Papers by Ke Wang
Harvesting and Refining Question-Answer Pairs for Unsupervised QA (2020.acl-main)
Copied to clipboard
| Challenge: | Recent research attempts to extend unsupervised question answering to settings with few or no labeled data available. |
| Approach: | They propose two approaches to improve unsupervised question answering . first, they harvest lexically and syntactically divergent Wikipedia questions to automatically construct a corpus of question-answer pairs . second, they take advantage of the QA model to extract more appropriate answers . |
| Outcome: | The proposed approach outperforms previous unsupervised approaches by a large margin and is competitive with early supervised models. |
Learning Task Decomposition to Assist Humans in Competitive Programming (2024.acl-long)
Copied to clipboard
| Challenge: | Using language models (LMs) to solve complex problems, humans might struggle to understand and repair flawed ones. |
| Approach: | They propose to automatically decompose complex problems into simpler pieces that correspond to specific subtasks and measure their assistive value. |
| Outcome: | The proposed method enables non-experts to solve 33.3% more problems and speeds them up by 3.3x . |
M3Seg: A Maximum-Minimum Mutual Information Paradigm for Unsupervised Topic Segmentation in ASR Transcripts (2023.emnlp-main)
Copied to clipboard
| Challenge: | Topic segmentation aims to split automatic speech recognition transcriptions into segments that are bounded by thematic meanings. |
| Approach: | They propose a Maximum-Minimum Mutual information paradigm for linear topic segmentation without using any parallel data. |
| Outcome: | The proposed paradigm outperforms the state-of-the-art methods by a significant margin. |
MathGenie: Generating Synthetic Data with Question Back-translation for Enhancing Mathematical Reasoning of LLMs (2024.acl-long)
Copied to clipboard
| Challenge: | Existing models have demonstrated outstanding capabilities in mathematical reasoning, but there is a performance gap between open-source models and closed-source ones. |
| Approach: | They propose a method for generating diverse and reliable math problems by leveraging the ground-truth solutions of the seed data. |
| Outcome: | The proposed model outperforms open-source models across five representative mathematical reasoning datasets. |
Sparse-RL: Breaking the Memory Wall in LLM Reinforcement Learning via Stable Sparse Rollouts (2026.acl-long)
Copied to clipboard
Sijia Luo, Xiaokang Zhang, Yuxuan Hu, Bohan Zhang, Ke Wang, Jinbo Su, Mengshu Sun, Lei Liang, Jing Zhang
| Challenge: | Existing methods for storing key-value caches during long-horizon rollouts cause performance collapses. |
| Approach: | They propose a new training paradigm that empowers stable RL training under sparse rollouts. |
| Outcome: | The proposed model reduces rollout overhead while maintaining the performance. |
From Outcome to Process: Optimizing MoE Load Balancing with MCTS (2026.findings-acl)
Copied to clipboard
Wenjun Ke, Hengyuan Xu, Ziyu Shang, Yao He, Jiahao Wang, Zijie Xu, Peng Wang, Yuhang Lou, Jiajun Liu
| Challenge: | Existing balancing strategies focus on constraining the final distribution of expert usage, but overlook the routing decisions made at each layer. |
| Approach: | They propose a three-stage framework that leverages process-level rewards to guide balanced expert routing. |
| Outcome: | Extensive experiments show that LayerMoE improves the performance of state-of-the-art LoRA-MoA baselines, yielding an average accuracy gain of 1.39%. |
SkipBERT: Efficient Inference with Shallow Layer Skipping (2022.acl-long)
Copied to clipboard
| Challenge: | Pre-trained language models have significant demands in computation and inference time, limiting their use in resource-constrained or latencysensitive applications. |
| Approach: | They propose to encode text chunks into independent representations and skip computation of shallow layers to accelerate inference. |
| Outcome: | The proposed approach can reduce latency by 65% without sacrificing performance. |
CAT: Confidence-Adaptive Thinking for Efficient Reasoning of Large Reasoning Models (2026.acl-industry)
Copied to clipboard
| Challenge: | Existing compression methods for large reasoning models rely on uniform length reduction or coarse-grained difficulty estimation, often leading to performance degradation on difficult problems. |
| Approach: | They propose a framework that incorporates model’s intrinsic self-certainty signals as confidence into the preference optimization process, which autonomously modulates reasoning lengths based on problem difficulty. |
| Outcome: | The proposed framework outperforms state-of-the-art models on reasoning accuracy across multiple benchmarks on different base models. |
Dual-Stage Multi-Task Syntax-Oriented Pre-Training for Syntactically Controlled Paraphrase Generation (2024.findings-acl)
Copied to clipboard
| Challenge: | Syntactically controlled paraphrase generation (SCPG) aims to generate sentences with syntactic structures resembling given exemplars. |
| Approach: | They propose a dual-stage multi-task pre-training scheme that uses a series of structure-oriented and syntax-oriented tasks to generate sentences with syntactic structures resembling given exemplars. |
| Outcome: | The proposed method outperforms existing methods on all possible variants of SCPG tasks and significantly outperformed the popular T5 model. |
PopAlign: Diversifying Contrasting Patterns for a More Comprehensive Alignment (2025.acl-long)
Copied to clipboard
Zekun Moore Wang, Shenzhi Wang, King Zhu, Jiaheng Liu, Ke Xu, Jie Fu, Wangchunshu Zhou, Wenhao Huang
| Challenge: | Typical approaches to training large language models rely on limited contrasting patterns . contrasting data is limited and models are susceptible to harmful response tendencies . |
| Approach: | They propose a framework that integrates contrasting patterns across the prompt, model, and pipeline levels. |
| Outcome: | The proposed framework outperforms existing methods in the comparison of RQ1 and RQ2 . the proposed framework significantly outperformed existing methods, leading to more comprehensive alignment. |
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. |
AgentBank: Towards Generalized LLM Agents via Fine-Tuning on 50000+ Interaction Trajectories (2024.findings-emnlp)
Copied to clipboard
Yifan Song, Weimin Xiong, Xiutian Zhao, Dawei Zhu, Wenhao Wu, Ke Wang, Cheng Li, Wei Peng, Sujian Li
| Challenge: | Existing studies focus on specialized agents designed for particular tasks. |
| Approach: | They propose to scale annotated interaction trajectories and fine-tune LLMs on AgentBank to get a series of agent models, Samoyed. |
| Outcome: | The proposed model can scale to get generalized agent capabilities. |
Draft
& Verify: Lossless Large Language Model Acceleration via Self-Speculative Decoding (2024.acl-long)
Copied to clipboard
| Challenge: | Existing methods for accelerating Large Language Models have been criticized for their inference costs and inefficient decoding. |
| Approach: | They propose a self-speculative decoding approach for accelerating Large Language Models without an auxiliary model. |
| Outcome: | The proposed method achieves a speedup of up to 1.99 with no additional neural network training and no extra memory footprint. |
Towards Medical Complex Reasoning with LLMs through Medical Verifiable Problems (2025.findings-acl)
Copied to clipboard
| Challenge: | OpenAI o1 has been a significant milestone in large language model development . however, most research in reasoning has focused on mathematical tasks . medical domains require robust reasoning to provide reliable answers . |
| Approach: | They propose a method to verify medical reasoning using a medical verifier . they also propose RL and reinforcement learning to enhance reasoning . |
| Outcome: | The proposed method outperforms general and medical-specific baselines using only 40K verifiable problems. |
On the Consistency of Commonsense in Large Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing evaluations of commonsense for large language models focus on downstream knowledge tasks, failing to probe whether LLMs truly understand and utilize knowledge or merely memorize it. |
| Approach: | They propose to automatically construct a large benchmark named CoCo which measures LLMs’ knowledge memorization, comprehension, and application and examines the consistency between these tasks. |
| Outcome: | The proposed benchmark systematically assesses LLMs’ knowledge memorization, comprehension, and application and examines the consistency between these tasks. |
Acquisition and Application of Novel Knowledge in Large Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for constructing new datasets rely on timestamps or simple templates that do not accurately reflect the real world. |
| Approach: | They propose a knowledge dataset construction approach that simulates biological evolution using knowledge graphs to generate synthetic entities with diverse attributes. |
| Outcome: | The proposed framework outperforms knowledge augmentation methods by 3.3%-38%. |
PositionID: LLMs can Control Lengths, Copy and Paste with Explicit Positional Awareness (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have impressive capabilities across various domains, including role-playing, creative writing, mathematical reasoning, and coding. |
| Approach: | They propose two methods to improve the model’s adherence to length constraints and copy-paste accuracy without compromising response quality. |
| Outcome: | The proposed methods improve the model’s adherence to length constraints and copy-paste accuracy without compromising response quality. |
