Recent studies on hallucination in large language models (LLMs) have been actively progressing in natural language processing. However, the impact of negated text on hallucination with LLMs remains largely unexplored. In this paper, we set three important yet unanswered research questions and aim to address them. To derive the answers, we investigate whether LLMs can recognize contextual shifts caused by negation and still reliably distinguish hallucinations comparable to affirmative cases. We also design the NegHalu dataset by reconstructing existing hallucination detection datasets with negated expressions. Our experiments demonstrate that LLMs struggle to detect hallucinations in negated text effectively, often producing logically inconsistent or unfaithful judgments. Moreover, we trace the internal state of LLMs as they process negated inputs at the token level and reveal the challenges of mitigating their unintended effects.

The Impact of Negated Text on Hallucination with Large Language Models

Retrieval-augmented generation (RAG)-based question answering (QA) has emerged as a powerful method for processing long industrial documents. However, conventional text chunking approaches often neglect complex and long industrial document structures, causing information loss and reduced answer quality. To address this, we introduce \textbf{MultiDocFusion}, a multimodal chunking pipeline that integrates: (i) detection of document regions using vision-based document parsing, (ii) text extraction from these regions via OCR, (iii) reconstruction of document structure into a hierarchical tree using large language model (LLM)-based document section hierarchical parsing (DSHP-LLM), and (iv) construction of hierarchical chunks through DFS-based grouping. Extensive experiments across industrial benchmarks demonstrate that \textbf{MultiDocFusion} improves retrieval precision by 8–15% and ANLS QA scores by 2–3% compared to baselines, emphasizing the critical role of explicitly leveraging document hierarchy for multimodal document-based QA. These significant performance gains underscore the necessity of structure-aware chunking in enhancing the fidelity of RAG-based QA systems.

MultiDocFusion : Hierarchical and Multimodal Chunking Pipeline for Enhanced RAG on Long Industrial Documents

Korean legal knowledge is subject to frequent temporal updates driven by societal needs and government policies. Even minor modifications to legal provisions can have significant consequences, yet continuously retraining large language models (LLMs) to incorporate such updates is resource-intensive and impractical. To address this, we propose KoLEG, an on-the-fly Korean Legal knowledge editing framework enhanced with continuous retrieval. KoLEG employs an Editing-Aware Learning Strategy and a LawEdit Retriever, which together adaptively integrate subtle linguistic nuances and continuous legislative amendments. To support this task, we construct the Korean Legislative Amendment Dataset, explicitly designed for continuous legal knowledge updates with attention to both temporal dynamics and linguistic subtleties. KoLEG outperforms existing locate-then-edit and retrieval-based editing methods, demonstrating superior effectiveness in legal knowledge editing while preserving linguistic capabilities. Furthermore, KoLEG maintains robust performance in sequential editing, improves performance on precedent application tasks, and is qualitatively validated by legal experts.

KoLEG: On-the-Fly Korean Legal Knowledge Editing with Continuous Retrieval

Recent frontier-level LLMs have saturated many previously difficult benchmarks, leaving little room for further differentiation. This progress highlights the need for challenging benchmarks that provide objective verification. In this paper, we introduce MCBench, a benchmark designed to evaluate whether LLMs can execute string-matching NLP metrics by strictly following step-by-step instructions. Unlike prior benchmarks that depend on subjective judgments or general reasoning, MCBench offers an objective, deterministic and code-verifiable evaluation. This setup allows us to systematically test whether LLMs can maintain accurate step-by-step execution, including instruction adherence, numerical computation, and long-range consistency in handling intermediate results. To ensure objective evaluation of these abilities, we provide a parallel reference code that can evaluate the accuracy of LLM output. We provide three evaluative metrics and three benchmark variants designed to measure the detailed instruction understanding capability of LLMs. Our analyses show that MCBench serves as an effective and objective tool for evaluating the capabilities of cutting-edge LLMs

Metric Calculating Benchmark: Code-Verifiable Complicate Instruction Following Benchmark for Large Language Models

Instruction tuning(IT) is an effective approach for aligning large language models (LLMs) with human intentions. There is ongoing discourse regarding the data quality for IT. As an effort to find the robust criteria of data quality for IT, we introduce LimaCost, a data quality measure that exhibits a strong correlation with model performance. LimaCost utilizes LIMA dataset, which effectiveness in IT has already been validated by several previous works. LimaCost then estimates the value of a given data by estimating how many LIMA data points might be needed to approximate its gradient. Our experiments reveal that LimaCost enables effective data selection that derive high alignment performance. We demonstrate that selecting data based on high LimaCost proves to be more effective than existing data selection strategies.

LimaCost: Data Valuation for Instruction Tuning of Large Language Models

Instruction tuning is crucial for adapting large language models (LLMs) to align with user intentions. Numerous studies emphasize the significance of the quality of instruction tuning (IT) data, revealing a strong correlation between IT data quality and the alignment performance of LLMs. In these studies, the quality of IT data is typically assessed by evaluating the performance of LLMs trained with that data. However, we identified a prevalent issue in such practice: hyperparameters for training models are often selected arbitrarily without adequate justification. We observed significant variations in hyperparameters applied across different studies, even when training the same model with the same data. In this study, we demonstrate the potential problems arising from this practice and emphasize the need for careful consideration in verifying data quality. Through our experiments on the quality of LIMA data and a selected set of 1,000 Alpaca data points, we demonstrate that arbitrary hyperparameter decisions can make any arbitrary conclusion.

