Papers by Jaewoo Kang
FaVIQ: FAct Verification from Information-seeking Questions (2022.acl-long)
Copied to clipboard
| Challenge: | Existing fact verification datasets with crowdsourced claims introduce subtle biases that are difficult to control for. |
| Approach: | They construct a large-scale fact verification dataset with ambiguous questions . they use a corpus of 188k claims to construct false and true claims . |
| Outcome: | The proposed dataset outperforms models trained on the dataset FEVER or in-domain data by up to 17% absolute. |
Outlier-Safe Pre-Training for Robust 4-Bit Quantization of Large Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Quantization is a practical solution for deploying Large Language Models in resource-constrained environments. |
| Approach: | They propose an outlier-safe pre-training approach that prevents outlier formation . they validate a 1.4B-parameter model on 1 trillion tokens with no outliers . |
| Outcome: | The proposed model achieves a 35.7 average score on 1 trillion tokens with 2% training overhead. |
Generating Information-Seeking Conversations from Unlabeled Documents (2022.emnlp-main)
Copied to clipboard
| Challenge: | a novel framework for conversational question answering from unlabeled documents has been proposed . a large-scale dataset of synthetic conversations is available for use in real-world applications . |
| Approach: | They propose a framework for conversational question answering from unlabeled documents . they propose 'SimSeek' framework that simulates conversation from unlabelled documents based on two scenarios . |
| Outcome: | The proposed framework achieves state-of-the-art performance on a recent CQA benchmark, QuAC. |
Rationale-Guided Retrieval Augmented Generation for Medical Question Answering (2025.naacl-long)
Copied to clipboard
Jiwoong Sohn, Yein Park, Chanwoong Yoon, Sihyeon Park, Hyeon Hwang, Mujeen Sung, Hyunjae Kim, Jaewoo Kang
| Challenge: | Large language models (LLMs) struggle with hallucinations and outdated knowledge. |
| Approach: | They propose a retrieval-augmented generation framework for enhancing the reliability of RAG in biomedical contexts. |
| Outcome: | The proposed framework outperforms the previous best medical RAG model by up to 5.6% across three medical question-answering benchmarks. |
Automatic Creation of Named Entity Recognition Datasets by Querying Phrase Representations (2023.acl-long)
Copied to clipboard
| Challenge: | Named entity recognition models rely on domain-specific dictionaries provided by experts . however, such dictionary sets are infeasible in many domains where they do not exist . |
| Approach: | They propose a framework that generates NER datasets with high-coverage pseudo-dictionaries . phrase retrieval models are used to retrieve popular entities rather than rare ones . |
| Outcome: | The proposed framework outperforms the previous best model by an average F1 score of 4.7 across five NER benchmark datasets. |
Benchmarking Direct Preference Optimization for Medical Large Vision–Language Models (2026.findings-eacl)
Copied to clipboard
| Challenge: | Large vision-language models (LVLMs) are gaining traction in clinical tasks such as diagnostic support, report generation, and medical question answering. |
| Approach: | They present a systematic evaluation of nine DPO variants applied to two leading medical LVLMs. |
| Outcome: | The proposed model improves alignment and reduces severe hallucinations, but yields inconsistent gains over supervised fine-tuning. |
ETHIC: Evaluating Large Language Models on Long-Context Tasks with High Information Coverage (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing evaluation methods do not assess whether large language models fully utilize contextual information. |
| Approach: | They introduce a new metric to assess LLMs' ability to fully utilize contextual information. |
| Outcome: | The proposed benchmark comprises 1,986 test instances spanning four long-context tasks with high IC scores in the domains of books, debates, medicine, and law. |
Ranking Paragraphs for Improving Answer Recall in Open-Domain Question Answering (D18-1)
Copied to clipboard
| Challenge: | Recent work has combined open-domain question answering with machine comprehension models to find answers in a large knowledge source. |
| Approach: | They propose a machine comprehension model that ranks paragraphs of retrieved documents for a higher answer recall with less noise. |
| Outcome: | The proposed model improves on four open-domain QA datasets by 7.8% on average. |
Learning Dense Representations of Phrases at Scale (2021.acl-long)
Copied to clipboard
| Challenge: | Existing phrase retrieval models rely on sparse representations and still underperform retriever-reader approaches. |
| Approach: | They propose a method to learn phrase representations from reading comprehension tasks using negative sampling methods. |