MathCoder-VL: Bridging Vision and Code for Enhanced Multimodal Mathematical Reasoning (2025.findings-acl)
Copied to clipboard
Ke Wang, Junting Pan, Linda Wei, Aojun Zhou, Weikang Shi, Zimu Lu, Han Xiao, Yunqiao Yang, Houxing Ren, Mingjie Zhan, Hongsheng Li
| Challenge: | Large Language Models (LMMs) struggle with simple tasks such as geometry, e.g., arithmetic, and reasoning. |
| Approach: | They propose to leverage code as supervision for cross-modal alignment . they propose to use FigCodifier and ImgCode-8.6M to synthesize novel mathematical figures . |
| Outcome: | The proposed model surpasses GPT-4o and Claude 3.5 Sonnet in the geometry problem-solving subset of MathVista, achieving improvements of 8.9% and 9.2%. |
Revisiting Large Language Models as Zero-shot Relation Extractors (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Recent studies show that large language models (LLMs) transfer well to new tasks out-of-the-box . relationship extraction (RE) involves a certain degree of labeled or unlabeled data even under zero-shot setting. |
| Approach: | They propose a simple prompt recursively using LLMs to transform RE inputs to QA format . they propose qq prompting and qt prompting to improve their results . |
| Outcome: | The proposed method improves on different model sizes, benchmarks and settings. |
CRPO: Confidence-Reward Driven Preference Optimization for Machine Translation (2025.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) have shown great potential in natural language processing tasks, but their application to machine translation remains challenging due to pretraining on predominantly English-centric datasets. |
| Approach: | They propose a method that combines reward scores with model confidence to improve model selection for fine-tuning. |
| Outcome: | The proposed method outperforms existing methods in translation accuracy and data efficiency. |
Unlocking LLMs’ Self-Improvement Capacity with Autonomous Learning for Domain Adaptation (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing models that use self-supervised and instruction fine-tuning can be trained using unlabeled corpora. |
| Approach: | They propose to use unlabeled target corpora to adapt large language models to new domains . they propose to employ self-supervised pre-training and instruction fine-tuning methods . |
| Outcome: | The proposed model can adapt to new domains using only a large amount of unlabeled target corpora. |
EVM-QuestBench: An Execution-Grounded Benchmark for Natural-Language Transaction Code Generation (2026.acl-long)
Copied to clipboard
| Challenge: | Existing evaluations of large language models overlook execution accuracy and safety. |
| Approach: | They propose an execution-grounded benchmark for natural-language transaction-script generation on EVM-compatible chains. |
| Outcome: | The proposed benchmark finds large performance gaps in the models with 5 independent rounds. |
ShieldLM: Empowering LLMs as Aligned, Customizable and Explainable Safety Detectors (2024.findings-emnlp)
Copied to clipboard
Zhexin Zhang, Yida Lu, Jingyuan Ma, Di Zhang, Rui Li, Pei Ke, Hao Sun, Lei Sha, Zhifang Sui, Hongning Wang, Minlie Huang
| Challenge: | Existing tools for detecting safety issues in LLMs are expensive and inefficient. |
| Approach: | They propose an LLM-based safety detector which annotates the safety of queries and provides explanations for its decisions. |
| Outcome: | The proposed detector outperforms baselines on four sets of query-response pairs and is effective as a safety evaluator for advanced LLMs. |
SciAssess: Benchmarking LLM Proficiency in Scientific Literature Analysis (2025.findings-naacl)
Copied to clipboard
Hengxing Cai, Xiaochen Cai, Junhan Chang, Sihang Li, Lin Yao, Wang Changxin, Zhifeng Gao, Hongshuai Wang, Li Yongge, Mujie Lin, Shuwen Yang, Jiankun Wang, Mingjun Xu, Jin Huang, Xi Fang, Jiaxi Zhuang, Yuqi Yin, Yaqi Li, Changhong Chen, Zheng Cheng, Zifeng Zhao, Linfeng Zhang, Guolin Ke
| Challenge: | Existing benchmarks fail to adequately evaluate the proficiency of Large Language Models (LLMs) Existing standards do not cover the skills needed to evaluate LLMs in scientific literature analysis. |
| Approach: | They propose a benchmark to evaluate the proficiency of large language models in scientific literature analysis. |
| Outcome: | SciAssess evaluates 11 LLMs on multiple tasks across scientific fields. |
QiMeng-PRepair: Precise Code Repair via Edit-Aware Reward Optimization (2026.acl-long)
Copied to clipboard
Changxin Ke, Rui Zhang, Jiaming Guo, Yuanbo Wen, Li Ding, Shuo Wang, Xuyuan Zhu, Xiong Peng, Di Huang, Zidong Du, Xing Hu, Qi Guo, Yunji Chen
| Challenge: | Existing approaches to program repair are based on correctness alone. |
| Approach: | They propose a framework that mitigates over-editing and improves repair accuracy by generating buggy programs and re-edits. |
| Outcome: | The proposed framework improves repair precision by 31.4% under fix1@1, a metric that considers repair correctness and extent, and significantly increases decoding throughput when combined with speculative editing. |
Computer Assisted Translation with Neural Quality Estimation and Automatic Post-Editing (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Using neural machine translation to approximate human parity is difficult due to the lack of parallel training corpora. |
| Approach: | They propose an end-to-end deep learning framework for quality estimation and automatic post-editing of machine translation output. |
| Outcome: | The proposed framework achieves state-of-the-art performance on the English–German dataset and human translators can significantly expedite their post-editing processing with the model. |
Mixture-of-Minds: Multi-Agent Reinforcement Learning for Table Understanding (2026.acl-long)
Copied to clipboard
Yuhang Zhou, Mingrui Zhang, Ke Li, Mingyi Wang, Qiao Liu, Qifei Wang, Jiayi Liu, Fei Liu, Serena Li, Weiwei LI, Mingze Gao, Abhishek Kumar, Xiangjun Fan, Zhuokai Zhao, Lizhu Zhang
| Challenge: | Large language models (LLMs) have shown promise on understanding and reasoning over tables, but current approaches remain limited. |
| Approach: | They propose a multi-agent framework that decomposes table reasoning into three specialized roles: planning, coding, and answering. |
| Outcome: | The proposed framework decomposes table reasoning into three specialized roles: planning, coding, and answering. |
Semi-Supervised Learning for Video Captioning (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Existing video captioning algorithms are heavily dependent on supervised training data. |
| Approach: | They propose to train the video captioning model on labeled and unlabeled data jointly in a semi-supervised learning manner. |
| Outcome: | The proposed model outperforms state-of-the-art semi-supervised learning approaches on VATEX, MSR-VTT and MSVD datasets. |
Towards Injecting Medical Visual Knowledge into Multimodal LLMs at Scale (2024.emnlp-main)
Copied to clipboard
Junying Chen, Chi Gui, Ruyi Ouyang, Anningzhe Gao, Shunian Chen, Guiming Chen, Xidong Wang, Zhenyang Cai, Ke Ji, Xiang Wan, Benyou Wang
| Challenge: | Multimodal large language models (MLLMs) lack visual knowledge in medical applications due to data privacy concerns and high annotation costs. |
| Approach: | They refined medical image-text pairs from PubMed and employed MLLMs (GPT-4V) to denoise and reformat the data. |
| Outcome: | The proposed model significantly improves the MMMU Health & Medicine track and shows that it can be used in multimodal scenarios. |
AutoDetect: Towards a Unified Framework for Automated Weakness Detection in Large Language Models (2024.findings-emnlp)
Copied to clipboard
Jiale Cheng, Yida Lu, Xiaotao Gu, Pei Ke, Xiao Liu, Yuxiao Dong, Hongning Wang, Jie Tang, Minlie Huang
| Challenge: | Large Language Models (LLMs) exhibit significant but subtle weaknesses, such as mistakes in instruction-following or coding tasks. |
| Approach: | They propose a framework to automatically expose weaknesses in Large Language Models (LLMs) they use three LLM-powered agents to perform comprehensive weakness identification . |
| Outcome: | The proposed framework shows that it is more effective than untargeted data augmentation methods like Self-Instruct to identify weaknesses in LLMs. |
Gated Mechanism Enhanced Multi-Task Learning for Dialog Routing (2022.coling-1)
Copied to clipboard
| Challenge: | Existing methods for dialog routing are mostly heuristic and cannot achieve high-quality performance. |
| Approach: | They propose a multi-task learning framework with a dialog encoder and two tailored gated mechanism modules to solve this problem. |
| Outcome: | The proposed model can play the role of hierarchical information filtering and is non-invasive to existing dialog systems. |
SeqXGPT: Sentence-Level AI-Generated Text Detection (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for sentence-level AIGT detection are weak . large language models (LLMs) can generate human-like content . |
| Approach: | They propose a sentence-level AIGT detection challenge using LLMs as log probability lists . they propose 'check' GPT' method that uses log probability list features to detect AIGT . |
| Outcome: | The proposed method surpasses baseline methods in sentence- and document-level detection challenges. |
Beyond Logits: Aligning Feature Dynamics for Effective Knowledge Distillation (2025.acl-long)
Copied to clipboard
Guoqiang Gong, Jiaxing Wang, Jin Xu, Deping Xiang, Zicheng Zhang, Leqi Shen, Yifeng Zhang, JunhuaShu JunhuaShu, ZhaolongXing ZhaolongXing, Zhen Chen, Pengzhang Liu, Ke Zhang
| Challenge: | Knowledge distillation (KD) compresses large language models into lightweight versions called student models. |
| Approach: | They propose to align the entire feature dynamics between teacher and student models by using two additional loss terms to achieve this. |
