | They propose a new metric which measures fairness in machine learning models based on the model’s prediction sensitivity to perturbations in input features. |
| Outcome: | The proposed metric can be linked with a specific notion of group fairness and individual fairness, and correlates well with humans’ perception of fairness. |
TextEE: Benchmark, Reevaluation, Reflections, and Future Challenges in Event Extraction (2024.findings-acl)
Copied to clipboard
Kuan-Hao Huang, I-Hung Hsu, Tanmay Parekh, Zhiyu Xie, Zixuan Zhang, Prem Natarajan, Kai-Wei Chang, Nanyun Peng, Heng Ji
| Challenge: | Recent studies suggest that event extraction evaluations may not accurately reflect the true performance. |
| Approach: | They propose a standardized, fair, and reproducible benchmark for event extraction . they use standardized scripts and splits for 16 datasets spanning eight domains . |
| Outcome: | The proposed benchmarks show that they struggle to achieve satisfactory performance. |
A Survey of Deep Learning for Mathematical Reasoning (2023.acl-long)
Copied to clipboard
| Challenge: | a survey of deep learning for mathematical reasoning examines the field . a comprehensive reading list is provided to assist readers interested in the field. |
| Approach: | They present a survey of deep learning for mathematical reasoning over the past decade . they outline directions for future research and highlight potential for further exploration . |
| Outcome: | The proposed framework is based on the results of a decade-long survey of deep learning for mathematical reasoning. |
Multilingual Generative Language Models for Zero-Shot Cross-Lingual Event Argument Extraction (2022.acl-long)
Copied to clipboard
| Challenge: | Existing models for zero-shot cross-lingual event argument extraction are based on pre-trained generative language models. |
| Approach: | They propose to use pre-trained generative language models to generate sentences that fill in a template with arguments extracted from the input passage. |
| Outcome: | The proposed model outperforms the current state-of-the-art models on zero-shot cross-lingual EAE. |
A Corpus to Learn Refer-to-as Relations for Nominals (L18-1)
Copied to clipboard
| Challenge: | Existing work on how to learn refer-to-as relations from large unlabeled corpora lacks coreferential information. |
| Approach: | They propose to use Wikipedia to generate coreferential neural embeddings for nominals . they use coreference resolution as a proxy to evaluate the neural embeds for noun phrases . |
| Outcome: | The proposed dataset can be leveraged to construct representations for coreferential nominals from Wikipedia. |
“Nice Try, Kiddo”: Investigating Ad Hominems in Dialogue Responses (2021.naacl-main)
Copied to clipboard
| Challenge: | Ad hominem attacks target a person's character instead of the position the person is maintaining. |
| Approach: | They propose to use salient n-gram similarity as a soft constraint to reduce the amount of ad hominems generated in Twitter conversations. |
| Outcome: | The proposed method reduces the amount of ad hominems generated in human and dialogue system responses to English Twitter posts by using salient n-gram similarity as a soft constraint. |
MetaKP: On-Demand Keyphrase Generation (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing keyphrase prediction methods only output a single set of keyphrases per document . however, existing methods fail to cater to diverse needs of users and downstream applications . |
| Approach: | They propose a method that requires keyphrases that conform to specific high-level goals or intents to generate on-demand keyphrase generation. |
| Outcome: | The proposed method surpasses the performance of a fully fine-tuned BART-base model in 0.548 SemF1 . it can be used in epidemic event detection from social media. |
The Male CEO and the Female Assistant: Evaluation and Mitigation of Gender Biases in Text-To-Image Generation of Dual Subjects (2025.acl-long)
Copied to clipboard
| Challenge: | Recent large-scale T2I models like DALLE-3 have made progress in reducing gender stereotypes when generating single-person images. |
| Approach: | They propose a framework that queries T2I models to depict two individuals with gender-stereotyped social identities to evaluate gender biases. |
| Outcome: | The proposed framework reduces gender stereotypes when generating images with more than one person. |
LOGAN: Local Group Bias Detection by Clustering (2020.emnlp-main)
Copied to clipboard
| Challenge: | a number of machine learning models inherit and amplify the societal biases in data. |
| Approach: | a new bias detection technique based on clustering is proposed to detect local biases in data . authors propose to use LOGAN to analyze local bias in data. |
| Outcome: | The proposed technique detects bias in a local region and allows better analysis of model predictions. |
ARES: Adaptive Red-Teaming and End-to-End Repair of Policy-Reward System (2026.acl-long)
Copied to clipboard
Jiacheng Liang, Yao Ma, Tharindu Kumarage, Satyapriya Krishna, Rahul Gupta, Kai-Wei Chang, Aram Galstyan, Charith Peris
