Papers by Joonsuk Park
Plug-and-Play Adaptation for Continuously-updated QA (2022.findings-acl)
Copied to clipboard
| Challenge: | Existing tasks to assess LMs’ efficacy as KBs do not adequately consider multiple large-scale updates. |
| Approach: | They propose a task where multiple large-scale updates are made to language models and plug-in modules are used to handle the updates. |
| Outcome: | The proposed method outperforms existing methods on zsRE QA and NQ datasets and is 4x more effective in terms of updates/forgets ratio compared to a fine-tuning baseline. |
AdvisorQA: Towards Helpful and Harmless Advice-seeking Question Answering with Collective Intelligence (2025.naacl-long)
Copied to clipboard
| Challenge: | AdvisorQA aims to improve LLMs’ capability to offer advice for deeply subjective concerns, utilizing the LifeProTips Reddit forum. |
| Approach: | They propose a dataset to train LLMs' ability to offer advice for deeply subjective concerns, utilizing the LifeProTips Reddit forum. |
| Outcome: | The proposed model improves usefulness through automatic metric, GPT-4 and human evaluations, and expands independent evaluation axis to include harmlessness. |
Argument Quality Assessment in the Age of Instruction-Following Large Language Models (2024.lrec-main)
Copied to clipboard
Henning Wachsmuth, Gabriella Lapesa, Elena Cabrio, Anne Lauscher, Joonsuk Park, Eva Maria Vecchi, Serena Villata, Timon Ziegenbein
| Challenge: | Argument quality assessment is critical for opinion formation, decision making, writing education, and the like. |
| Approach: | They propose to use large language models to leverage knowledge across contexts to enable a much more reliable assessment. |
| Outcome: | The proposed approach improves the quality of argumentation and the ability to leverage knowledge across contexts. |
Return of EM: Entity-driven Answer Set Expansion for QA Evaluation (2025.coling-main)
Copied to clipboard
| Challenge: | Recent studies show that using large language models (LLMs) is the most reliable method to evaluate QA models, but suffers from limited interpretability, high cost, and environmental harm. |
| Approach: | They propose to use soft exact match (EM) with entity-driven answer set expansion to expand gold answer set to include diverse surface forms. |
| Outcome: | The proposed method outperforms traditional evaluation methods while offering the benefits of high interpretability and reduced environmental harm. |
Hierarchical Deconstruction of LLM Reasoning: A Graph-Based Framework for Analyzing Knowledge Utilization (2024.emnlp-main)
Copied to clipboard
| Challenge: | Despite advances in large language models, how they use their knowledge for reasoning is not yet well understood. |
| Approach: | They propose a method that deconstructs complex real-world questions into a graph . they quantify forward discrepancy, a discrepany in LLM performance on simpler sub-problems . |
| Outcome: | The proposed method shows that smaller models exhibit more discrepancies than larger models . it also shows that guiding models from simpler to complex questions improves performance . |
DADA: Distribution-Aware Domain Adaptation of PLMs for Information Retrieval (2024.findings-acl)
Copied to clipboard
| Challenge: | Pre-trained language models struggle with out-of-domain data due to distribution shifts . generative domain adaptation (DA) methods have been proposed to address these issues . |
| Approach: | They propose a distribution-aware domain adaptation method to address distribution shifts in domains . they use observation-level feedback and observation- level feedback to adapt to the target domain . |
| Outcome: | The proposed method adapts to the domain distribution knowledge at the level of a single document and the corpus and expands document representation to unseen gold query terms using domain and observation feedback. |
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. |
Enhancing Hallucination Detection via Future Context (2026.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are widely used to generate plausible text on online platforms, without revealing the generation process. |
| Approach: | They propose a framework for detection of hallucinations in black-box generators by analyzing future contexts. |
| Outcome: | The proposed framework improves on existing methods and demonstrates that it is feasible to integrate it with other models. |
Critic-Guided Decoding for Controlled Text Generation (2023.findings-acl)
Copied to clipboard
| Challenge: | Recent work has demonstrated reinforcement learning and weighted decoding as effective approaches to achieve a higher level of language control and quality with pros and cons. |
| Approach: | They propose a method that combines reinforcement learning and weighted decoding to train a critic from reward models. |
| Outcome: | The proposed method generates more coherent and well-controlled texts than previous methods on three controlled generation tasks, topic control, sentiment control, and detoxification. |