| Outcome: | The proposed method matches the entire feature dynamics between teacher and student models rather than just the final states. |
Watermarking LLMs with Weight Quantization (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Large language models are being deployed at an astonishing speed, exposing users to high risks. |
| Approach: | They propose a method that plants watermarks in quantization process of large language models without pre-defined triggers during inference. |
| Outcome: | The proposed method protects model weights without pre-defined triggers . it works when the model is used in the fp32 mode and remains hidden when the models are quantized to int8 . |
RACQC: Advanced Retrieval-Augmented Generation for Chinese Query Correction (2025.findings-emnlp)
Copied to clipboard
Jinbo Su, Lingzhe Gao, Wei Li, Shihao Liu, Haojie Lei, Xinyi Wang, Yuanzhao Guo, Ke Wang, Daiting Shi, Dawei Yin
| Challenge: | Large language models (LLMs) exhibit remarkable capabilities across many tasks, but face critical challenges in the CSC scenario: (1) poor generalization to rare entities in open-domain searches; and (2) failure to adapt to temporal entity variations due to static parameters, resulting in serious over-correction issues. |
| Approach: | They propose a Chinese Spelling Check system with RAG and multi-task learning that integrates dynamic knowledge retrieval and entity-centric RAG to address rare entities. |
| Outcome: | The proposed system outperforms existing baselines in the CSC task and achieves a maximum improvement of +9.92% on the search scenario benchmark and +3.2% on the general-domain dataset. |
Response-G1: Explicit Scene Graph Modeling for Proactive Streaming Video Understanding (2026.acl-long)
Copied to clipboard
Ke Ma, Jiaqi Tang, Bin Guo, Xueting Han, Ruonan Xu, Qingfeng He, Ziheng Wang, Xu Wang, Qifeng Chen, Zhiwen Yu, Yunhao Liu
| Challenge: | Existing methods for streaming video understanding are query-agnostic and implicitly model video evidence. |
| Approach: | They propose a framework that establishes explicit, structured alignment between the accumulated video evidence and the query’s expected response conditions via scene graphs. |
| Outcome: | The proposed model achieves more interpretable and accurate response timing decisions on both proactive and reactive tasks. |
MTP-RL: Acceleration of Reinforcement Learning Rollouts with Policy-Aligned Multi-Token Prediction (2026.findings-acl)
Copied to clipboard
| Challenge: | Reinforcement learning (RL) is widely applied to boost the performance of pretrained models, yet its training efficiency is severely constrained by rollout generation. |
| Approach: | They propose a framework that accelerates the rollout phase for diverse models by equipping a pipeline to equip the multi-layer parameter-sharing MTP for all models and an advantage-aware MTP optimization strategy. |
| Outcome: | The proposed framework achieves stable growth of acceptance length during RL training, and also accelerates RL rollouts, achieving an average 23.1%–55.3% reduction in rollout time compared to baselines. |
AlignBench: Benchmarking Chinese Alignment of Large Language Models (2024.acl-long)
Copied to clipboard
Xiao Liu, Xuanyu Lei, Shengyuan Wang, Yue Huang, Andrew Feng, Bosi Wen, Jiale Cheng, Pei Ke, Yifan Xu, Weng Lam Tam, Xiaohan Zhang, Lichao Sun, Xiaotao Gu, Hongning Wang, Jing Zhang, Minlie Huang, Yuxiao Dong, Jie Tang
| Challenge: | Effective evaluation of alignment for emerging Chinese LLMs is still significantly lacking, calling for real-scenario grounded, open-ended, challenging and automatic evaluations tailored for alignment. |
| Approach: | They propose a multi-dimensional benchmark for evaluating LLMs’ alignment in Chinese with 8 main categories, 683 real-scenario rooted queries and corresponding human verified references. |
| Outcome: | The benchmark uses a human-in-the-loop data curation pipeline, 683 real-scenario rooted queries and human verified references. |
Adversarial Text Generation via Sequence Contrast Discrimination (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to generate human-like texts are auto-regressive, but they suffer from exposure bias due to the dependence on the previous sampled output during the inferring phase. |
| Approach: | They propose a sequence contrast loss driven text generation framework which learns the difference between real texts and generated texts and uses that difference. |
| Outcome: | The proposed framework improves training stability and quality of generated texts and avoids the time-consuming sampling process. |
McBE: A Multi-task Chinese Bias Evaluation Benchmark for Large Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing datasets on bias evaluation for large language models focus on English and North American culture and are limited to one task. |
| Approach: | They propose to evaluate Chinese language models' biases from multiple perspectives using a multi-task Chinese Bias Evaluation Benchmark. |
| Outcome: | The proposed model covers 12, 82 subcategories and 5 evaluation tasks covering a wide range of categories and content diversity. |
DecompEval: Evaluating Generated Texts as Unsupervised Decomposed Question Answering (2023.acl-long)
Copied to clipboard
| Challenge: | Existing evaluation metrics for natural language generation (NLG) tasks face the challenges on generalization ability and interpretability. |
| Approach: | They propose a metric that evaluates natural language generation tasks as an instruction-style question answering task and utilizes instruction-tuned pre-trained language models without training on evaluation datasets. |
| Outcome: | The proposed metric achieves state-of-the-art performance in untrained metrics for evaluating text summarization and dialogue generation, which exhibits strong dimension-level / task-level generalization ability and interpretability. |
PACE: Prefix-Protected and Difficulty-Aware Compression for Efficient Reasoning (2026.findings-acl)
Copied to clipboard
Ruixiang Feng, Yuntao Wen, Silin Zhou, Ke Shi, Yifan Wang, Ran Le, Zhenwei An, Zongchao Chen, Chen Yang, Guangyue Peng, Yiming Jia, Dongsheng Wang, Tao Zhang, Lisi Chen, Yang Song, Shen Gao, Shuo Shang
| Challenge: | Existing LRMs often suffer from "overthinking" and excessively long reasoning traces . a dual-level framework for length compression of LRM is proposed . |
| Approach: | They propose a framework for prefix-protected and difficulty-aware compression under hierarchical supervision. |
| Outcome: | The proposed framework reduces token usage while improving accuracy on math benchmarks. |
Agent-Pro: Learning to Evolve via Policy-Level Reflection and Optimization (2024.acl-long)
Copied to clipboard
Wenqi Zhang, Ke Tang, Hai Wu, Mengna Wang, Yongliang Shen, Guiyang Hou, Zeqi Tan, Peng Li, Yueting Zhuang, Weiming Lu
| Challenge: | Large Language Models (LLMs) are designed as specific task solvers with sophisticated prompt engineering, but are inherently incapacitating to address complex dynamic scenarios. |
| Approach: | They propose an LLM-based agent with policy-level reflection and optimization that can learn from interactive experiences and progressively elevate its behavioral policy. |
| Outcome: | The proposed agent outperforms vanilla LLM and specialized models in blackjack and Texas hold’em. |
RAG-Instruct: Boosting LLMs with Diverse Retrieval-Augmented Instructions (2025.emnlp-main)
Copied to clipboard
| Challenge: | Retrieval-Augmented Generation (RAG) has emerged as a key paradigm for enhancing large language models by incorporating external knowledge. |
| Approach: | They propose a method for synthesizing diverse and high-quality RAG instruction data based on any source corpus. |
| Outcome: | The proposed method outperforms existing methods in multiple tasks and achieves strong zero-shot performance. |
PROSE: A Pronoun Omission Solution for Chinese-English Spoken Language Translation (2023.emnlp-main)
Copied to clipboard
| Challenge: | Pro-drop (‘pronoun-dropping’) language requires NMT systems to recover omitted pronouns, but this task lacks sufficient datasets for benchmarking . |
| Approach: | They propose a benchmarking method that leverages the semantic embedding of dropped pronouns to augment training pairs to alleviate the negative impact introduced by pro-drop . |
| Outcome: | The proposed method outperforms existing methods regarding omitted pronoun retrieval and overall translation quality on four Chinese-English translation corpora. |
Probability-Consistent Preference Optimization for Enhanced LLM Reasoning (2025.findings-acl)
Copied to clipboard
Yunqiao Yang, Houxing Ren, Zimu Lu, Ke Wang, Weikang Shi, Aojun Zhou, Junting Pan, Mingjie Zhan, Hongsheng Li
| Challenge: | Recent advances in preference optimization have demonstrated significant potential for improving mathematical reasoning capabilities in large language models. |
| Approach: | They propose a framework that establishes two quantitative metrics for preference selection: surface-level answer correctness and intrinsic token-level probability consistency. |
| Outcome: | The proposed framework outperforms existing outcome-only criterion approaches across a diverse range of LLMs and benchmarks. |
InferAligner: Inference-Time Alignment for Harmlessness through Cross-Model Guidance (2024.emnlp-main)
Copied to clipboard
Pengyu Wang, Dong Zhang, Linyang Li, Chenkun Tan, Xinghao Wang, Mozhi Zhang, Ke Ren, Botian Jiang, Xipeng Qiu
| Challenge: | Existing methods for enhancing harmlessness and helpfulness of large language models (LLMs) involve complex and resource-intensive training processes. |
| Approach: | They propose a method that decouples harmlessness from helpfulness during inference phase. |