| Challenge: | Existing red-teaming approaches focus on policy-level weaknesses, but they overlook systemic weaknesses . aRES exploits dual-targeting weaknesses in both the core LLM and the RM simultaneously. |
| Approach: | a new framework uncovers weaknesses in both the core and the reward models simultaneously . a "Safety Mentor" generates semantically coherent adversarial prompts . |
| Outcome: | ARES uncovers weaknesses in both the core LLM and the RM simultaneously . it fine-tunes the LM to detect harmful content, then optimizes the core model . |
Con-ReCall: Detecting Pre-training Data in LLMs via Contrastive Decoding (2025.coling-main)
Copied to clipboard
| Challenge: | Existing methods analyze training data with member and non-member contexts, overlooking potential insights from both member and not-member. |
| Approach: | They propose a method that leverages asymmetric distributional shifts induced by member and non-member contexts through contrastive decoding to enhance membership inference. |
| Outcome: | The proposed approach outperforms the current state-of-the-art on the WikiMIA benchmark and is robust against various text manipulation techniques. |
Evaluating the Values of Sources in Transfer Learning (2021.naacl-main)
Copied to clipboard
| Challenge: | Transfer learning is a form of learning that adapts a model trained on data-rich sources to low-resource targets. |
| Approach: | They propose a source valuation framework that quantifies the usefulness of the sources in transfer learning by using the Shapley value method. |
| Outcome: | The proposed framework is effective in choosing useful transfer sources and the source values match the intuitive source-target similarity. |
Select, Extract and Generate: Neural Keyphrase Generation with Layer-wise Coverage Attention (2021.acl-long)
Copied to clipboard
| Challenge: | Generally, documents are truncated before being inputs to deep neural networks, resulting in missing keyphrases . evaluators use layer-wise coverage attention to cover all the critical points in a document . |
| Approach: | They propose a neural keyphrase generation model that identifies the salient sentences in a document and an extractor-generator that jointly extracts and generates keyphrases from the selected sentences. |
| Outcome: | The proposed model outperforms the state-of-the-art keyphrase generation methods on keyphrases generated from scientific and web documents. |
Medical Vision-Language Pre-Training for Brain Abnormalities (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing vision-language models lack expertise for medical applications due to the scarcity and complexity of data. |
| Approach: | They propose a pipeline to collect medical image-text aligned data for pretraining from public resources such as PubMed and build a high-performance vision-language model tailored to specific medical tasks. |
| Outcome: | The proposed model is based on a large brain image-text dataset and will be released to the public. |
How well can Text-to-Image Generative Models understand Ethical Natural Language Interventions? (2022.emnlp-main)
Copied to clipboard
| Challenge: | Text-to-image generative models can generate high-quality photo-realistic images conditional on natural language text descriptions in a zero-shot fashion. |
| Approach: | They propose an Ethical NaTural Language Interventions in Text-to-Image GENeration benchmark dataset to evaluate the change in image generation conditional on ethical interventions across three social axes – gender, skin color, and culture. |
| Outcome: | The proposed model generations cover diverse social groups while preserving image quality. |
Symbolic Chain-of-Thought Distillation: Small Models Can Also “Think” Step-by-Step (2023.acl-long)
Copied to clipboard
| Challenge: | Symbolic Chain-of-thought Distillation (SCoTD) is a method to train a smaller student model on rationalizations sampled from a significantly larger teacher model. |
| Approach: | They propose a method to train a smaller student model on rationalizations from a larger teacher model. |
| Outcome: | The proposed method improves the performance of a student model in supervised and few-shot settings and especially for challenge sets. |
Examining Gender Bias in Languages with Grammatical Gender (D19-1)
Copied to clipboard
| Challenge: | Existing studies on gender bias in word embeddings focus on English . however, these studies cannot be extended to languages with morphological agreement on gender . |
| Approach: | They propose new metrics to evaluate gender bias in word embeddings of English and Spanish . they extend existing approaches to mitigate gender bias while preserving original embeddables . |
| Outcome: | The proposed methods reduce gender bias while preserving the original embeddings. |
A Corpus of Drug Usage Guidelines Annotated with Type of Advice (L18-1)
Copied to clipboard
| Challenge: | Current research indicates patients are often unaware of such critical information / advice related to their prescription drugs due to lack of communication with their doctors and/or pharmacists. |
| Approach: | They propose an annotation scheme for annotating safety critical advice from drug usage guidelines and an annotated dataset containing drug usage guideline data. |
