Papers by Sungroh Yoon
GrounDial: Human-norm Grounded Safe Dialog Response Generation (2024.findings-eacl)
Copied to clipboard
| Challenge: | Recent conversational AI systems generate unsafe responses agreeing to offensive user input or including toxic content. |
| Approach: | They propose a method where response safety is achieved by grounding responses to commonsense social rules without fine-tuning. |
| Outcome: | The proposed approach is quantitatively and qualitatively safer even without additional data or tuning. |
Interactive Text-to-Image Retrieval with Large Language Models: A Plug-and-Play Approach (2024.acl-long)
Copied to clipboard
| Challenge: | primarily addressed in text-to-image retrieval task using dialogue-form context query . conventionally, text-based retrieval methods rely on initial text descriptions . |
| Approach: | They propose a plug-based retrieval method that uses large language models as questioners to generate non-redundant questions about the attributes of the target image. |
| Outcome: | The proposed method performs better than zero-shot and fine-tuned baselines in benchmarks. |
Entity-level Factual Adaptiveness of Fine-tuning based Abstractive Summarization Models (2024.eacl-long)
Copied to clipboard
| Challenge: | Abstractive summarization models generate factually inconsistent content when parametric knowledge conflicts with knowledge in the input document. |
| Approach: | They propose a method to enhance factual adaptiveness while achieving factual consistency on original datasets. |
| Outcome: | The proposed method improves factual adaptiveness while achieving factual consistency on original datasets. |
Rare Tokens Degenerate All Tokens: Improving Neural Text Generation via Adaptive Gradient Gating for Rare Token Embeddings (2022.acl-long)
Copied to clipboard
| Challenge: | Recent studies have determined that the learned token embeddings of large-scale neural language models are degenerated to be anisotropic with a narrow-cone shape. |
| Approach: | They propose a method to degenerate the learning gradient for rare token embeddings by gating the specific part of the gradient for all tokens during training stage. |
| Outcome: | The proposed method improves the performance of the models but lacks the training dynamics needed to solve the representation degeneration problem. |
Exploring the Potential of LLMs as Personalized Assistants: Dataset, Evaluation, and Analysis (2025.acl-long)
Copied to clipboard
| Challenge: | Personalized AI assistants are a challenging application that intertwines multiple problems in LLM research. |
| Approach: | They propose a Llama-3.2-based automated evaluation model that matches human preferences to a conversational dataset. |
| Outcome: | HiCUPID provides a conversational dataset tailored for personalization . the evaluation model closely mirrors human preferences, the researchers show . |
EdiText: Controllable Coarse-to-Fine Text Editing with Diffusion Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for text editing have been proposed for various types of data with diverse attributes. |
| Approach: | They propose a novel text editing method that modifies reference text to desired attributes at various scales. |
| Outcome: | The proposed method is capable of making precise adjustments within the desired range while maintaining the accuracy of the reference text. |
Semantic Token Reweighting for Interpretable and Controllable Text Embeddings in CLIP (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Despite the varying significance of textual elements within a sentence depending on the context, efforts to account for variation of importance in constructing text embeddings have been lacking. |
| Approach: | They propose a framework for Semantic Token Reweighting to build Interpretable text embeddings which incorporates controllability as well. |
| Outcome: | The proposed framework improves the text encoding process in CLIP by differentially weighting semantic elements based on contextual importance, enabling finer control over emphasis responsive to data-driven insights and user preferences. |
Large-scale Lifelong Learning of In-context Instructions and How to Tackle It (2023.acl-long)
Copied to clipboard
| Challenge: | In-context instruction learning is a method to improve the target PLM’s instance- and task-level generalization performance as it observes more tasks. |
| Approach: | They propose to fine-tune a Pre-trained Language Model (PLM) on a set of tasks with in-context instructions and to extend this property to a scenario in which tasks are fed to the target PLM in a sequential manner. |
| Outcome: | The proposed method achieves noticeable improvements in both types of generalization, nearly reaching the upper bound performance obtained through joint training. |
Unleashing Multi-Hop Reasoning Potential in Large Language Models through Repetition of Misordered Context (2025.findings-naacl)
Copied to clipboard
| Challenge: | Multi-hop reasoning requires multi-step reasoning based on supporting documents within a given context. |
| Approach: | They propose a method that prompts the model by repeatedly presenting the context. |
| Outcome: | The proposed method improves the F1 score by 30%p on multi-hop QA tasks and increases accuracy by 70%p on a synthetic task. |