| Outcome: | The proposed method significantly reduces the attack success rate (ASR) of harmful instructions and jailbreak instructions while maintaining almost unchanged performance in downstream tasks. |
FacLens: Transferable Probe for Foreseeing Non-Factuality in Fact-Seeking Question Answering of Large Language Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing non-factuality detection methods require response generation, which incurs significant computational overhead. |
| Approach: | They propose a lightweight model called Factuality Lens which effectively probes hidden representations of fact-seeking questions for the NFP task. |
| Outcome: | The proposed model is able to probe hidden representations of fact-seeking questions and reduce development costs. |
Adaptive Contrastive Knowledge Distillation for BERT Compression (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing knowledge distillation methods for BERT implicitly learn discriminative student features by mimicking the teacher features. |
| Approach: | They propose a new knowledge distillation approach called adaptive contrastive knowledge distilling for BERT compression using hidden state features in BERT as explicit supervision to learn discriminative student features. |
| Outcome: | The proposed approach improves on multiple natural language processing tasks. |
Improving Neural Machine Translation by Multi-Knowledge Integration with Prompting (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods focus on how to integrate multiple types of knowledge into NMT models . |
| Approach: | They propose a framework that integrates multiple types of knowledge into NMT models . they use multiple types as prefix-prompts of input for the encoder and decoder . |
| Outcome: | The proposed framework outperforms baselines on English-Chinese and English-German translation. |
Training Language Model to Critique for Better Refinement (2025.findings-acl)
Copied to clipboard
Tianshu Yu, Chao Xiang, Mingchuan Yang, Pei Ke, Bosi Wen, Cunxiang Wang, Jiale Cheng, Li Zhang, Xinyu Mu, Chuxiong Sun, Minlie Huang
| Challenge: | Large language models (LLMs) have remarkable evaluation and critique capabilities, providing insightful feedback and identifying flaws in various tasks. |
| Approach: | They propose a framework to train critic models using refinement signals to generate feedback loops where critiques guide the model in refining its responses. |
| Outcome: | The proposed framework outperforms traditional methods and open-source models in terms of critique quality and refinement outcomes. |
Speculating LLMs’ Chinese Training Data Pollution from Their Tokens (2025.emnlp-main)
Copied to clipboard
Qingjie Zhang, Di Wang, Haoting Qian, Liu Yan, Tianwei Zhang, Ke Xu, Qi Li, Minlie Huang, Hewu Li, Han Qiu
| Challenge: | Experiments on GPT and other 23 LLMs indicate that tokens widely exist while GPT’s vocabulary behaves the worst: more than 23% long Chinese tokens (i.e., a token with more than two Chinese characters) are either porn or online gambling. |
| Approach: | They propose to locate Polluted Chinese (PoC) tokens in LLMs and build a PoC token detector to label them in vocabularies by considering each token’s semantics and related contents from the search engines. |
| Outcome: | The proposed method predicts that the ratio of “*” related webpages in GPT-4o's training data is around 0.5%. |
Breaking the Evaluation Paradox: Evaluating High-Entropy Search with Computationally Irreducible Constraints (2026.findings-acl)
Copied to clipboard
| Challenge: | a new framework for evaluation of exhaustive search capabilities is needed . high-entropy enumeration tasks make such ground truth impossible for humans to create . VERITAS is a framework built on the principle of computationally irreducible constraints . |
| Approach: | They propose a framework that uses non-optimizable constraints to create verifiable searches . VERITAS can generate infinite number of test cases with perfect ground truth and precise difficulty control . |
| Outcome: | a new evaluation framework for large language models is based on non-optimizable constraints . the framework can generate infinite number of test cases with perfect ground truth and precise difficulty control . |
MathCanvas: Intrinsic Visual Chain-of-Thought for Multimodal Mathematical Reasoning (2026.acl-long)
Copied to clipboard
Weikang Shi, Aldrich Yu, Rongyao Fang, Houxing Ren, Ke Wang, Aojun Zhou, Changyao Tian, Xinyu Fu, Yuxuan Hu, Zimu Lu, Linjiang Huang, Si Liu, Rui Liu, Hongsheng Li
| Challenge: | Existing approaches to visual chain-of-thought are limited by external tools or fail to generate high-fidelity diagrams. |
| Approach: | They propose a framework to enable large multimodal models with VCoT capabilities . they pre-train a model on a 15.2M-pair corpus and teach it how to leverage visual aids . |
| Outcome: | The proposed framework unlocks complex, human-like visual reasoning in large language models . it pre-trains the model on a 15.2M-pair corpus and fine-tunes it on MathCanvas-Instruct . |
Comprehensive Benchmarking of Long-Form Speech Generation in Diverse Scenarios (2026.findings-acl)
Copied to clipboard
Changhao Pan, Rui Yang, Han Wang, Zhuan Zhou, Xuming He, Wenxiang Guo, Ziyue Jiang, Ruiqi Li, Yu Zhang, Chenyuhao Wen, Ke Lei, Xiang Yin, Jingyu Lu, Zhiyuan Zhu, Zhou Zhao
| Challenge: | Existing evaluation benchmarks for long-form speech are limited to limited domains, creating a significant gap with the diverse downstream applications. |
| Approach: | They propose a benchmark that decomposes "long-form speech quality" into specific, disentangled dimensions. |
| Outcome: | The proposed benchmark decomposes “long-form speech quality” into specific, disentangled dimensions. |
GR1: Reinforcement-Enhanced LLM for Geoscience Reasoning (2026.findings-acl)
Copied to clipboard
Yule Xie, Jiaxin Ding, Cheng Deng, Shiqing Gao, Junran Zhang, Sibo Zhang, Zeyuan Wang, Ke Wu, Xin Ding, Luoyi Fu, Meng Jin, Xinbing Wang
| Challenge: | Recent advances in large language models have demonstrated RL's substantial capacity to enhance multi-step reasoning beyond what supervised instruction tuning achieves. |
| Approach: | They propose a framework that converts multimodal questions into descriptive text . they propose RL-enhanced geoscience reasoning that can be fine-tuned to a text-only level . |
| Outcome: | The proposed framework improves accuracy and accuracy on multimodal questions while preserving answerability and difficulty. |
T2A-Feedback: Improving Basic Capabilities of Text-to-Audio Generation via Fine-grained AI Feedback (2025.acl-long)
Copied to clipboard
Zehan Wang, Ke Lei, Chen Zhu, Jiawei Huang, Sashuai Zhou, Luping Liu, Xize Cheng, Shengpeng Ji, Zhenhui Ye, Tao Jin, Zhou Zhao
| Challenge: | Text-to-audio (T2A) models still struggle to satisfy human preferences for prompt-following and acoustic quality when generating complex multi-event audio. |
| Approach: | They propose to use AI feedback learning to enhance basic capabilities of text-to-audio models . they use a large audio preference dataset to evaluate the model's capabilities . |
| Outcome: | The proposed model improves in simple and complex scenarios with AI feedback learning. |
C3KG: A Chinese Commonsense Conversation Knowledge Graph (2022.findings-acl)
Copied to clipboard
| Challenge: | Existing commonsense knowledge bases organize tuples in an isolated manner, causing problems for chatbots . |
| Approach: | They create a Chinese commonsense conversation knowledge graph which integrates social commonsensm and dialog flow information. |
| Outcome: | The proposed graph incorporates social commonsense knowledge and dialog flow information. |
Black-Box Prompt Optimization: Aligning Large Language Models without Model Training (2024.acl-long)
Copied to clipboard
| Challenge: | Large language models are often not well aligned with human intents, which requires additional training. |
| Approach: | They propose to use Black-Box Prompt Optimization (BPO) to perform alignments on large language models that are not well aligned with human intents. |
| Outcome: | The proposed model outperforms existing models and is model-agnostic. |
MdEval: Massively Multilingual Code Debugging (2026.findings-acl)
Copied to clipboard
Shukai Liu, Linzheng Chai, Jian Yang, Jiajun Shi, He Zhu, Liran Wang, Jin Ke, Wei Zhang, Hualei Zhu, Shuyue Guo, Tao Sun, Jiaheng Liu, Yunlong Duan, Yu Hao, Liqun Yang, Guanglin Niu, Ge Zhang, Zhoujun Li
| Challenge: | Existing benchmarks primarily focus on Python and are limited in terms of language diversity. |
| Approach: | They propose a multilingual debugging benchmark that includes 3.9K test samples of 20 programming languages and introduces the debug instruction corpora MdEval-Instruct by injecting bugs into the correct multilingual queries and solutions. |
| Outcome: | The proposed benchmark includes 3.9K test samples of 20 programming languages and covers the automated program repair task, bug localization task, and bug identification task. |
GigaSpeech 2: An Evolving, Large-Scale and Multi-domain ASR Corpus for Low-Resource Languages with Automated Crawling, Transcription and Refinement (2025.acl-long)
Copied to clipboard
Yifan Yang, Zheshu Song, Jianheng Zhuo, Mingyu Cui, Jinpeng Li, Bo Yang, Yexing Du, Ziyang Ma, Xunying Liu, Ziyuan Wang, Ke Li, Shuai Fan, Kai Yu, Wei-Qiang Zhang, Guoguo Chen, Xie Chen
| Challenge: | GigaSpeech 2 is a large-scale, multi-domain, multilingual speech recognition corpus for low-resource languages. |
| Approach: | They propose a large-scale, multi-domain, multilingual speech recognition corpus for low-resource languages and an automated pipeline for data crawling, transcription, and label refinement. |