| Outcome: | The proposed dataset will accelerate further release of annotated drug usage guideline datasets and research on automatically filtering safety critical information from these documents. |
Model Editing Harms General Abilities of Large Language Models: Regularization to the Rescue (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods that edit large language models with updated knowledge can cause side effects on the general abilities of LLMs such as reasoning, natural language inference, and question answering. |
| Approach: | They propose to regularize the edit update weights by imposing constraints on their complexity based on the RElative Change in weighT. |
| Outcome: | The proposed method can significantly mitigate the side effects while maintaining over 94% editing performance. |
Improving the Adversarial Robustness of NLP Models by Information Bottleneck (2022.findings-acl)
Copied to clipboard
| Challenge: | Existing studies have shown that adversarial examples can be directly attributed to the presence of non-robust features. |
| Approach: | They propose to capture task-specific robust features while eliminating non-robust ones . they show that models can achieve significant improvement in robust accuracy . |
| Outcome: | The proposed method outperforms all defense methods on SST-2, AGNEWS and IMDB datasets while achieving no performance drop. |
Resolving Ambiguities in Text-to-Image Generative Models (2023.acl-long)
Copied to clipboard
Ninareh Mehrabi, Palash Goyal, Apurv Verma, Jwala Dhamala, Varun Kumar, Qian Hu, Kai-Wei Chang, Richard Zemel, Aram Galstyan, Rahul Gupta
| Challenge: | ambiguities can lead to misinterpretation and miscommunication in natural language . resolving ambiguity is notoriously hard for machines . |
| Approach: | They propose a framework to disambiguate prompts given to generative models by soliciting clarifications from the end user. |
| Outcome: | The proposed framework generates more faithful images better aligned with user intention in the presence of ambiguities. |
The Tail Wagging the Dog: Dataset Construction Biases of Social Bias Benchmarks (2023.acl-short)
Copied to clipboard
| Challenge: | omnipresence of large pre-trained language models has fueled concerns regarding systematic biases carried over from underlying data into the applications they are used in. |
| Approach: | They propose to compare social biases with non-social biase masked by alternate constructions that maintain the essence of their social bias. |
| Outcome: | The proposed benchmarks underestimate or overestimate the social bias in a given model. |
Retrofitting Contextualized Word Embeddings with Paraphrases (D19-1)
Copied to clipboard
| Challenge: | Contextualized word embeddings can be useful for downstream applications, but they can be over-sensitive to contexts. |
| Approach: | They propose a method to retrofit contextualized word embeddings with paraphrases to minimize the variance of word representations on paraphrased contexts. |
| Outcome: | The proposed method improves on sentence classification and inference tasks. |
On Difficulties of Cross-Lingual Transfer with Order Differences: A Case Study on Dependency Parsing (N19-1)
Copied to clipboard
| Challenge: | Existing studies on crosslingual transfer have focused on word-level information sharing, but words are not independent in sentences; their combinations form larger linguistic units, known as context. |
| Approach: | They propose to use orderagnostic models to transfer word order to distant languages . they train dependency parsers on an English corpus and evaluate their transfer performance on 30 other languages. |
| Outcome: | The proposed model performs better on languages with different word orders than on other languages. |
Towards Safety Reasoning in LLMs: AI-agentic Deliberation for Policy-embedded CoT Data Creation (2025.findings-acl)
Copied to clipboard
Tharindu Kumarage, Ninareh Mehrabi, Anil Ramakrishna, Xinyan Zhao, Richard Zemel, Kai-Wei Chang, Aram Galstyan, Rahul Gupta, Charith Peris
| Challenge: | Safety reasoning paradigms require high-quality policy-embedded chain-of-thought datasets . generating such data through human annotations is prohibitively expensive . |
| Approach: | They propose AIDSAFE: Agentic Iterative Deliberation for Safety Reasoning . AIDS AFE leverages multi-agent deliberation to iteratively expand reasoning on safety policies . |
| Outcome: | The proposed model improves policy adherence and reasoning quality while maintaining acceptable utility and over-refusal accuracy. |
Robust Text Classifier on Test-Time Budgets (D19-1)
Copied to clipboard
| Challenge: | Recent advances in deep neural networks (DNNs) achieve high accuracy on many text classification tasks. |
| Approach: | They propose a generic framework for learning a robust text classification model . they use a data aggregation method to train the classifier on a large corpus of text . |
| Outcome: | The proposed framework achieves consistent speedup with little degradation in accuracy on four benchmark text classification tasks. |
Open-Domain Safety Policy Construction (2026.findings-eacl)
Copied to clipboard
| Challenge: | Moderation layers are core component of many products built on user-generated content. |
| Approach: | They propose a system that drafts a content moderation policy based on human-written seed domain information. |
| Outcome: | The proposed system outperforms definition-only and in-context learning baselines on openAI undesired content benchmarks and an in-house multimodal advertisement moderation benchmark. |
Towards Controllable Biases in Language Generation (2020.findings-emnlp)
Copied to clipboard
| Challenge: | a new method to induce societal biases in natural language generation is being developed . a method to equalize the amount of biased text across demographics is effective . |
| Approach: | They propose a method to induce societal biases in natural language generation by using demographic inequalities. |
| Outcome: | The proposed method is effective at equalizing biases across demographics while generating less negatively biased text overall. |
BLUR: A Bi-Level Optimization Approach for LLM Unlearning (2026.eacl-long)
Copied to clipboard
Hadi Reisizadeh, Jinghan Jia, Zhiqi Bu, Bhanukiran Vinzamuri, Anil Ramakrishna, Kai-Wei Chang, Volkan Cevher, Sijia Liu, Mingyi Hong
| Challenge: | Existing algorithms to unlearn knowledge and capabilities from large datasets are unclear how to best formulate the unlearning problem. |
| Approach: | They propose to model the hierarchical structure of the unlearning problem, where the forget problem takes priority over the retain problem, and propose an algorithm that aims to unlearn knowledge and capabilities. |
| Outcome: | The proposed algorithm outperforms all state-of-the-art algorithms across unlearning tasks, models, and metrics. |
Unsupervised Syntactically Controlled Paraphrase Generation with Abstract Meaning Representations (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to syntactically controlled paraphrase generation require annotated paraphrase pairs for training and are costly to extend to new domains. |
| Approach: | They propose to leverage Abstract Meaning Representations (AMR) to improve the performance of unsupervised syntactically controlled paraphrase generation. |
| Outcome: | The proposed model generates more accurate syntactically controlled paraphrases, both quantitatively and qualitatively, compared to the existing unsupervised approaches. |
Comparing Bad Apples to Good Oranges Aligning Large Language Models via Joint Preference Optimization (2025.findings-acl)
Copied to clipboard
| Challenge: | Recent studies have shown that acquiring human preferences by comparing generations is not effective for large language models. |
| Approach: | They propose a preference optimization objective that elicits preferences jointly over the instruction-response pairs. |
| Outcome: | The proposed approach outperforms prior preference optimizations by 5.2% and 3.3% in summarization and open-ended dialogue datasets. |
Gold-Medal-Level Olympiad Geometry Solving with Efficient Heuristic Auxiliary Constructions (2026.acl-long)
Copied to clipboard
Boyan Duan, Xiao Liang, Shuai Lu, Yaoxiang Wang, Yelong Shen, Kai-Wei Chang, Ying Nian Wu, Mao Yang, Weizhu Chen, Yeyun Gong
| Challenge: | Existing methods for geometry theorem proving in Euclidean geometry are challenging and require a neural network to perform. |
| Approach: | They propose a method for adding auxiliary points in geometry that runs on CPUs without relying on neural network-based inference. |
| Outcome: | The proposed method achieves silver-medal-level human performance on IMO-30 benchmark. |
Robustness and Adversarial Examples in Natural Language Processing (2021.emnlp-tutorials)
Copied to clipboard
| Challenge: | This tutorial aims to raise awareness of practical concerns about NLP robustness . it aims at addressing the weaknesses of NLP systems when faced with adversarial inputs and data with a distribution shift . |
| Approach: | This tutorial aims to bring awareness of practical concerns about NLP robustness . it reviews recent studies on analyzing the weakness of NLP systems when facing adversarial inputs . |
| Outcome: | This tutorial aims to bring awareness of practical concerns about NLP robustness . it will examine the weaknesses of NLP systems when faced with adversarial inputs and data with a distribution shift . |
UniFine: A Unified and Fine-grained Approach for Zero-shot Vision-Language Understanding (2023.findings-acl)
Copied to clipboard
| Challenge: | supervised methods for vision-language tasks have been well-studied, but they lack the fine-grained information needed for semantics understanding. |
| Approach: | They propose a framework to take advantage of fine-grained information for zero-shot vision-language learning, covering multiple tasks such as VQA, SNLI-VE, and VCR. |
| Outcome: | The proposed framework outperforms previous zero-shot methods on VQA and achieves substantial improvement on SNLI-VE and VCR. |
METAL: A Multi-Agent Framework for Chart Generation with Test-Time Scaling (2025.acl-long)
Copied to clipboard