Call for Rigor in Reporting Quality of Instruction Tuning Data

Through numerous endeavors, large language models (LLMs) have witnessed significant advancements in their instruction-following capability. However, we discern that LLMs are prone to generate responses to instruction-formatted statements in an instinctive manner, rather than comprehending the underlying user intention reside within the given instructions. We also recognize that the significance of instruction understanding capability is largely overlooked in most of LLM evaluation benchmarks. To ensure more comprehensive evaluation on the instruction understanding capability of LLM, we propose Intention of Instruction (IntInst) benchmark, which primary objective is to distinguish the appropriate instruction that accurately instruct to generate a given context. IntInst presents four instruction candidates and requires LLMs to select one among them. Through extensive experiments with several instruction-tuned LLMs, we reveal that most LLMs struggle to grasp the actual intention concealed in the instruction and thoroughly analyze the factors influencing instruction understanding.

Find the Intention of Instruction: Comprehensive Evaluation of Instruction Understanding for Large Language Models

Large language models (LLMs) often retain outdated or incorrect information from pre-training, which undermines their reliability. While model editing methods have been developed to address such errors without full re-training, they frequently suffer from knowledge conflicts, where outdated information interferes with new knowledge. In this work, we propose Conflict-free Model Editing (CoME), a novel framework that enhances the accuracy of knowledge updates in LLMs by selectively removing outdated knowledge. CoME leverages unlearning to mitigate knowledge interference, allowing new information to be integrated without compromising relevant linguistic features. Through experiments on GPT-J and LLaMA-3 using Counterfact and ZsRE datasets, we demonstrate that CoME improves both editing accuracy and model reliability when applied to existing editing methods. Our results highlight that the targeted removal of outdated knowledge is crucial for enhancing model editing effectiveness and maintaining the model's generative performance.

CoME: An Unlearning-based Approach to Conflict-free Model Editing

The evolution of large language models (LLMs) has culminated in a multitask model paradigm where prompts drive the generation of user-specific outputs. However, this advancement has revealed a critical challenge: LLMs frequently produce outputs against socially acceptable commonsense standards in various scenarios. To address this gap in commonsense reasoning, we present KoCommonGEN v2, a fine-grained benchmark dataset focused on Korean commonsense reasoning. This dataset, enriched with human annotations, comprises multiple-choice questions across seven error categories. These categories include commonsense memorization, numerical commonsense, toxic speech, and more, which are vulnerable to undermining the reliability of LLMs' commonsense reasoning capabilities. The empirical results present that LLMs struggle with Korean commonsense reasoning. With human accuracy benchmarked at approximately 85%, GPT-4's performance lags at about 74%, and other LLMs demonstrate an average accuracy of around 42%. Our findings emphasize the need for targeted improvements in Korean commonsense reasoning within LLMs, paving the way for more socially and contextually sensitive AI models.

KoCommonGEN v2: A Benchmark for Navigating Korean Commonsense Reasoning Challenges in Large Language Models

Byte Pair Encoding is an effective approach in machine translation across several languages. However, our analysis indicates that BPE is prone to over-segmentation in the morphologically rich language, Korean, which can erode word semantics and lead to semantic confusion during training. This semantic confusion, stemming from over-segmentation, ultimately contributes to a degradation of overall translation quality. To address this issue, we introduce Length-aware Subword Vocabulary Construction (LeVoC), a novel approach strategically incorporating longer words into the vocabulary. By utilizing an external monolingual Korean corpus, LeVoC extracts and integrates long words, effectively preserving morphological information and reducing semantic confusion. Our experiments demonstrate that LeVoC not only significantly outperforms BPE, but also can be applied to and surpass current state-of-the-art morpheme-aware subword tokenization methods. We provide evidence that the difficulty in translating sentences with long words in Korean is associated with morphological compositionality, and LeVoC's ability to reduce semantic confusion during training leads to improved translation quality.

Length-aware Byte Pair Encoding for Mitigating Over-segmentation in Korean Machine Translation

The recent advancements in the realm of Automatic Speech Recognition (ASR) post-processing have been primarily driven by sequence-to-sequence paradigms. Despite their effectiveness, these methods often demand substantial amounts of data, necessitating the expensive recruitment of phonetic transcription experts to rectify the erroneous outputs of ASR systems, thereby creating the desired training data. Back TranScription (BTS) alleviates this issue by generating ASR inputs from clean text via a Text-to-Speech (TTS) system. While initial studies on BTS exhibited promise, they were constrained by a limited dataset of just 200,000 sentence pairs, leaving the scalability of this method in question. In this study, we delve into the potential scalability of BTS. We introduce the "Hyper-BTS" dataset, a corpus approximately five times larger than that utilized in prior research. Additionally, we present innovative criteria for categorizing error types within ASR post-processing. This not only facilitates a more comprehensive qualitative analysis, which was absent in preceding studies, but also enhances the understanding of ASR error patterns. Our empirical results, both quantitative and qualitative, suggest that the enlarged scale of the Hyper-BTS dataset sufficiently addresses a vast majority of the ASR error categories. We make the Hyper-BTS dataset publicly available.