| Outcome: | The proposed corpus reduces the word error rate for Thai, Indonesian, and Vietnamese on a realistic YouTube test set by 25% to 40% compared to Whisper large-v3. |
EPO: Explicit Policy Optimization for Strategic Reasoning in LLMs via Reinforcement Learning (2025.acl-long)
Copied to clipboard
Xiaoqian Liu, Ke Wang, Yongbin Li, Yuchuan Wu, Wentao Ma, Aobo Kong, Fei Huang, Jianbin Jiao, Junge Zhang
| Challenge: | Existing methods for strategic reasoning face challenges in adaptability, scalability, and transferring strategies to new contexts. |
| Approach: | They propose an explicit policy optimization model that provides strategies in open-ended action space and can be plugged into arbitrary LLM agents to motivate goal-directed behavior. |
| Outcome: | The proposed model provides strategies in open-ended action space and can be plugged into arbitrary LLM agents to motivate goal-directed behavior. |
Capability Decomposition for Unified Information Extraction via Hierarchical Mixture-of-Experts (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for IE tasks suffer from inconsistent schema representation and implicitly intermediate reasoning . UC-UIE adopts a low-rank adapted hierarchical Mixture-of-Experts adapter for UIE tasks . |
| Approach: | They propose a framework that decomposes IE reasoning into three universal capabilities . UC-UIE adopts a low-rank Adaptation adapter to fine-tune LLMs for IE tasks . |
| Outcome: | The proposed framework outperforms full-parameter tuning methods with 1.24% trainable parameters and outperformed existing methods in generalization and interpretability. |
Pyramid: A Layered Model for Nested Named Entity Recognition (2020.acl-main)
Copied to clipboard
| Challenge: | Named Entity Recognition (NER) is a fundamental NLP task. |
| Approach: | They propose a pyramid-like layered model for Nested Named Entity Recognition . token or text region embeddings are recursively inputted into L flat NER layers . |
| Outcome: | The proposed model achieves state-of-the-art F1 scores in nested NER on ACE-2004, ACE 2005, GENIA, and NNE. |
LM-Searcher: Cross-domain Neural Architecture Search with LLMs via Unified Numerical Encoding (2025.emnlp-main)
Copied to clipboard
Yuxuan Hu, Jihao Liu, Ke Wang, Jinliang Zheng, Weikang Shi, Manyuan Zhang, Qi Dou, Rui Liu, Aojun Zhou, Hongsheng Li
| Challenge: | Recent advances in Large Language Models have opened new avenues for solving complex optimization problems, including Neural Architecture Search (NAS). |
| Approach: | They propose a framework that leverages LLMs for cross-domain neural architecture optimization without extensive domain-specific tuning. |
| Outcome: | The proposed framework achieves competitive performance in both in-domain and out-of-domain tasks. |
Understanding the Dark Side of LLMs’ Intrinsic Self-Correction (2025.acl-long)
Copied to clipboard
Qingjie Zhang, Di Wang, Haoting Qian, Yiming Li, Tianwei Zhang, Minlie Huang, Ke Xu, Hewu Li, Liu Yan, Han Qiu
| Challenge: | Recent studies show that LLMs’ intrinsic self-correction fails without oracle labels as feedback. |
| Approach: | They propose to use one simple task and three complex tasks with state-of-the-art LLMs like ChatGPT, Llama, and DeepSeek to interpret LLM's intrinsic self-correction. |
| Outcome: | The proposed methods reveal the dark side of LLMs’ intrinsic self-correction for different tasks, especially for those failure cases. |
Learning to Maximize Mutual Information for Chain-of-Thought Distillation (2024.findings-acl)
Copied to clipboard
| Challenge: | Knowledge distillation is a technique of transferring knowledge from large, complex models to smaller ones. |
| Approach: | They propose a method utilizing chain-of-thought distillation to transfer knowledge from large, complex models to smaller ones by maximizing mutual information of the representation features of the two tasks. |
| Outcome: | The proposed method outperforms the state-of-the-art knowledge distillation method on four datasets. |
SDPO: Segment-Level Direct Preference Optimization for Social Agents (2025.acl-long)
Copied to clipboard
Aobo Kong, Wentao Ma, Shiwan Zhao, Yongbin Li, Yuchuan Wu, Ke Wang, Xiaoqian Liu, Qicheng Li, Yong Qin, Fei Huang
| Challenge: | Direct Preference Optimization (DPO) has proven effective in aligning LLM behavior with human preferences across various tasks, but is limited in multi-turn social interactions. |
| Approach: | They propose a method which dynamically selects key segments within interactions to optimize multi-turn agent behavior. |
| Outcome: | The proposed methods outperform existing methods and proprietary LLMs on the SOTOPIA benchmark and show that they can improve social intelligence. |
MIO: A Foundation Model on Multimodal Tokens (2025.emnlp-main)
Copied to clipboard
Zekun Moore Wang, King Zhu, Chunpu Xu, Wangchunshu Zhou, Jiaheng Liu, Yibo Zhang, Jessie Wang, Ning Shi, Siyu Li, Yizhi Li, Haoran Que, Zhaoxiang Zhang, Yuanxing Zhang, Ge Zhang, Ke Xu, Jie Fu, Wenhao Huang
| Challenge: | Existing models lack multimodal understanding capabilities, resulting in closed-source model that does not support multimodal interleaved sequences. |
| Approach: | They propose a foundation model built on multimodal tokens capable of understanding and generating speech, text, images, and videos in an end-to-end, autoregressive manner. |
| Outcome: | The proposed model is able to understand speech, text, images, and videos in an end-to-end, autoregressive manner. |
IF-CRITIC: Towards a Fine-Grained LLM Critic for Instruction-Following Evaluation (2026.acl-long)
Copied to clipboard
Bosi Wen, Yilin Niu, Cunxiang Wang, Pei Ke, Xiaoying Ling, Ying Zhang, Aohan Zeng, Hongning Wang, Minlie Huang
| Challenge: | Existing evaluation models for instruction-following have many shortcomings, such as substantial costs and unreliable assessments. |
| Approach: | They propose an LLM critic for fine-grained instruction-following evaluation using a checklist generator and a constraint-level preference optimization method. |
| Outcome: | The proposed model beats strong LLM-as-a-Judge baselines in evaluations under lower computational overhead compared to baselines. |
Watch Every Step! LLM Agent Learning via Iterative Step-level Process Refinement (2024.emnlp-main)
Copied to clipboard
| Challenge: | Recent approaches to enhance agent performance focus on outcome rewards, which may lead to errors or suboptimal actions due to the absence of process supervision signals. |
| Approach: | They propose a step-level framework that provides detailed step-by-step guidance to enhance agent training by using Monte Carlo methods. |
| Outcome: | The proposed framework outperforms strong baselines on three tasks and shows that it is effective in augmenting efficiency and its applicability to diverse models. |
FRAG: A Flexible Modular Framework for Retrieval-Augmented Generation based on Knowledge Graphs (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for retrieval-augmented generation struggle with a trade-off between flexibility and retrieval quality. |
| Approach: | They propose a flexible modular KG-RAG framework that uses query text instead of KGs . they propose to use query text to infer the structural information of reasoning paths . |
| Outcome: | The proposed method achieves state-of-the-art performance with high efficiency and low resource consumption. |
IF-RewardBench: Benchmarking Judge Models for Instruction-Following Evaluation (2026.acl-long)
Copied to clipboard
| Challenge: | Existing benchmarks for instruction-following lack data coverage and oversimplified pairwise evaluation paradigms that misalign with model optimization scenarios. |
| Approach: | They propose a meta-evaluation benchmark for instruction-following that covers diverse instruction and constraint types and a preference graph for each instruction. |
| Outcome: | Extensive experiments on IF-RewardBench show that the proposed benchmark achieves a stronger positive correlation with downstream task performance compared to existing benchmarks. |
Pre-training with Meta Learning for Chinese Word Segmentation (2021.naacl-main)
Copied to clipboard
| Challenge: | Recent studies show that pre-trained models are beneficial to Chinese Word Segmentation (CWS). However, these models lack task-specific prior segmentation knowledge. |
| Approach: | They propose a pre-trained Chinese word segmentation model MetaSeg which incorporates meta learning into a multi-criteria pre-training task. |
| Outcome: | Empirical results show that MetaSeg can achieve new state-of-the-art performance on twelve widely-used CWS datasets and significantly improve model performance in low-resource settings. |
Alignment with Fill-In-the-Middle for Enhancing Code Generation (2025.emnlp-main)
Copied to clipboard
Houxing Ren, Zimu Lu, Weikang Shi, Haotian Hou, Yunqiao Yang, Ke Wang, Aojun Zhou, Junting Pan, Mingjie Zhan, Hongsheng Li
| Challenge: | Existing methods for generating test cases with limited training data are not reliable and may be counterproductive. |
| Approach: | They propose a method that splits code snippets into smaller, granular blocks, creating more diverse DPO pairs from the same test cases. |
| Outcome: | The proposed approach shows significant improvements in code generation tasks on benchmark datasets such as HumanEval (+), MBPP (+), and APPS. |
ORCHID: A Chinese Debate Corpus for Target-Independent Stance Detection and Argumentative Dialogue Summarization (2023.emnlp-main)
Copied to clipboard
| Challenge: | Recent development of large language models (LLMs) have boosted interest on dialogue agents . however, research on these tasks is limited by the insufficiency of public datasets . stance detection and debate summarization are key for engaging argumentative dialogues - but are not available for non-English languages. |