| Challenge: | Chart generation requires strong visual design skills and precise coding capabilities that embed the desired visual properties into code. |
| Approach: | They propose a vision-language model-based multi-agent framework for effective automatic chart generation. |
| Outcome: | The proposed framework achieves a 5.2% improvement in the F1 score over the current best chart generation task. |
TAGPRIME: A Unified Framework for Relational Structure Extraction (2023.acl-long)
Copied to clipboard
| Challenge: | Existing models for natural language processing (NLP) do not address common tasks. |
| Approach: | They propose to take a unified view of all the tasks and introduce a model that appends priming words about the condition to the input text. |
| Outcome: | The proposed model is based on ten datasets across five different languages and covers ten tasks that cover ten languages. |
Understanding ME? Multimodal Evaluation for Fine-grained Visual Commonsense (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing models that understand image and text but also cross-reference in-between are lacking in evaluation data resources. |
| Approach: | They propose a multimodal evaluation pipeline to automatically generate question-answer pairs to test models’ understanding of the visual scene, text, and related knowledge. |
| Outcome: | The proposed model can answer the highly semantic VCR question correctly but fails to answer related visual question (Q2), textual question (q3), and background knowledge question ( Q4) as shallow mappings with language priors and unbalanced utilization of information between modalities. |
Are Personalized Stochastic Parrots More Dangerous? Evaluating Persona Biases in Dialogue Systems (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Recent advances in Large Language Models enable them to follow freeform instructions, including imitating generic or specific demographic personas in conversations. |
| Approach: | They propose to investigate persona biases by experimenting with UNIVERSALPERSONA, a model that incorporates both generic and specific personas. |
| Outcome: | The proposed model systematically measures persona biases in harmful expression and harmful agreement. |
Building Language Models for Text with Named Entities (P18-1)
Copied to clipboard
| Challenge: | Existing language models fail to predict the entity names due to their wide variations. |
| Approach: | They propose a language model which can learn the entity names by leveraging their entity type information. |
| Outcome: | The proposed model achieves 52.2% better perplexity in recipe generation and 22.06% on code generation than state-of-the-art language models. |
DRS: Deep Question Reformulation With Structured Output (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing models like GPT-3 and Instruct-GPT lack the ability to reformulate unanswerable questions. |
| Approach: | They propose a zero-shot method that combines the strengths of LLMs with a DFS-based algorithm to iteratively explore potential entity combinations and constrain outputs using predefined entities. |
| Outcome: | The proposed method outperforms all baselines, including the GPT-3.5 model, on the unanswerable question reformulation task. |
A Transformer-based Approach for Source Code Summarization (2020.acl-main)
Copied to clipboard
| Challenge: | Generating a readable summary that describes the functionality of a program is known as source code summarization. |
| Approach: | They propose a Transformer model that uses a self-attention mechanism to capture long-range dependencies by encoding source code tokens relative to the code token position. |
| Outcome: | The proposed model outperforms the state-of-the-art methods by a significant margin. |
Enhancing Unsupervised Semantic Parsing with Distributed Contextual Representations (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to learn models on corpus of pairs of sentences require labor-intensive annotation. |
| Approach: | They propose to leverage distributed contextual word and phrase representations pre-trained on unlabelled texts to deal with homonymy and polysemy. |
| Outcome: | The proposed model achieves better accuracy on question-answering and relation extraction tasks. |
Dataset Bias Mitigation in Multiple-Choice Visual Question Answering and Beyond (2023.findings-emnlp)
Copied to clipboard
Zhecan Wang, Long Chen, Haoxuan You, Keyang Xu, Yicheng He, Wenhao Li, Noel Codella, Kai-Wei Chang, Shih-Fu Chang
| Challenge: | Existing studies have examined dataset biases in VQA benchmarks with short-phrase answers Multiple-choice Question with the LONG Answers (VCR, VLEP, etc.) |
| Approach: | They propose to use Adversarial Data Synthesis (ADS) to generate synthetic training and debiased evaluation data and introduce Intra-sample Counterfactual Training (ICT) to assist models in utilizing synthesized training data. |
| Outcome: | The proposed approach improves model performance even in domain-shifted scenarios. |
Ethical-Advice Taker: Do Language Models Understand Natural Language Interventions? (2021.findings-acl)
Copied to clipboard
| Challenge: | a new task is proposed to modify a model's unethical behavior by communicating context-specific principles of ethics and equity to it. |
| Approach: | They propose a task to amend a questionanswering model's unethical behavior . they use social stereotypes to augment social stereotype models with ethical interventions . |
| Outcome: | The proposed task improves model behavior but fails to generalize to new domains . it is based on a zero-shot evaluation of social stereotypes in reading comprehension systems . |
“The Boating Store Had Its Best Sail Ever”: Pronunciation-attentive Contextualized Pun Recognition (2020.acl-main)
Copied to clipboard
| Challenge: | Identifying and modeling puns is challenging as they involve implicit semantic or phonological tricks. |
| Approach: | They propose a method to detect puns in a sentence and then locate them in it . they propose to capture phonetic associations between the context and phonetic symbols . |
| Outcome: | The proposed method outperforms state-of-the-art methods in pun detection and location tasks. |
Generating Natural Language Adversarial Examples (D18-1)
Copied to clipboard
| Challenge: | Recent research has shown that deep neural networks are vulnerable to adversarial examples, perturbations to correctly classified examples which can cause the model to misclassify. |
| Approach: | They propose to generate adversarial examples that fool well-trained sentiment analysis and textual entailment models by using a black-box population-based optimization algorithm. |
| Outcome: | The proposed model is able to fool well-trained sentiment analysis and textual entailment models with success rates of 97% and 70%, respectively. |
VisRet: Visualization Improves Knowledge-Intensive Text-to-Image Retrieval (2026.acl-long)
Copied to clipboard
| Challenge: | Text-to-image retrieval is challenging because of cross-modal embeddings are bags of concepts, underrepresenting structured visual relationships. |
| Approach: | They propose a retrieval paradigm that embeds textual queries into the image modality via T2I generation and performs retrieval within the image mode to bypass weaknesses of cross-modal retrievers in recognizing subtle visual-spatial features. |
| Outcome: | The proposed retrieval paradigm outperforms previous approaches in visual-spatial retrieval benchmarks. |
Event Detection from Social Media for Epidemic Prediction (2024.naacl-long)
Copied to clipboard
Tanmay Parekh, Anh Mac, Jiarui Yu, Yuxuan Dong, Syed Shahriar, Bonnie Liu, Eric Yang, Kuan-Hao Huang, Wei Wang, Nanyun Peng, Kai-Wei Chang
| Challenge: | Social media is an easy-to-access platform providing timely updates about societal trends and events. |
| Approach: | They propose a framework to extract epidemic-related events from social media posts to provide early warnings. |
| Outcome: | The proposed framework can detect epidemic events for three unseen epidemics of Monkeypox, Zika, and Dengue while existing models fail miserably. |
Retrieval Augmented Code Generation and Summarization (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Software developers often recall parts of source code or code summaries that they had written in the past while implementing software or documenting them. |
| Approach: | They propose a retrieval augmented framework that retrieves relevant code or summaries from a database and provides them as a supplement to code generation or summarization models. |
| Outcome: | The proposed framework can search for relevant code or summaries from retrieval databases and can work with unimodal (only code or natural language description) or bimodal instances (code-description pairs). |
SWAN: Semantic Watermarking with Abstract Meaning Representation (2026.acl-long)
Copied to clipboard
Ziping Ye, Gourab Dey, Christos Christodoulopoulos, Charith Peris, Anil Ramakrishna, Weitong Ruan, Aram Galstyan, Kai-Wei Chang, Rahul Gupta, Ninareh Mehrabi
| Challenge: | Existing methods to embed signatures by adjusting token selection preferences during text generation are highly sensitive to paraphrasing and synonyms. |
| Approach: | They propose a framework that embeds watermark signatures into the semantic structure of a sentence using Abstract Meaning Representation (AMR). |
| Outcome: | Empirical evaluation shows SWAN matches state-of-the-art detection performance on unaltered watermarked text while improving robustness against paraphrasing. |
Socially Aware Bias Measurements for Hindi Language Representations (2022.naacl-main)
Copied to clipboard
| Challenge: | Language representations are an efficient tool used across NLP, but they are strife with encoded societal biases. |
| Approach: | They investigate the encoded biases in Hindi language representations based on cultural and historical contexts . they emphasize the necessity of social-awareness along with linguistic and grammatical artefacts when modeling language representation . |
| Outcome: | The proposed model reflects the cultural and cultural diversity of the region in which it is used . the model is based on the language and culture of the language being used based upon the study . |
Synchronous Faithfulness Monitoring for Trustworthy Retrieval-Augmented Generation (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies show that RALMs generate baseless information or contradicts with the retrieved context. |