Does Your Voice Assistant Remember? Analyzing Conversational Context Recall and Utilization in Voice Interaction Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Recent advances in multi-turn voice interaction models have improved user-model communication, but whether open-source models share this ability remains unexplored. |
| Approach: | They propose to use ContextDialog to evaluate open-source interaction models' ability to recall past utterances to identify key limitations. |
| Outcome: | The proposed model retains and recalls past utterances better than closed-source models, but still struggles with questions about past . findings highlight key limitations in open-source model and suggest ways to improve memory retention and retrieval robustness. |
Controlled Text Generation for Black-box Language Models via Score-based Progressive Editor (2024.acl-long)
Copied to clipboard
| Challenge: | Existing methods to control text generation are inapplicable to black-box models or suffer a trade-off between control and fluency. |
| Approach: | They propose a new approach to control text generation that modifies context at the token level during the generation process of a backbone language model and guides subsequent text to naturally include the target attributes. |
| Outcome: | The proposed method can regulate the attributes of the generated text while utilizing the capability of the backbone large language models. |
Model Intrinsic Features of Fine-tuning based Text Summarization Models for Factual Consistency (2023.findings-acl)
Copied to clipboard
| Challenge: | a summarization model with relatively low factual consistency is more likely to model summaries that are not conditional to the documents. |
| Approach: | They analyze the model intrinsic features by varying the fine-tuning objectives and datasets. |
| Outcome: | The proposed models have a high inductive bias for factual consistency and are more likely to model summaries that are not conditional to the documents. |
AligNART: Non-autoregressive Neural Machine Translation by Jointly Learning to Estimate Alignment and Translate (2021.emnlp-main)
Copied to clipboard
| Challenge: | Non-autoregressive neural machine translation models suffer from the multi-modality problem . aligNART leverages full alignment information to explicitly reduce the modality of the target distribution . |
| Approach: | They propose an alignment decomposition method which explicitly reduces the modality of the target distribution. |
| Outcome: | The proposed model outperforms previous models that focus on modality reduction on two translation tasks. |
Interpretation of NLP models through input marginalization (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods to interpret NLP predictions replace each token with a predefined value, resulting in misleading interpretations. |
| Approach: | They propose to marginalize each token out of the training data distribution to demystify the "black box" property of deep neural networks for natural language processing. |
| Outcome: | The proposed method marginalizes each token out of the training data distribution. |
Correcting Negative Bias in Large Language Models through Negative Attention Score Alignment (2025.naacl-long)
Copied to clipboard
Sangwon Yu, Jongyoon Song, Bongkyu Hwang, Hoyoung Kang, Sooah Cho, Junhwa Choi, Seongho Joe, Taehee Lee, Youngjune Gwon, Sungroh Yoon
| Challenge: | Experimental results show that large language models exhibit a negative bias in binary decision tasks . hallucination is a factor that degrades reliability of LLMs . |
| Approach: | They propose a negative attention score to systematically and quantitatively formulate negative bias by using a parameter-efficient fine-tuning technique. |
| Outcome: | The proposed method reduces the gap between precision and recall caused by negative bias while preserving generalization abilities. |
Still Between Us? Evaluating and Improving Voice Assistant Robustness to Third-Party Interruptions (2026.acl-long)
Copied to clipboard
| Challenge: | Recent Spoken Language Models lack the capability to discern Third-Party Interruptions (TPI) from the primary user’s ongoing flow, leaving them vulnerable to contextual failures. |
| Approach: | They propose a dataset with speaker-aware hard negatives to enforce acoustic cue prioritization for interruption handling and a framework to measure the interruption-handling strategy and precise speaker discrimination in deceptive contexts. |
| Outcome: | The proposed framework mitigates semantic shortcut learning while neglecting acoustic signals essential for discerning speaker changes. |
Verbal-R3: Verbal Reranker as the Missing Bridge between Retrieval and Reasoning (2026.acl-long)
Copied to clipboard
| Challenge: | Existing paradigms of Retrieval-Augmented Generation (RAG) are suboptimal due to exposure bias, a mismatch between pre-training data distribution and retrieved information. |
| Approach: | They propose to bridge retrieval results and the LLM’s reasoning ability through Verbal Annotations, analytic narratives that explicitly articulate the logical connection between a search query and retrieved contexts. |
| Outcome: | The proposed framework achieves state-of-the-art performance on complex Question Answering benchmarks validating the effectiveness of the proposed framework. |