| Approach: | They propose to use ORCHID to benchmark stance detection and debate summarization in Chinese debates. |
| Outcome: | The proposed task is based on 1,218 real-world debates conducted in Chinese on 476 unique topics. |
SATBench: Benchmarking LLMs’ Logical Reasoning via Automated Puzzle Generation from SAT Formulas (2025.emnlp-main)
Copied to clipboard
Anjiang Wei, Yuheng Wu, Yingjia Wan, Tarun Suresh, Huanmi Tan, Zhanke Zhou, Sanmi Koyejo, Ke Wang, Alex Aiken
| Challenge: | SATBench is a benchmark for evaluating the logical reasoning capabilities of large language models (LLMs) through logical puzzles derived from Boolean satisfiability (SAT) problems. |
| Approach: | They propose a benchmark to evaluate logical reasoning capabilities of large language models (LLMs) using logical puzzles derived from Boolean satisfiability problems. |
| Outcome: | The proposed model achieves 65.0% accuracy on hard UNSAT problems, close to the baseline of 50%. |
Enhancing the General Agent Capabilities of Low-Paramter LLMs through Tuning and Multi-Branch Reasoning (2024.findings-naacl)
Copied to clipboard
| Challenge: | Open-source pre-trained Large Language Models exhibit strong language understanding and generation capabilities, making them highly successful in a variety of tasks. |
| Approach: | They propose a method to construct agent-specific data using GPT-4 and supervised fine-tuning . they find that supervised tunning can significantly reduce hallucination outputs and formatting errors in agent tasks . |
| Outcome: | The proposed method improves on five agent tasks of AgentBench. |
Towards Robust Real-World Spreadsheet Understanding with Multi-Agent Multi-Format Reasoning (2026.acl-long)
Copied to clipboard
| Challenge: | Spreadsheets are among the most widely used data formats in real-world applications . existing large language models treat tables as plain text, overlooking layout cues and visual semantics. |
| Approach: | They propose a two-stage multi-agent framework for spreadsheet understanding that adopts a step-by-step reading and reasoning paradigm. |
| Outcome: | Extensive experiments on two spreadsheet datasets show the proposed framework outperforms existing methods on Spreadsheet Bench. |
LLM-Guided Semantic-Aware Clustering for Topic Modeling (2025.acl-long)
Copied to clipboard
| Challenge: | Experimental results show that topic modeling is competitive compared to closed-source methods. |
| Approach: | They propose a semi-supervised topic modeling method that combines LLMs with clustering to improve topic generation and distribution. |
| Outcome: | The proposed method outperforms state-of-the-art methods that utilize GPT-4 on topic alignment and exhibits competitive performance compared to Neural Topic Models on topic quality. |
JointGT: Graph-Text Joint Representation Learning for Text Generation from Knowledge Graphs (2021.findings-acl)
Copied to clipboard
| Challenge: | Existing pre-trained models for knowledgegraph-to-text generation ignore graph structure during encoding and lack elaborate pre-training tasks to explicitly model graph-text alignments. |
| Approach: | They propose a graph-text joint representation learning model called JointGT which incorporates a structure-aware semantic aggregation module into each Transformer layer to preserve the graph structure. |
| Outcome: | The proposed model achieves state-of-the-art performance on various KG-to-text datasets. |
TransSum: Translating Aspect and Sentiment Embeddings for Self-Supervised Opinion Summarization (2021.findings-acl)
Copied to clipboard
| Challenge: | Existing studies focus on unsupervised opinion summarization and treat it as a normal multi-document summarizing task. |
| Approach: | They propose a selfsupervised opinion summarization framework TransSum that learns crucial aspect and sentiment embeddings of reviews using intra- and inter-group invariances. |
| Outcome: | The proposed framework outperforms baselines in generating informative, relevant and low-redundant summaries on three domains. |
Boosting Textural NER with Synthetic Image and Instructive Alignment (2024.findings-acl)
Copied to clipboard
| Challenge: | Named entity recognition (NER) is a key task reliant on textual data. |
| Approach: | They propose a method to transform NER into a multimodal task by using images from the internet as auxiliaries. |
| Outcome: | The proposed method surpasses all text-only baselines and improves F1 score by 1.4% to 2.3% on prominent MNER datasets. |
Gloss-Free End-to-End Sign Language Translation (2023.acl-long)
Copied to clipboard
| Challenge: | a study of sign language translation without gloss annotations focuses on the problem of gloss annotation . gloss annotation is hard to acquire, especially in large quantities, and limits the domain coverage of translation datasets . |
| Approach: | They propose a gloss-free end-to-end sign language translation framework to solve this problem . gloss annotations are hard to acquire, especially in large quantities, they argue . |
| Outcome: | The proposed framework improves sign language translation performance on large-scale datasets . gloss annotations are hard to acquire, especially in large quantities . |
SimRPD: Optimizing Recruitment Proactive Dialogue Agents through Simulator-Based Data Evaluation and Selection (2026.acl-industry)
Copied to clipboard
Zhiyong Cao, Dunqiang Liu, Qi Dai, Haojun Xu, Huai Yuen Khor, Hao Wang, Huan He, Yafei Liu, Ke Ma, Ruqian Shi, Sicheng Zhou, Sijia Yao
| Challenge: | High-quality data in training proactive dialogue agents is scarce, despite fine-tuning and reinforcement learning . a recent study has shown that the effectiveness of supervised fine-touring is limited by the lack of high-quality, domain-specific training data. |
| Approach: | They propose a framework for training recruitment proactive dialogue agents using a high-fidelity user simulator and a multi-dimensional evaluation framework based on Chain-of-Intention. |
| Outcome: | The proposed framework outperforms existing simulator-based data selection strategies in a real-world recruitment scenario. |
Fin-STAR: Structure-as-Semantics to Resolve Implicitness in Financial Retrieval (2026.findings-acl)
Copied to clipboard
Yu Zou, Yan Chen, Lida He, Qi Zhou, Xiaorui Zhou, Aixi Zhong, Yi Wang, Wei Li, Qingyu Wang, Jiatao Li, Wei Gong, Jialei Zeng, Jingmei Zhao, Ke Jiang, Qing Li
| Challenge: | Existing Retrieval-Augmented Generation systems treat structure as a physical navigational skeleton rather than intrinsic semantic knowledge. |
| Approach: | They propose a framework that redefining hierarchy as intrinsic semantics and uses snippets to enrich hierarchical lineage. |
| Outcome: | The proposed framework outperforms state-of-the-art hierarchical and graph-based benchmarks on FinTierQA Gold. |
SEA: Supervised Embedding Alignment for Token-Level Visual-Textual Integration in MLLMs (2025.emnlp-main)
Copied to clipboard
Yuanyang Yin, Yaqi Zhao, Yajie Zhang, Yuanxing Zhang, Ke Lin, Jiahao Wang, Xin Tao, Pengfei Wan, Wentao Zhang, Feng Zhao
| Challenge: | Multimodal Large Language Models (MLLMs) integrate visual and textual inputs, yet modality alignment remains one of the most challenging aspects. |
| Approach: | They propose a token-level supervision alignment method that enables more precise visual-text alignment during pretraining. |
| Outcome: | The proposed method improves performance across various model sizes, with smaller models benefiting the most. |
PerturbScore: Connecting Discrete and Continuous Perturbations in NLP (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Natural language processing (NLP) applications are growing rapidly due to discrete nature of texts. |
| Approach: | They propose to connect discrete perturbations with continuous perturbations to help understand discrete ones in NLP models. |
| Outcome: | The proposed method surpasses methods used in discrete perturbation measuring and can be generalized to different datasets, perturbation methods. |
When 20 Agents Fail to Sort: The Distributed Sorting Benchmark for Scalable Multi-Agent Systems (2026.findings-acl)
Copied to clipboard
| Challenge: | MAS-BENCH isolates coordination under explicit communication constraints . CAMOC significantly improves coordination success and efficiency across backends . |
| Approach: | They propose a distributed-sorting benchmark that isolates coordination under explicit communication constraints. |
| Outcome: | MAS-BENCH improves coordination success and efficiency across backends . CAMOC significantly improves efficiency under shared-state interaction . |
MOA: Multi-Objective Alignment for Role-Playing Agents (2026.acl-long)
Copied to clipboard
| Challenge: | Prior work on role-playing agents relies on supervised fine-tuning or reinforcement learning with scalarized rewards, but these approaches do not address the coordination of multiple reward dimensions during optimization. |
| Approach: | They propose a reinforcement-learning framework that enables multi-dimensional, fine-grained rubric optimization for general RPAs. |
| Outcome: | Experiments on PersonaGym and RoleMRC show that MOA improves multi-dimensional role-playing performance over supervised and standard RL baselines. |
Hierarchical Verbalizer for Few-Shot Hierarchical Text Classification (2023.acl-long)
Copied to clipboard
| Challenge: | Existing work on the hierarchical text classification problem is limited due to the complexity of label hierarchy and intensive labeling cost. |
| Approach: | They propose a path-based few-shot setting and a strict path-basic evaluation metric to further explore few- shot HTC tasks. |
| Outcome: | The proposed framework outperforms those who inject hierarchy through graph encoders on three popular HTC datasets under the few-shot setting. |