| Approach: | They propose a lightweight monitor that leverages fine-grained decoding dynamics to synchronously detect unfaithful sentences. |
| Outcome: | Empirical results show that SynCheck outperforms baseline faithfulness detection and FOD outperformed traditional strategies in terms of faithfulness. |
The Woman Worked as a Babysitter: On Biases in Language Generation (D19-1)
Copied to clipboard
| Challenge: | a systematic study of biases in natural language generation (NLG) is presented . a study of language models in NLG is conducted by examining language models. |
| Approach: | They propose a systematic study of biases in natural language generation by analyzing text generated from prompts that contain mentions of different demographic groups. |
| Outcome: | The proposed method reveals biases in natural language generation (NLG) by analyzing text generated from demographic prompts. |
Knowledge Control for Responsible Generative AI: Bridging Academia, Industry, and Society (2026.acl-tutorials)
Copied to clipboard
Zheyuan Liu, Yixin Wan, Kai-Wei Chang, Meng Jiang, Jieyu Zhao, Nouha Dziri, Yuning Mao, Jia-Chen Gu, Jindong Gu
| Challenge: | This tutorial introduces the foundations of post-training knowledge control and showcases recent frontier methods. |
| Approach: | This tutorial introduces the foundations of post-training knowledge control and showcases recent frontier methods. |
| Outcome: | This tutorial introduces the foundations of post-training knowledge control and showcases recent frontier methods . key motivations and failure modes, harmful generation and stereotype reinforcement, are addressed . core methods such as machine unlearning, knowledge editing, and inference-time interventions are also included . |
Disentangling Semantics and Syntax in Sentence Embeddings with Pre-trained Language Models (2021.naacl-main)
Copied to clipboard
| Challenge: | Pre-trained language models have been successful on a wide range of NLP tasks . however, contextual representations from pre-trated models contain entangled semantic and syntactic information. |
| Approach: | They propose a semantic sentence embedding model that disentangles semantics and syntax from pre-trained models. |
| Outcome: | The proposed model outperforms state-of-the-art models on unsupervised semantic similarity tasks. |
Generating Sports News from Live Commentary: A Chinese Dataset for Sports Game Summarization (2020.aacl-main)
Copied to clipboard
| Challenge: | Existing methods to generate sports summarization tasks are laborintensive and infeasible. |
| Approach: | They propose a Chinese dataset for sports game summarization and a model that consists of a selector and rewriter to evaluate the correctness of generated sports summaries. |
| Outcome: | The proposed model performs better on ROUGE and the two designed scores. |
BILLY: Steering Large Language Models via Merging Persona Vectors for Creative Generation (2026.eacl-long)
Copied to clipboard
| Challenge: | Multi-LLM systems enhance creativity of large language models by simulating human collective intelligence but suffer from significant drawbacks, such as high computational costs and inference latency. |
| Approach: | They propose a training-free framework that captures the benefits of multi-LLM collaboration by extracting and blending multiple distinct persona vectors directly in the model’s activation space. |
| Outcome: | The proposed framework surpasses model prompting and traditional multi-LLM approaches while significantly reducing inference time and computational costs. |
Towards Adversarially Robust Text Classifiers by Learning to Reweight Clean Examples (2022.findings-acl)
Copied to clipboard
| Challenge: | Existing defense methods improve the adversarial robustness by making models adapt to training set augmented with some adversarials. |
| Approach: | They propose to introduce a reweighting mechanism to calibrate the training distribution to obtain robust models. |
| Outcome: | The proposed method minimizes the loss of validation set mixed with clean examples and adversarial ones in an online learning manner. |
Few-Shot Representation Learning for Out-Of-Vocabulary Words (P19-1)
Copied to clipboard
| Challenge: | Existing methods for learning word embedding assume there are enough occurrences for each word in the corpus to accurately estimate the representation of words. |
| Approach: | They propose to fit a representation function to predict an oracle embedding vector based on limited contexts. |
| Outcome: | The proposed model outperforms existing methods in constructing an accurate embedding for OOV words and improves downstream tasks when the embeddable is utilized. |
Tree-of-Traversals: A Zero-Shot Reasoning Algorithm for Augmenting Black-box Language Models with Knowledge Graphs (2024.acl-long)
Copied to clipboard
Elan Markowitz, Anil Ramakrishna, Jwala Dhamala, Ninareh Mehrabi, Charith Peris, Rahul Gupta, Kai-Wei Chang, Aram Galstyan
| Challenge: | Knowledge graphs (KGs) complement Large Language Models (LLMs) by providing reliable, structured, domain-specific, and up-to-date external knowledge. |