| Outcome: | The proposed model improves over previous models by 15%-25% absolute accuracy and matches the performance of state-of-the-art retrieval models. |
Learn to Resolve Conversational Dependency: A Consistency Training Framework for Conversational Question Answering (2021.acl-long)
Copied to clipboard
| Challenge: | Existing approaches do not explicitly train QA models on how to resolve conversational dependency, and thus these models are limited in understanding human dialogues. |
| Approach: | They propose a framework that generates self-contained questions that can be understood without the conversation history and then trains a QA model with the pairs of original and self-constructed questions using a consistency-based regularizer. |
| Outcome: | The proposed framework improves the models’ performance by up to 1.2 F1 on QuAC, and 5.2 F1 for CANARD, while addressing the limitations of the existing approaches. |
CLAG: Adaptive Memory Organization via Agent-Driven Clustering for Small Language Model Agents (2026.findings-acl)
Copied to clipboard
| Challenge: | Large language model agents rely on external memory to support knowledge reuse and reasoning tasks. |
| Approach: | They propose a CLustering-based AGentic memory framework where an agent actively organizes memory . they employ an SLM-agent driven router to assign each new memory to a semantically coherent cluster . |
| Outcome: | The proposed framework improves answer quality and robustness over previous memory systems. |
ToxReason: A Benchmark for Mechanistic Chemical Toxicity Reasoning via Adverse Outcome Pathway (2026.findings-acl)
Copied to clipboard
| Challenge: | Recent advances in large language models have enabled molecular reasoning for property prediction. however, toxicity arises from complex biological mechanisms, necessitating mechanistic reasoning for reliable prediction. |
| Approach: | They propose a benchmark that evaluates organ-level toxicity reasoning across multiple organs . they find strong predictive performance does not necessarily imply reliable reasoning . |
| Outcome: | The proposed benchmark evaluates toxicity prediction performance and reasoning quality across LLMs. |
Biomedical NER for the Enterprise with Distillated BERN2 and the Kazu Framework (2022.emnlp-industry)
Copied to clipboard
| Challenge: | a low overall error rate is unavoidable in the field of bioNLP, argues a new study. |
| Approach: | They propose a highly extensible open source framework that supports BioNLP for the pharmaceutical sector. |
| Outcome: | The proposed framework is extensible and scalable and can support BioNLP for the pharmaceutical sector. |
CompAct: Compressing Retrieved Documents Actively for Question Answering (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods to condense extensive documents with no loss of information are difficult to implement in real-world scenarios. |
| Approach: | They propose a framework that employs an active strategy to condense extensive documents without losing key information. |
| Outcome: | The proposed framework improves performance and compression rate on multi-hop question-answering benchmarks. |
Tree of Clarifications: Answering Ambiguous Questions with Retrieval-Augmented Large Language Models (2023.emnlp-main)
Copied to clipboard
| Challenge: | In open-domain question answering, users often ask ambiguous questions (AQs) . one approach is to identify all possible interpretations of the AQ and generate a long-form answer addressing them all. |
| Approach: | They propose a framework that generates a long-form answer addressing all possible interpretations of an ambiguous question. |
| Outcome: | The proposed framework outperforms baselines on ASQA in a few-shot setup across metrics while surpassing fully-supervised baselines trained on the whole training set in terms of Disambig-F1 and Disambigo-ROUGE. |
Consistency Training with Virtual Adversarial Discrete Perturbation (2022.naacl-main)
Copied to clipboard
| Challenge: | Existing methods for regularizing a model are agnostic to the training model and may not be effective for perturbed inputs. |
| Approach: | They propose an augmentation method of adding a discrete noise that would incur the highest divergence between predictions by replacing tokens while keeping original semantics. |
| Outcome: | The proposed method outperforms baselines on semi-supervised text classification tasks and a robustness benchmark. |
Learning from Negative Samples in Biomedical Generative Entity Linking (2025.findings-acl)
Copied to clipboard
| Challenge: | Generative models are usually trained only with positive samples and do not explicitly learn from hard negative samples, which are entities that look similar but have different meanings. |
| Approach: | They propose a framework that trains generative BioEL models using negative samples to learn from hard negative samples. |
| Outcome: | The proposed framework outperforms baseline models by up to an average top-1 accuracy of 1.4% on five benchmarks. |
CookingSense: A Culinary Knowledgebase with Multidisciplinary Assertions (2024.lrec-main)
Copied to clipboard
| Challenge: | CookingSense is a descriptive collection of knowledge assertions in the culinary domain extracted from various sources, including web data, scientific papers, and recipes. |
| Approach: | They introduce CookingSense, a descriptive collection of knowledge assertions in the culinary domain extracted from various sources, including web data, scientific papers, and recipes. |
| Outcome: | The proposed system improves retrieval augmented language models and food decision support systems. |
Adversarial Subword Regularization for Robust Neural Machine Translation (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for segmenting words into subword units are not robust enough to handle multiple subword candidates. |
| Approach: | They propose to regularize subword segmentations that maximize the translation loss by using gradient signals during training to prevent erroneous segmentations of unseen words. |
| Outcome: | The proposed method improves the performance of NMT models on low-resource and out-domain datasets. |
Simple Questions Generate Named Entity Recognition Datasets (2022.emnlp-main)
Copied to clipboard
| Challenge: | Recent named entity recognition models rely on human-annotated datasets . however, in-domain dictionaries and sentences are often unavailable or expensive to construct for many entity types. |
| Approach: | They propose an ask-to-generate approach which automatically generates NER datasets by asking natural language questions to an open-domain question answering system. |
| Outcome: | The proposed model outperforms the previous best model by 19.5 F1 score on six benchmarks and achieves state-of-the-art performance. |
Assessing LLM Reasoning Steps via Principal Knowledge Grounding (2025.findings-emnlp)
Copied to clipboard
Hyeon Hwang, Yewon Cho, Chanwoong Yoon, Yein Park, Minju Song, Kyungjae Lee, Gangwoo Kim, Jaewoo Kang
| Challenge: | Step-by-step reasoning has become a standard approach for large language models to tackle complex tasks. |
| Approach: | They propose a framework that assesses the knowledge grounding of intermediate reasoning by using a large-scale repository of atomic knowledge essential for reasoning. |
| Outcome: | The evaluation suite identifies missing or misapplied knowledge elements and provides crucial insights for uncovering fundamental reasoning deficiencies in LLMs. |
Fine-tuning CLIP Text Encoders with Two-step Paraphrasing (2024.findings-eacl)
Copied to clipboard
| Challenge: | Contrastive language-image pre-training models have demonstrated considerable success across various vision-language tasks, such as text-to-image retrieval. |
| Approach: | They propose a fine-tuning approach to enhance the representations of CLIP models for paraphrases by leveraging large language models. |
| Outcome: | The proposed model improves on baseline models across paraphrased retrieval, visual genome relation and attribution, and seven semantic textual similarity tasks. |
Tracing Mathematical Proficiency Through Problem-Solving Processes (2026.findings-acl)
Copied to clipboard
| Challenge: | Knowledge Tracing (KT) models a learner's evolving knowledge state over time, but lacks the rich information embedded in students' problem-solving processes. |
| Approach: | They propose a framework that uses a teacher-student-teacher pipeline to extract students’ Mathematical Proficiency (MP) as intermediate representation. |
| Outcome: | The proposed framework improves the prediction performance of existing KT methods and provides interpretable explanations by explicitly modeling students’ mathematical proficiency. |
“Killing Me” Is Not a Spoiler: Spoiler Detection Model using Graph Neural Networks with Dependency Relation-Aware Attention Mechanism (2021.eacl-main)
Copied to clipboard
| Challenge: | Several attention-based spoiler detection models are insufficient for utilizing dependency relations between context words. |
| Approach: | They propose a new spoiler detection model called SDGNN that uses syntax-aware graph neural networks to detect dependency relations between context words. |
| Outcome: | The proposed model outperforms existing models on two real-world benchmark datasets. |
SCRIPTMIND: Crime Script Inference and Cognitive Evaluation for LLM-based Social Engineering Scam Detection System (2026.eacl-industry)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have shown promise in identifying deception, but their cognitive assistance potential remains underexplored. |
| Approach: | They propose a framework for LLM-based scam detection that bridges automated reasoning and human cognition. |
| Outcome: | The proposed framework outperforms GPT-4o in the Korean scam detection and phone scam simulations. |
Med-PRM: Medical Reasoning Models with Stepwise, Guideline-verified Process Rewards (2025.emnlp-main)
Copied to clipboard
Jaehoon Yun, Jiwoong Sohn, Jungwoo Park, Hyunjae Kim, Xiangru Tang, Daniel Shao, Yong Hoe Koo, Ko Minhyeok, Qingyu Chen, Mark Gerstein, Michael Moor, Jaewoo Kang
| Challenge: | Large language models have shown promise in clinical decision making, but current approaches struggle to localize and correct reasoning errors at specific steps of the reasoning process. |
| Approach: | They propose a process reward modeling framework that leverages retrieval-augmented generation to verify each reasoning step against established medical knowledge bases. |
| Outcome: | The proposed model improves on five medical QA benchmarks and two open-ended diagnostic tasks by 13.50% on MedQA. |
Contextualized Sparse Representations for Real-Time Open-Domain Question Answering (2020.acl-main)
Copied to clipboard
| Challenge: | Existing phrase retrieval models suffer from low accuracy due to limited scalability and speed . 'Open-domain question answering' is a task of answering generic factoid questions by looking up a large knowledge source, typically unstructured text corpora such as Wikipedia. |
| Approach: | They aim to augment existing phrase retrieval models with contextualized sparse representations to improve the quality of each phrase embedding. |
| Outcome: | The proposed model improves CuratedTREC and SQuAD-Open by 4% and 45x faster inference speeds over the existing model. |
Can Language Models be Biomedical Knowledge Bases? (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies have focused on probing LMs in the general domain but little attention has been given to whether they can be used as domain knowledge bases. |
| Approach: | They propose to use 49K biomedical factual knowledge triples to probe LMs for biomedically . they find that biomedic LM can achieve up to 18.51% Acc@5 on retrieving biomedcial knowledge. |
| Outcome: | The proposed biomedical factual knowledge probing benchmark achieves 18.51% Acc@5 on biomedically-relevant knowledge retrieval. |
Ask Optimal Questions: Aligning Large Language Models with Retriever’s Preference in Conversation (2025.findings-naacl)
Copied to clipboard
| Challenge: | Existing methods to perform conversational search are sub-optimal due to the limited ability to incorporate signals from the retrieval results. |
| Approach: | They propose to optimize a language model for reformulating search queries in line with retrievers’ preferences by combining a large-scale dataset with Retrievers’ Feedback. |
| Outcome: | The proposed framework outperforms existing methods on two benchmarks and surpasses the state-of-the-art methods. |
Optimizing Test-Time Query Representations for Dense Retrieval (2023.findings-acl)
Copied to clipboard
| Challenge: | Recent developments of dense retrieval rely on quality representations of queries and contexts from pre-trained query and context encoders. |
| Approach: | They propose a test-time optimization of query representations that provides fine-grained pseudo labels over retrieval results. |
| Outcome: | The proposed algorithm improves open-domain question answering accuracy and direct re-ranking by up to 2.0% while running 1.3–2.4x faster with an efficient implementation. |
Biomedical Entity Representations with Synonym Marginalization (2020.acl-main)
Copied to clipboard
| Challenge: | Biomedical named entities are often used as key features in biomedical text mining. |
| Approach: | They propose to use a model-based candidate selection to maximize the marginal likelihood of the synonyms present in top candidates. |
| Outcome: | The proposed model outperforms previous state-of-the-art models on four biomedical entity normalization datasets with three different entity types. |
Does Time Have Its Place? Temporal Heads: Where Language Models Recall Time-specific Information (2025.acl-long)
Copied to clipboard
| Challenge: | Temporal Heads are attention heads that primarily handle temporal knowledge. |
| Approach: | They discover Temporal Heads, specific attention heads that primarily handle temporal knowledge, through circuit analysis. |
| Outcome: | The proposed models can handle temporal knowledge without compromising time-invariant and question-answering performances. |
Look at the First Sentence: Position Bias in Question Answering (2020.emnlp-main)
Copied to clipboard
| Challenge: | Extractive question answering models are trained to predict start and end positions of answers . recent QA models outperform humans in some datasets due to their simplicity and effectiveness. |
| Approach: | They propose to use prior distribution of answer positions as a bias model to reduce position bias. |
| Outcome: | The proposed model outperforms BERT from 37.48% to 81.64% when trained on a biased SQUAD dataset. |