RoleLLM: Benchmarking, Eliciting, and Enhancing Role-Playing Abilities of Large Language Models (2024.findings-acl)
Copied to clipboard
Noah Wang, Z.y. Peng, Haoran Que, Jiaheng Liu, Wangchunshu Zhou, Yuhan Wu, Hongcheng Guo, Ruitong Gan, Zehao Ni, Jian Yang, Man Zhang, Zhaoxiang Zhang, Wanli Ouyang, Ke Xu, Wenhao Huang, Jie Fu, Junran Peng
| Challenge: | Large Language Models (LLMs) have paved the way for complex tasks such as role-playing. |
| Approach: | They propose a framework to benchmark, elicit, and enhance role-playing abilities in Large Language Models. |
| Outcome: | The proposed framework improves role-playing abilities with 168,093 samples. |
Flexibly Utilize Memory for Long-Term Conversation via a Fragment-then-Compose Framework (2025.emnlp-main)
Copied to clipboard
Cai Ke, Yiming Du, Bin Liang, Yifan Xiang, Lin Gui, Zhongyang Li, Baojun Wang, Yue Yu, Hui Wang, Kam-Fai Wong, Ruifeng Xu
| Challenge: | Large language models extract useful information from conversation history to enhance the response in long-term conversations. |
| Approach: | They propose a Fragment-then-Compose framework to optimize memory utilization for long-term open-domain conversation. |
| Outcome: | The proposed framework can be used to extract useful information from conversation history . it can be adapted to different situations and improve response generation . |
Defending Large Language Models Against Jailbreaking Attacks Through Goal Prioritization (2024.acl-long)
Copied to clipboard
| Challenge: | Despite the success of jailbreaking attacks, there is a lack of effort in defending against them. |
| Approach: | They propose to integrate goal prioritization at both training and inference stages to counteract this conflict between the goals of being helpful and ensuring safety. |
| Outcome: | The proposed approach reduces the Attack Success Rate (ASR) of jailbreaking from 66.4% to 3.6% for ChatGPT and reduces it from 71.0% to 6.6% for Llama2-13B. |
TabEmb: Joint Semantic-Structure Embedding for Table Annotation (2026.acl-long)
Copied to clipboard
| Challenge: | Existing tables learn by linearizing the 2D table into a 1D token sequence and encoding it with pretrained language models (PLMs) such as BERT, but this leads to limited semantic quality and weaker generalization to unseen or rare values compared to modern LLMs. |
| Approach: | They propose a table annotation module called TabEmb which decouples semantic encoding from structural modeling by creating semantically rich embeddings for each column. |
| Outcome: | Experiments show that TabEmb outperforms baselines on different table annotation tasks. |
Beyond Glass-Box Features: Uncertainty Quantification Enhanced Quality Estimation for Neural Machine Translation (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Quality Estimation (QE) is an essential role in applications of Machine Translation (MT). |
| Approach: | They propose to fuse uncertainty quantification into a pre-trained cross-lingual language model to predict the translation quality. |
| Outcome: | The proposed method achieves state-of-the-art on the datasets of WMT 2020 QE shared task. |
DuSQL: A Large-Scale and Pragmatic Chinese Text-to-SQL Dataset (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing text-to-SQL parsing methods mainly focus on English, but there is no labeled data available for the language . a larges-scale and pragmatic Chinese dataset is used for cross-domain text- to-Sql task . |
| Approach: | They propose a larges-scale Chinese dataset for a cross-domain text-to-SQL task . they analyze questions from several representative applications and use an effective data construction framework . |
| Outcome: | The proposed dataset contains 200 databases, 813 tables, and 23,797 question/SQL pairs. |
HPSS: Heuristic Prompting Strategy Search for LLM Evaluators (2025.findings-acl)
Copied to clipboard
Bosi Wen, Pei Ke, Yufei Sun, Cunxiang Wang, Xiaotao Gu, Jinfeng Zhou, Jie Tang, Hongning Wang, Minlie Huang
| Challenge: | Existing efforts to optimize text evaluation prompts neglect the combinatorial impact of multiple factors, leading to insufficient optimization of the evaluation pipeline. |
| Approach: | They propose to integrate 8 key factors for evaluation prompts and integrate them into an algorithm that searches for well-behaved prompting strategies for LLM evaluators. |
| Outcome: | The proposed method outperforms existing methods and human-designed evaluation prompts on four evaluation tasks. |
CritiqueLLM: Towards an Informative Critique Generation Model for Evaluation of Large Language Model Generation (2024.acl-long)
Copied to clipboard
Pei Ke, Bosi Wen, Andrew Feng, Xiao Liu, Xuanyu Lei, Jiale Cheng, Shengyuan Wang, Aohan Zeng, Yuxiao Dong, Hongning Wang, Jie Tang, Minlie Huang
| Challenge: | Existing models for NLP evaluations lack the ability to generate informative critiques in pointwise grading and pairwise comparison especially without references. |
| Approach: | They propose a method which can acquire pointwise grading critiques with pseudo references and revise these critiques via multi-path prompting to obtain informative evaluation data in different tasks and settings. |
| Outcome: | The proposed method outperforms all open-source models and even GPT-4 in system-level correlations of pointwise grading. |
Unlocking Instructive In-Context Learning with Tabular Prompting for Relational Triple Extraction (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing methods for relational triple extraction (RTE) are unnatural and recast RTE tasks to text-to-text prompting formats. |
| Approach: | They propose a tabular prompting for RTE which frames RTE task into a table generation task and propose an instructive in-context learning which only selects and annotates samples considering triple semantics in massive unlabeled samples. |
| Outcome: | The proposed prompting for RTE with TableIE achieves state-of-the-art performance compared to other methods . the proposed prompts are based on off-the shelf LLMs and are scalable to multiple scenarios . |
CharacterGLM: Customizing Social Characters with Large Language Models (2024.emnlp-industry)
Copied to clipboard
Jinfeng Zhou, Zhuang Chen, Dazhen Wan, Bosi Wen, Yi Song, Jifan Yu, Yongkang Huang, Pei Ke, Guanqun Bi, Libiao Peng, JiaMing Yang, Xiyao Xiao, Sahand Sabour, Xiaohan Zhang, Wenjing Hou, Yijia Zhang, Yuxiao Dong, Hongning Wang, Jie Tang, Minlie Huang
| Challenge: | Character-based dialogue systems (CharacterDial) allow users to customize social characters for social interactions. |
| Approach: | They will collect a large-scale Chinese corpus of characters with diverse categories and behaviors and develop CharacterGLM models to address these challenges. |
| Outcome: | Experiments show that CharacterGLM outperforms most popular open- and closed-source LLMs and performs comparable to GPT-4. |
Disambiguated Lexically Constrained Neural Machine Translation (2023.findings-acl)
Copied to clipboard
| Challenge: | Current approaches to LCNMT assume that pre-specified lexicon constraints are contextually appropriate. |
| Approach: | They propose a framework that disambiguates constraints based on contexts at first and integrates them into LCNMT. |
| Outcome: | The proposed approach outperforms baseline approaches on benchmark datasets and comprehensive experiments in multiple target constraints. |
Exploring Layer Activation Dynamic of CoT via Knowledge Probe (2026.acl-long)
Copied to clipboard
| Challenge: | Chain-of-thought reasoning has emerged as a crucial paradigm for multi-step reasoning tasks. |
| Approach: | They propose a multi-stage probing framework that enforces structured reasoning with three explicit stages: keyword extraction, theorem generation, and computation execution. |
| Outcome: | The proposed framework enforces structured reasoning with three explicit stages: keyword extraction, theorem generation, and computation execution. |
GUI0: Self-Evolving Foundational GUI Agents in Super App Ecosystems (2026.acl-long)
Copied to clipboard
Xinyi Wang, Wei Dai, Kyle Qiao, Ke Wang, Peng Chen, Gang Cao, null Kangqin, Zhongpu Wang, Xiaode Zhang, Yanming Liu, Jihao Gu, Jingtao Xu, Gong Zhi
| Challenge: | Automated interaction with graphical user interfaces (GUIs) is central to general artificial intelligence, but remains challenging within Super App ecosystems. |
| Approach: | They propose a framework synergizing autonomous data synthesis with dual-agent co-evolution . GUI0 establishes a domain-aware foundation model via synthesized corpora and employs curriculum-driven reinforcement learning . |
| Outcome: | The proposed framework outperforms Gemini-2.5-Pro and Claude-4-Sonnet in the SuperAPP benchmark and has universal efficacy across base models. |
Adapt Once, Thrive with Updates: Transferable Parameter-Efficient Fine-Tuning on Evolving Base Models (2025.acl-long)
Copied to clipboard
| Challenge: | Parameter-efficient fine-tuning (PEFT) is a common method for fine- tuning large language models . however, once updated, PEFT modules suffer performance degradation on newer versions . |
| Approach: | They propose a method that enhances the PEFT module by focusing on the task-specific pattern while reducing its dependence on certain knowledge in the base model. |
| Outcome: | Experiments show that PEFT modules can maintain performance on updated models without re-tuning . the proposed approach can be used in real-world applications with large model sizes . |
Bridging Internal Consistency and External Alignment: A Causal and Dynamic Interpretability Framework for LLM Generation (2026.acl-long)
Copied to clipboard
| Challenge: | Existing interpretability methods focus on internal and external aspects of the model . existing explanations often focus on surface correlations or static dependencies . |