Hyper-BTS Dataset: Scalability and Enhanced Analysis of Back TranScription (BTS) for ASR Post-Processing

Educational question-answer generation has been extensively researched owing to its practical applicability. However, we have identified a persistent challenge concerning the evaluation of such systems. Existing evaluation methods often fail to produce objective results and instead exhibit a bias towards favoring high similarity to the ground-truth question-answer pairs. In this study, we demonstrate that these evaluation methods yield low human alignment and propose an alternative approach called Generative Interpretation (GI) to achieve more objective evaluations. Through experimental analysis, we reveal that GI outperforms existing evaluation methods in terms of human alignment, and even shows comparable performance with GPT3.5, only with BART-large.

Generative Interpretation: Toward Human-Like Evaluation for Educational Question-Answer Pair Generation

Large language models (LLMs) have demonstrated impressive capabilities in natural language processing. However, their ability to establish causal relationships, particularly in the context of temporal interventions and language hallucinations, remains challenging. This paper presents \textbf{CReTIHC}, a novel dataset designed to test and enhance the causal reasoning abilities of LLMs. The dataset is constructed using a unique approach that incorporates elements of verbal hallucinations and temporal interventions through the reengineering of existing causal inference datasets. This transformation creates complex scenarios that push LLMs to critically evaluate the information presented and identify cause-and-effect relationships. The CReTIHC dataset serves as a pioneering tool for improving LLM's causal inference capabilities, paving the way for a more nuanced understanding of causal relationships in natural language processing (NLP) tasks. The whole dataset is publicly accessible at: (https://github.com/ChangwooChun/CReTIHC)

CReTIHC: Designing Causal Reasoning Tasks about Temporal Interventions and Hallucinated Confoundings

Korean morphological variations present unique opportunities and challenges in natural language processing (NLP), necessitating an advanced understanding of morpheme-based sentence construction. The complexity of morphological variations allows for diverse sentence forms based on the syntactic-semantic integration of functional morphemes (i.e., affixes) to lexical morphemes (i.e., roots). With this in mind, we propose a method - CHEF, replicating the morphological transformations inherent in sentences based on lexical and functional morpheme combinations through generative data augmentation. CHEF operates using a morpheme blender and a label discriminator, thereby enhancing the diversity of Korean sentence forms by capturing the properties of agglutination while maintaining label consistency. We conduct experiments on Korean multiple classification datasets, improving model performance in full- and few-shot settings. Our proposed method boosts performance beyond the preceding data augmentation methods without incurring external data usage. We demonstrate that our approach achieves comparable results yielded by augmentation techniques that use large language models (LLMs).

CHEF in the Language Kitchen: A Generative Data Augmentation Leveraging Korean Morpheme Ingredients

Automatic Speech Recognition (ASR) systems are instrumental across various applications, with their performance being critically tied to user satisfaction. Conventional evaluation metrics for ASR systems produce a singular aggregate score, which is insufficient for understanding specific system vulnerabilities. Therefore, we aim to address the limitations of the previous ASR evaluation methods by introducing the Korean Error Explainable Benchmark Dataset for ASR and Post-processing (KEBAP). KEBAP enables comprehensive analysis of ASR systems at both speech- and text levels, thereby facilitating a more balanced assessment encompassing speech recognition accuracy and user readability. KEBAP provides 37 newly defined speech-level resources incorporating diverse noise environments and speaker characteristics categories, also presenting 13 distinct text-level error types. This paper demonstrates detailed statistical analyses of colloquial noise categories and textual error types. Furthermore, we conduct extensive validation and analysis on commercially deployed ASR systems, providing valuable insights into their performance. As a more fine-grained and real-world-centric evaluation method, KEBAP contributes to identifying and mitigating potential weaknesses in ASR systems.

KEBAP: Korean Error Explainable Benchmark Dataset for ASR and Post-processing | VIDEO

Critical error detection (CED) aims to identify the presence of catastrophic meaning distortion in machine translation. Fatal errors require significant attention because of their potential to cause personal or societal harm. The CED for Korean, an agglutinative language, is particularly highlighted, as minor variations in morphemes often bring substantial shifts in semantic interpretation. However, research on Korean is still underexplored and has room for improvement. In this study, we conduct the first investigation of CED for English-Korean to the best of our knowledge. We adopt prompt-based fine-tuning and propose various informative evidence to incorporate into the input prompt. Subsequently, we perform comprehensive verification and analysis to identify the most helpful guidance for detecting critical errors. The experimental results show that prompt-based fine-tuning with informative evidence outperforms standard fine-tuning by a large margin, demonstrating its remarkable effectiveness in English-Korean CED.

Jaehyung Seo

17

5

SHORT BIO

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES