Despite the fact that large language models (LLMs) show exceptional skill in instruction following tasks, this strength can turn into a vulnerability when the models are required to disregard certain instructions.Instruction following tasks typically involve a clear task description and input text containing the target data to be processed.However, when the input itself resembles an instruction, confusion may arise, even if there is explicit prompting to distinguish between the task instruction and the input.We refer to this phenomenon as instructional distraction.In this paper, we introduce a novel benchmark, named DIM-Bench, specifically designed to assess LLMs' performance under instructional distraction.The benchmark categorizes real-world instances of instructional distraction and evaluates LLMs across four instruction tasks: rewriting, proofreading, translation, and style transfer-alongside five input tasks: reasoning, code generation, mathematical reasoning, bias detection, and question answering.Our experimental results reveal that even the most advanced LLMs are susceptible to instructional distraction, often failing to accurately follow user intent in such cases. Instruction Input ExampleRewrite Reasoning Instruction: Paraphrase the following text.Input: Laundry detergents were once manufactured to contain high ... which would a lake become as a result of the phosphorous in the detergent?Options : A. canyon B. desert C. swamp D. river

Recently, large vision–language models (LVLMs) have emerged as the preferred tools for judging text–image alignment, yet their robustness along the visual modality remains underexplored. This work is the first study to address a key research question: Can adversarial visual manipulations systematically fool LVLM judges into assigning unfairly inflated scores? We define potential image-induced biases within the context of T2I evaluation and examine how these biases affect the evaluations of LVLM judges. Moreover, we introduce a novel, fine-grained, multi-domain meta-evaluation benchmark named FRAME, which is deliberately constructed to exhibit diverse score distributions. By introducing the defined biases into the benchmark, we reveal that all tested LVLM judges exhibit vulnerability across all domains, consistently inflating scores for manipulated images. Further analysis reveals that combining multiple biases amplifies their effects, and pairwise evaluations are similarly susceptible. Moreover, we observe that visual biases persist despite prompt-based mitigation strategies, highlighting the vulnerability of current LVLM evaluation systems and underscoring the urgent need for more robust LVLM judges.

Dongryeol Lee, Yerin Hwang, Yongil Kim, Joonsuk Park, Kyomin Jung. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Existing English-based text similarity measurements primarily focus on the semantic dimension, neglecting the unique linguistic attributes found in languages like Korean, where honorific expressions are explicitly integrated. To address this limitation, this study proposes Kosmic, a novel Korean text-similarity metric that encompasses the semantic and tonal facets of a given text pair. For the evaluation, we introduce a novel benchmark annotated by human experts, empirically showing that Kosmic outperforms the existing method. Moreover, by leveraging Kosmic, we assess various Korean paraphrasing methods to determine which techniques are most effective in preserving semantics and tone.

Despite advancements in on-topic dialogue systems, effectively managing topic shifts within dialogues remains a persistent challenge, largely attributed to the limited availability of training datasets.To address this issue, we propose Multi-Passage to Dialogue (MP2D), a data generation framework that automatically creates conversational questionanswering datasets with natural topic transitions.By leveraging the relationships between entities in a knowledge graph, MP2D maps the flow of topics within a dialogue, effectively mirroring the dynamics of human conversation.It retrieves relevant passages corresponding to the topics and transforms them into dialogues through the passage-to-dialogue method.Through quantitative and qualitative experiments, we demonstrate MP2D's efficacy in generating dialogue with natural topic shifts.Furthermore, this study introduces a novel benchmark for topic shift dialogues, TS-WikiDialog.Utilizing the dataset, we demonstrate that even Large Language Models (LLMs) struggle to handle topic shifts in dialogue effectively, and we showcase the performance improvements of models trained on datasets generated by MP2D across diverse topic shift dialogue tasks.

To address the data scarcity issue in Conversational question answering (ConvQA), a dialog inpainting method, which utilizes documents to generate ConvQA datasets, has been proposed.However, the original dialog inpainting model is trained solely on the dialog reconstruction task, resulting in the generation of questions with low contextual relevance due to insufficient learning of question-answer alignment.To overcome this limitation, we propose a novel framework called Dialogizer, which has the capability to automatically generate ConvQA datasets with high contextual relevance from textual sources.The framework incorporates two training tasks: question-answer matching (QAM) and topic-aware dialog generation (TDG).Moreover, re-ranking is conducted during the inference phase based on the contextual relevance of the generated questions.Using our framework, we produce four Con-vQA datasets by utilizing documents from multiple domains as the primary source.Through automatic evaluation using diverse metrics, as well as human evaluation, we validate that our proposed framework exhibits the ability to generate datasets of higher quality compared to the baseline dialog inpainting model.