Papers by Kang-il Lee
Target-Agnostic Gender-Aware Contrastive Learning for Mitigating Bias in Multilingual Machine Translation (2023.emnlp-main)
Copied to clipboard
| Challenge: | Gender bias is a significant issue in machine translation, but most studies focus on debiasing bilingual models without consideration for multilingual systems. |
| Approach: | They propose a method which debiases bilingual models for unambiguous cases where there is a single correct translation. |
| Outcome: | The proposed method improves gender accuracy by a wide margin without hampering translation performance. |
Weakly Supervised Semantic Parsing with Execution-based Spurious Program Filtering (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods to train a semantic parser from weak supervision focus on exploiting similarities between examples based on domain-specific knowledge. |
| Approach: | They propose a domain-agnostic filtering mechanism based on program execution results to identify and filter out programs with significantly different semantics from the other programs. |
| Outcome: | The proposed method improves the performance of existing weakly-supervised parsers by incorporating a majority vote on the program search results. |
Drift: Decoding-time Personalized Alignments with Implicit User Preferences (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Drift personalizes large language models at decoding time with implicit user preferences . Unlike traditional Reinforcement Learning from Human Feedback, Drift operates in a training-free manner . |
| Approach: | They propose a framework that personalizes large language models at decoding time with implicit user preferences. |
| Outcome: | The proposed framework personalizes large language models at decoding time with implicit user preferences. |
VLind-Bench: Measuring Language Priors in Large Vision-Language Models (2025.findings-naacl)
Copied to clipboard
| Challenge: | Large Vision-Language Models suffer from a problem known as language prior . such language priors can lead to undesirable biases and hallucinations when dealing with images that are out of distribution. |
| Approach: | They propose a benchmark to measure the language priors of Large Vision-Language Models. |
| Outcome: | The proposed benchmark is the first specifically designed to measure the language priors, or blindness, of LVLMs. |
Mitigating Hallucinations in Large Vision-Language Models via Summary-Guided Decoding (2025.findings-naacl)
Copied to clipboard
| Challenge: | Large Vision-Language Models (LVLMs) generate detailed and coherent responses from visual inputs but are prone to generate hallucinations due to an over-reliance on language priors. |
| Approach: | They propose a method that reduces the text context and controls only the image-related POS tokens to maintain text quality by reducing the text contextualization. |
| Outcome: | The proposed method achieves state-of-the-art performance on object hallucination benchmarks and achieves Pareto optimality among the existing methods. |
Generating Diverse Hypotheses for Inductive Reasoning (2025.naacl-long)
Copied to clipboard
| Challenge: | Recent studies suggest that large language models (LLMs) can engage in inductive reasoning by sampling multiple hypotheses about the rules and selecting the one that best explains the observations. |
| Approach: | They propose to increase the temperature parameter to enhance diversity by sampling multiple hypotheses and selecting the one that best explains the observations. |
| Outcome: | The proposed method improves diversity while maintaining text quality while increasing temperature. |
IterCQR: Iterative Conversational Query Reformulation with Retrieval Guidance (2024.naacl-long)
Copied to clipboard
| Challenge: | Existing methods for conversational query reformulation depend on human annotations. |
| Approach: | They propose a method that reformulates context-dependent conversational queries without relying on human rewrites. |
| Outcome: | The proposed method shows state-of-the-art performance on two widely-used datasets. |