| Approach: | They propose a zero-shot reasoning algorithm that augments black-box LLMs with one or more KGs. |
| Outcome: | The proposed algorithm significantly improves performance on question answering and KG question answering tasks. |
On the Sensitivity and Stability of Model Interpretations in NLP (2022.acl-long)
Copied to clipboard
| Challenge: | Recent years have witnessed the emergence of post-hoc interpretations that aim to uncover how NLP models make predictions. |
| Approach: | They propose two new criteria that provide complementary notions of faithfulness to removal-based criteria. |
| Outcome: | The proposed methods overcome limitations of gradient-based methods on removal-based criteria and overcome limitations in the proposed methods. |
Vulnerability of Large Language Models to Output Prefix Jailbreaks: Impact of Positions on Safety (2025.findings-naacl)
Copied to clipboard
| Challenge: | Previous research on jailbreak attacks has focused on optimizing the adversarial snippet content injected into input prompts to expose LLM security vulnerabilities. |
| Approach: | They propose to use a simple adversarial snippet at the beginning of output to expose LLM security vulnerabilities. |
| Outcome: | The proposed approach exposes LLM security vulnerabilities much faster than input suffix attacks or prompt-based output jailbreaks. |
Are LLMs Capable of Data-based Statistical and Causal Reasoning? Benchmarking Advanced Quantitative Reasoning with Data (2024.findings-acl)
Copied to clipboard
| Challenge: | Quantitative reasoning with data is a critical skill to analyze data, yet the assessment of such ability remains limited. |
| Approach: | They propose a quantitative reasoning with data benchmark to evaluate Large Language Models' ability in statistical and causal reasoning with real-world data. |
| Outcome: | The proposed model GPT-4 achieves an accuracy of 58%, while open-source model Deepseek-coder-instruct gets the highest accuracy of 37%. |
Indirectly Supervised Natural Language Processing (2023.acl-tutorials)
Copied to clipboard
| Challenge: | a tutorial on indirect supervision addresses challenges in ML for NLP . conventional approaches to NLP use taskspecific labeled examples of a large volume . indirect supervision is useful for a wide range of NLP tasks, but it is not enough for decoders . |
| Approach: | This tutorial aims to address questions about indirect supervision in machine learning . authors discuss indirect supervision from T′ that handles T with outputs spanning from a moderate size to an open space . |
| Outcome: | This tutorial aims to answer questions about how to provide supervision for ML tasks . it will discuss indirect supervision from T′ that handles T with outputs spanning from a moderate size to an open space . |
What Does BERT with Vision Look At? (2020.acl-main)
Copied to clipboard
| Challenge: | Pre-trained visual grounded language models have improved performance on vision-and-language tasks but what they learn during pre-training remains unclear. |
| Approach: | They show that attention heads of visual grounded language models actively ground elements of language to image regions. |
| Outcome: | The attention heads of a visual grounded language model can ground elements to image regions, demonstrating their ability to detect syntactic relations between non-entity words and image regions. |
White Men Lead, Black Women Help? Benchmarking and Mitigating Language Agency Social Biases in LLMs (2025.acl-long)
Copied to clipboard
| Challenge: | Social biases manifest in language agency, but there is no comprehensive benchmark for evaluating such biase in language models. |
| Approach: | They propose a benchmark to evaluate language agency biases in large language models . they propose 'Mitigation via Selective Rewrite' to selectively revise parts of generated texts . |
| Outcome: | The proposed language agency bias evaluation benchmark identifies gender, racial, and intersectional biases in 3 recent LLMs. |
Summarize and Generate to Back-translate: Unsupervised Translation of Programming Languages (2023.eacl-main)
Copied to clipboard
| Challenge: | Recent developments of multilingual pre-trained sequence-to-sequence models for programming languages have been effective for a broad spectrum of downstream software engineering tasks. |
| Approach: | They propose to combine a source-to-target model with a target-tosource model trained in parallel. |
| Outcome: | The proposed approach performs competitively with state-of-the-art methods. |
Societal Biases in Language Generation: Progress and Challenges (2021.acl-long)
Copied to clipboard
| Challenge: | Language generation techniques can produce undesirable societal biases that can negatively impact marginalized populations. |
| Approach: | They propose to examine how decoding techniques contribute to biases in language generation . they also conduct experiments to quantify the effects of these techniques . |
| Outcome: | The proposed methods can reduce biases and improve user experience, the authors argue . they also show that the proposed techniques can reduce societal biase . |