| Approach: | They propose a causal and dynamic interpretability framework for Large Language Models . they characterize backdoor-adjusted causal effects of generated prefix and prompt . |
| Outcome: | The proposed framework provides a unified causal view of internal consistency and external alignment in LLM generation dynamics. |
FlowBench: Revisiting and Benchmarking Workflow-Guided Planning for LLM-based Agents (2024.findings-emnlp)
Copied to clipboard
| Challenge: | LLM-based agents are susceptible to undesired planning hallucinations when lacking specific knowledge for expertise-intensive tasks. |
| Approach: | They propose a benchmark to evaluate the efficacy of workflow-guided agent planning by formalizing different formats of workflow knowledge. |
| Outcome: | The proposed benchmark aims to improve the planning reliability of LLM-based agents by incorporating external workflow-related knowledge. |
An Electoral Approach to Diversify LLM-based Multi-Agent Collective Decision-Making (2024.emnlp-main)
Copied to clipboard
| Challenge: | Recent advances in large language models have sparked interest in collaborative LLM agents. |
| Approach: | They propose to integrate various ordinal preferential voting mechanisms into LLMs to improve reasoning capabilities and robustness. |
| Outcome: | The proposed method improves reasoning capabilities and robustness of leading LLMs without complex system designs. |
EviReport: From Reasoned Outlines to Evidence Tracked Long-Form Reports (2026.findings-acl)
Copied to clipboard
Zihan Liu, Jianhui li, Zexin Wang, Fei Sun, Jingjing LI, Zheyuan Li, Ke Xiang, Hang Cui, Houhua Gong, Changhua Pei, Gaogang Xie
| Challenge: | Evidence-intensive reports often produce fluent but under-supported drafts . eviReport is an evidence-grounded workflow for automated long-form report generation . |
| Approach: | They propose an evidence-tracked workflow that organizes corpus evidence into compact, traceable units and retrieves query-relevant subgraphs into retrieval-ready packages. |
| Outcome: | The proposed workflow outperforms baselines in factual coverage, factual accuracy and visual evidence integration. |
ProtoCycle: Reflective Tool-Augmented Planning for Text-Guided Protein Design (2026.findings-acl)
Copied to clipboard
Yutang Ge, Guojiang Zhao, Sihang Li, Zheng Cheng, Zifeng Zhao, Hanchen Xia, Guolin Ke, Linfeng Zhang, Zhifeng Gao, Yu Guang Wang
| Challenge: | Recent deep generative models have already shown encouraging * Equal contribution. |
| Approach: | They propose to use generic instruction-tuned LLMs as direct text-to-sequence generators to achieve this goal. |
| Outcome: | Recent studies show that reflection improves sequence quality and alignment while maintaining competitive foldability. |
CoD, Towards an Interpretable Medical Agent using Chain of Diagnosis (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing large language models (LLMs) are proving to be effective in medical automatic diagnosis, but their interpretability remains unaddressed. |
| Approach: | They propose to use a "Chain-of-Diagnosis" approach to enhance the interpretability of medical automatic diagnosis by outputting the disease confidence distribution. |
| Outcome: | The proposed model outperforms other LLMs on automatic diagnostic tasks across three real-world benchmarks and provides interpretability while ensuring controllability in diagnostic rigor. |
OssCSE: Overcoming Surface Structure Bias in Contrastive Learning for Unsupervised Sentence Embedding (2023.emnlp-main)
Copied to clipboard
Zhan Shi, Guoyin Wang, Ke Bai, Jiwei Li, Xiang Li, Qingjun Cui, Belinda Zeng, Trishul Chilimbi, Xiaodan Zhu
| Challenge: | Recent studies show that contrastive learning is effective in sentence representation learning . but, the surface structure bias is a problem in the current model . |
| Approach: | They propose to combine a sentence with a sub-semantic sentence to investigate the surface structure bias. |
| Outcome: | The proposed model achieves state-of-the-art on standard semantic textual similarity tasks using different pre-trained backbones. |
SAM Decoding: Speculative Decoding via Suffix Automaton (2025.acl-long)
Copied to clipboard
| Challenge: | Speculative decoding (SD) methods are inefficient and rely on single retrieval resources. |
| Approach: | They propose a retrieval-based speculative decoding method that adapts the suffix automaton for efficient draft generation by utilizing the generating text sequence and static text corpus. |
| Outcome: | The proposed method can find the longest suffix match and can be integrated with existing methods to generalize to broader domains. |
At Your Own PACE: A Causal Framework for Evaluating EQ in LLMs (2026.findings-acl)
Copied to clipboard
| Challenge: | Emotional Quotient (EQ) has emerged as a competency for seamless human-AI integration. |
| Approach: | They propose a framework for a closed-loop EQ evaluation using a PACE taxonomy to define four dimensions of LLM EQ. |
| Outcome: | The proposed framework achieves high alignment of 89.31% with human preferences while maintaining robust consistency of 83.6%. |
Easy Guided Decoding in Providing Suggestions for Interactive Machine Translation (2023.acl-long)
Copied to clipboard
| Challenge: | In order to improve translation efficiency, human translators perform post-editing on machine translations to correct errors. |
| Approach: | They propose to use the parameterized objective function of neural machine translation to deal with the TS problem without additional training. |
| Outcome: | The proposed method improves translation quality by 10.6 BLEU and reduces time overhead by 63.4% on benchmark datasets. |
Harmonizing Dense and Sparse Signals in Multi-turn RL: Dual-Horizon Credit Assignment for Industrial Sales Agents (2026.acl-industry)
Copied to clipboard
| Challenge: | Large language models for industrial sales require balancing long-term commercial objectives with immediate linguistic constraints such as fluency and compliance. |
| Approach: | They propose a framework that disentangles optimization across time scales by normalizing advantages from turn-level and session-level rewards before fusion. |
| Outcome: | The proposed framework outperforms the state-of-the-art GRPO model in conversion rate and identity detection rate. |
EquiBench: Benchmarking Large Language Models’ Reasoning about Program Semantics via Equivalence Checking (2025.emnlp-main)
Copied to clipboard
Anjiang Wei, Jiannan Cao, Ran Li, Hongyu Chen, Yuhui Zhang, Ziheng Wang, Yuan Liu, Thiago S. F. X. Teixeira, Diyi Yang, Ke Wang, Alex Aiken
| Challenge: | EquiBench is a new benchmark to evaluate large language models' ability to reason about program semantics . Unlike natural language, code is executable. |
| Approach: | They propose a benchmark to evaluate large language models through equivalence checking . EquiBench consists of 2400 program pairs across four languages and six categories . |
| Outcome: | The proposed benchmark consists of 2400 program pairs across four languages and six categories. |
CMNEE:A Large-Scale Document-Level Event Extraction Dataset Based on Open-Source Chinese Military News (2024.lrec-main)
Copied to clipboard
| Challenge: | Current research focuses on the general news or financial domains, with relatively few studies for military domain. |
| Approach: | They propose to annotate Chinese military news events from documents using a schema for the military domain. |
| Outcome: | The proposed dataset is large-scale, document-level open-source for the military domain . it contains 17,000 documents and 29,223 events, which are all manually annotated . |
Open World Classification with Adaptive Negative Samples (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing models with no effective open category data during training are limited by the lack of effective open categories data during the training stage. |
| Approach: | They propose an approach to generate effective open category samples in the training stage and without requiring prior knowledge or external datasets. |
| Outcome: | The proposed approach generates effective synthetic open category samples in the training stage and without requiring any prior knowledge or external datasets. |
FuseSearch: Learning Adaptive Parallel Execution for Efficient Code Localization (2026.findings-acl)
Copied to clipboard
Ke Xu, Siyang Xiao, Ming Liang, Yichen Yu, Zhixiang Wang, Jingxuan Xu, Dajun Chen, Wei Jiang, Yong Li
| Challenge: | Existing parallel code localization agents suffer from a 34.9% redundant tool invocation rate . specialized localization agent that operate as dedicated search components is needed to achieve high localization accuracy. |
| Approach: | They propose a parallel code localization system that reframes parallel code execution as a quality–efficiency co-optimization problem. |
| Outcome: | The proposed method matches SOTA performance while being 93.6% faster. |
NeKo: Cross-Modality Post-Recognition Error Correction with Tasks-Guided Mixture-of-Experts Language Model (2025.acl-industry)
Copied to clipboard
Yen-Ting Lin, Zhehuai Chen, Piotr Zelasko, Zhen Wan, Xuesong Yang, Zih-Ching Chen, Krishna C Puvvada, Ke Hu, Szu-Wei Fu, Jun Wei Chiu, Jagadeesh Balam, Boris Ginsburg, Yu-Chiang Frank Wang, Chao-Han Huck Yang
| Challenge: | Existing methods to train a model on a mixture of domain datasets require separate correction language models. |
| Approach: | They propose a multi-task correction MoE that trains experts to become an "expert" of speech-to-text, language-totext and vision-to text datasets by learning to route each dataset’s tokens to its mapped expert. |
| Outcome: | The proposed model outperforms GPT-3.5 and Claude-3.5-Sonnet on the Open ASR Leaderboard and reaches an average relative 5.0% WER reduction and substantial improvements in BLEU scores. |