KoSBI: A Dataset for Mitigating Social Bias Risks Towards Safer Large Language Model Applications (2023.acl-industry)
Copied to clipboard
| Challenge: | Existing research and resources are not readily applicable in South Korea due to the differences in language and culture, both of which significantly affect the biases and targeted demographic groups. |
| Approach: | They propose a social bias dataset of 34k pairs of contexts and sentences in Korean covering 72 demographic groups in 15 categories. |
| Outcome: | The proposed dataset reduces social biases by 16.47%p on average for HyperClova (30B and 82B), and GPT-3. |
Automated Fact-Checking of Claims from Wikipedia (2020.lrec-1)
Copied to clipboard
| Challenge: | Fact checking datasets such as FEVER and SNLI suffer from limited applicability due to synthetic nature of claims and/or evidence written by annotators that differ from real claims and evidence on the internet. |
| Approach: | They present a dataset of 124k+ triples consisting of a claim, context and an evidence document extracted from English Wikipedia articles and citations. |
| Outcome: | The proposed dataset is the largest fact checking dataset consisting of real claims and evidence to date. |
LifeTox: Unveiling Implicit Toxicity in Life Advice (2024.naacl-short)
Copied to clipboard
| Challenge: | Existing safety benchmarks and red teaming prompts fail to capture implicit toxicity in complex real-life advice-seeking scenarios. |
| Approach: | They propose a dataset designed for identifying implicit toxicity within advice-seeking scenarios. |
| Outcome: | The proposed dataset matches or surpasses the zero-shot performance of large language models in toxicity classification tasks. |
Argument Mining for Review Helpfulness Prediction (2022.emnlp-main)
Copied to clipboard
| Challenge: | Argumentational features have been shown to be promising indicators of product review helpfulness, but their utility has been limited due to the lack of resources and large-scale experiments investigating their utility. |
| Approach: | They present an argumentational argumentation model that annotates 878 Amazon reviews on headphones and uses it to evaluate argument quality. |
| Outcome: | The proposed model improves the state-of-the-art model under text-only and text-and-image settings. |
Relevance-assisted Generation for Robust Zero-shot Retrieval (2023.emnlp-industry)
Copied to clipboard
| Challenge: | Despite strong in-domain performance, dense retrievers have shown poor generalization to out-of-domain zero-shot tasks where no training queries are available. |
| Approach: | They propose to generate domain-specific pseudo queries for fine-tuning with domain-relevant relevance between PQ and documents. |
| Outcome: | The proposed approach is more robust to domain shifts, validated on BEIR zero-shot tasks. |
ClaimDiff: Comparing and Contrasting Claims on Contentious Issues (2023.findings-acl)
Copied to clipboard
| Challenge: | Using fact verification tasks, however, can not detect subtle differences in factually consistent claims, which might bias the readers. |
| Approach: | They propose a novel dataset that primarily focuses on comparing the nuance between claim pairs. |
| Outcome: | The proposed dataset shows that human-labeled 2,941 claim pairs are weaker than baselines, showing a 19% absolute gap with the baselines. |
mRedditSum: A Multimodal Abstractive Summarization Dataset of Reddit Threads with Images (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing summarization datasets do not cover multimodal discussions, multiple modalities, or both . mRedditSum consists of 3,033 discussion threads and images with human-written summaries. |
| Approach: | They propose a multimodal discussion summarization dataset that annotates 3,033 discussion threads with a human-written summary. |
| Outcome: | The proposed method outperforms existing models and serves as competitive baseline for future work. |
SQuARe: A Large-Scale Dataset of Sensitive Questions and Acceptable Responses Created through Human-Machine Collaboration (2023.acl-long)
Copied to clipboard
Hwaran Lee, Seokhee Hong, Joonsuk Park, Takyoung Kim, Meeyoung Cha, Yejin Choi, Byoungpil Kim, Gunhee Kim, Eun-Ju Lee, Yong Lim, Alice Oh, Sangchul Park, Jung-Woo Ha
| Challenge: | Existing studies focus on coping with social harms that large language models pose . however, discussions on sensitive issues can become toxic even if the users are well-intentioned. |
| Approach: | They propose to use Korean dataset to test whether LLMs can generate offensive content and propagate prejudices. |
| Outcome: | The proposed dataset shows that acceptable response generation improves for HyperCLOVA and GPT-3. |
ArgInstruct: Specialized Instruction Fine-Tuning for Computational Argumentation (2025.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) have been trained to follow instructions for many NLP tasks, including several tasks from computational argumentation (CA), the computational analysis and synthesis of natural language arguments. |
| Approach: | They propose a specialized instruction fine-tuning for the domain of computational argumentation (CA) they synthesized 52k CA-related instructions and used them to train a CA-specialized instruction-following LLM. |
| Outcome: | The proposed benchmarks show that the LLMs can tackle unseen and seen tasks while maintaining generalization capabilities. |
Asking Clarification Questions to Handle Ambiguity in Open-Domain QA (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Ambiguous questions persist in open-domain question answering because formulating a precise question with a unique answer is often challenging. |
| Approach: | They propose to ask a clarification question where the user’s response will help identify the interpretation that best aligns with the user's intention. |
| Outcome: | The proposed approach achieves F1 of 61.3, 25.1, and 40.5 on the three tasks, demonstrating the need for further improvements while providing competitive baselines for future work. |
Are LLM-Judges Robust to Expressions of Uncertainty? Investigating the effect of Epistemic Markers on LLM-based Evaluation (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing efforts to train large language models to generate outputs containing epistemic markers have been largely overlooked. |
| Approach: | They propose a benchmark to assess the robustness of LLM-judges to epistemic markers. |
| Outcome: | EMBER benchmarks show that LLM-judges lack robustness in presence of epistemic markers . EMber QA (2,000 instances) and IF (2823 instances) are used to evaluate outputs containing epistemological markers. |
From Values to Opinions: Predicting Human Behaviors and Stances Using Value-Injected Large Language Models (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing large-scale surveys soliciting opinions on issues can be costly and laborious. |
| Approach: | They propose to use value-injected large language models to inject a target value distribution into large language model (LLM) and have them predict opinions and behaviors of people with similar values. |
| Outcome: | The proposed method significantly outperforms baseline methods on four tasks and the results suggest opinions and behaviors can be better predicted using value-injected LLMs. |
tRAG: Term-level Retrieval-Augmented Generation for Domain-Adaptive Retrieval (2025.naacl-long)
Copied to clipboard
| Challenge: | Neural retrieval models suffer when there is a domain shift between training and test data distributions. |
| Approach: | They propose to generate domain-adapted pseudo-queries using large language models (LLMs) to improve term recall of unseen query terms by using term-level Retrieval-Augmented Generation (tRAG). |
| Outcome: | The proposed method significantly improves recall for unseen terms by 10.6% and outperforms LLM and retrieval-augmented generation baselines on overall retrieval performance. |
Masked Summarization to Generate Factually Inconsistent Summaries for Improved Factual Consistency Checking (2022.findings-naacl)
Copied to clipboard
| Challenge: | Abstractive summarization systems generate paraphrases, but they often contain information inconsistent with the source text. |
| Approach: | They propose to generate factually inconsistent summaries using source texts and reference summary with key information masked to train a factual consistency classifier. |
| Outcome: | The proposed method outperforms existing models and shows a competitive correlation with human judgments. |
ReSCORE: Label-free Iterative Retriever Training for Multi-hop Question Answering with Relevance-Consistency Supervision (2025.acl-long)
Copied to clipboard
| Challenge: | Multi-hop question answering requires reasoning across multiple documents to answer complex questions. |
| Approach: | They propose a method for training dense retrievers for multi-hop question answering . they leverage large language models to measure document-question relevance with answer consistency . their results lead to state-of-the-art Exact Match and F1 scores for MHQA . |
| Outcome: | Evaluated on three MHQA benchmarks, the proposed method improves retrieval performance . it leads to state-of-the-art Exact Match and F1 scores for the proposed technique . |
A Corpus of eRulemaking User Comments for Measuring Evaluability of Arguments (L18-1)
Copied to clipboard
| Challenge: | eRulemaking is a way for government agencies to directly reach citizens to solicit their opinions and experiences regarding newly proposed rules. |
| Approach: | They propose an argument mining corpus annotated with argumentative structure information capturing the evaluability of arguments. |
| Outcome: | The proposed corpus contains 731 user comments on consumer debt collection practices rule by the Consumer Financial Protection Bureau. |