Taeuk Kim

Kang Min Yoo, Junyeob Kim, Hyuhng Joon Kim et al.

Despite recent explosion of interests in in-context learning, the underlying mechanism and the precise impact of the quality of demonstrations remain elusive. Intuitively, ground-truth labels should have as much impact in in-context learning (ICL) as supervised learning, but recent work reported that the input-label correspondence is significantly less important than previously thought. Intrigued by this counter-intuitive observation, we re-examine the importance of ground-truth labels in in-context learning. With the introduction of two novel metrics, namely Label-Correctness Sensitivity and Ground-truth Label Effect Ratio (GLER), we were able to conduct quantifiable analysis on the impact of ground-truth label demonstrations. Through extensive analyses, we find that the correct input-label mappings can have varying impacts on the downstream in-context learning performances, depending on the experimental configuration. Through additional studies, we identify key components, such as the verbosity of prompt templates and the language model size, as the controlling factor to achieve more noise-resilient ICL.

Hyuhng Joon Kim, Hyunsoo Cho, Junyeob Kim et al.

Large-scale pre-trained language models (PLMs) are well-known for being capable of solving a task simply by conditioning a few input-label pairs dubbed demonstrations on a prompt without being explicitly tuned for the desired downstream task. Such a process (i.e., in-context learning), however, naturally leads to high reliance on the demonstrations which are usually selected from external datasets. In this paper, we propose self-generated in-context learning (SG-ICL), which generates demonstrations for in-context learning from PLM itself to minimize the reliance on the external demonstration. We conduct experiments on four different text classification tasks and show SG-ICL significantly outperforms zero-shot learning and is generally worth approximately 0.6 gold training samples. Moreover, our generated demonstrations show more consistent performance with low variance compared to randomly selected demonstrations from the training dataset.

Hyunsoo Cho, Hyuhng Joon Kim, Junyeob Kim et al.

Through in-context learning (ICL), large-scale language models are effective few-shot learners without additional model fine-tuning. However, the ICL performance does not scale well with the number of available training samples as it is limited by the inherent input length constraint of the underlying language model. Meanwhile, many studies have revealed that language models are also powerful feature extractors, allowing them to be utilized in a black-box manner and enabling the linear probing paradigm, where lightweight discriminators are trained on top of the pre-extracted input representations. This paper proposes prompt-augmented linear probing (PALP), a hybrid of linear probing and ICL, which leverages the best of both worlds. PALP inherits the scalability of linear probing and the capability of enforcing language models to derive more meaningful representations via tailoring input into a more conceivable form. Throughout in-depth investigations on various datasets, we verified that PALP significantly enhances the input representations closing the gap between ICL in the data-hungry scenario and fine-tuning in the data-abundant scenario with little training overhead, potentially making PALP a strong alternative in a black-box scenario.

Youna Kim, Hyuhng Joon Kim, Cheonbok Park et al.

When using large language models (LLMs) in knowledge-intensive tasks, such as open-domain question answering, external context can bridge the gap between external knowledge and the LLMs' parametric knowledge. Recent research has been developed to amplify contextual knowledge over the parametric knowledge of LLMs with contrastive decoding approaches. While these approaches could yield truthful responses when relevant context is provided, they are prone to vulnerabilities when faced with noisy contexts. We extend the scope of previous studies to encompass noisy contexts and propose adaptive contrastive decoding (ACD) to leverage contextual influence effectively. ACD demonstrates improvements in open-domain question answering tasks compared to baselines, especially in robustness by remaining undistracted by noisy contexts in retrieval-augmented generation.

Youmin Ko, Hyemin Yang, Taeuk Kim et al.

Fine-tuning pre-trained language models (PLMs) has recently shown a potential to improve knowledge graph completion (KGC). However, most PLM-based methods focus solely on encoding textual information, neglecting the long-tailed nature of knowledge graphs and their various topological structures, e.g., subgraphs, shortest paths, and degrees. We claim that this is a major obstacle to achieving higher accuracy of PLMs for KGC. To this end, we propose a Subgraph-Aware Training framework for KGC (SATKGC) with two ideas: (i) subgraph-aware mini-batching to encourage hard negative sampling and to mitigate an imbalance in the frequency of entity occurrences during training, and (ii) new contrastive learning to focus more on harder in-batch negative triples and harder positive triples in terms of the structural properties of the knowledge graph. To the best of our knowledge, this is the first study to comprehensively incorporate the structural inductive bias of the knowledge graph into fine-tuning PLMs. Extensive experiments on three KGC benchmarks demonstrate the superiority of SATKGC. Our code is available.

Hyunsoo Cho, Choonghyun Park, Jaewook Kang et al.

Out-of-distribution (OOD) detection aims to discern outliers from the intended data distribution, which is crucial to maintaining high reliability and a good user experience. Most recent studies in OOD detection utilize the information from a single representation that resides in the penultimate layer to determine whether the input is anomalous or not. Although such a method is straightforward, the potential of diverse information in the intermediate layers is overlooked. In this paper, we propose a novel framework based on contrastive learning that encourages intermediate features to learn layer-specialized representations and assembles them implicitly into a single representation to absorb rich information in the pre-trained language model. Extensive experiments in various intent classification and OOD datasets demonstrate that our approach is significantly more effective than other works.

Hyuhng Joon Kim, Hyunsoo Cho, Sang-Woo Lee et al.

When deploying machine learning systems to the wild, it is highly desirable for them to effectively leverage prior knowledge to the unfamiliar domain while also firing alarms to anomalous inputs. In order to address these requirements, Universal Domain Adaptation (UniDA) has emerged as a novel research area in computer vision, focusing on achieving both adaptation ability and robustness (i.e., the ability to detect out-of-distribution samples). While UniDA has led significant progress in computer vision, its application on language input still needs to be explored despite its feasibility. In this paper, we propose a comprehensive benchmark for natural language that offers thorough viewpoints of the model's generalizability and robustness. Our benchmark encompasses multiple datasets with varying difficulty levels and characteristics, including temporal shifts and diverse domains. On top of our testbed, we validate existing UniDA methods from computer vision and state-of-the-art domain adaptation techniques from NLP literature, yielding valuable findings: We observe that UniDA methods originally designed for image input can be effectively transferred to the natural language domain while also underscoring the effect of adaptation difficulty in determining the model's performance.

Hyuhng Joon Kim, Youna Kim, Cheonbok Park et al.

In interactions between users and language model agents, user utterances frequently exhibit ellipsis (omission of words or phrases) or imprecision (lack of exactness) to prioritize efficiency. This can lead to varying interpretations of the same input based on different assumptions or background knowledge. It is thus crucial for agents to adeptly handle the inherent ambiguity in queries to ensure reliability. However, even state-of-the-art large language models (LLMs) still face challenges in such scenarios, primarily due to the following hurdles: (1) LLMs are not explicitly trained to deal with ambiguous utterances; (2) the degree of ambiguity perceived by the LLMs may vary depending on the possessed knowledge. To address these issues, we propose Alignment with Perceived Ambiguity (APA), a novel pipeline that aligns LLMs to manage ambiguous queries by leveraging their own assessment of ambiguity (i.e., perceived ambiguity). Experimental results on question-answering datasets demonstrate that APA empowers LLMs to explicitly detect and manage ambiguous queries while retaining the ability to answer clear questions. Furthermore, our finding proves that APA excels beyond training with gold-standard labels, especially in out-of-distribution scenarios. The data and code are available at https://github.com/heyjoonkim/APA.

Taeuk Kim

Constituency Parse Extraction from Pre-trained Language Models (CPE-PLM) is a recent paradigm that attempts to induce constituency parse trees relying only on the internal knowledge of pre-trained language models. While attractive in the perspective that similar to in-context learning, it does not require task-specific fine-tuning, the practical effectiveness of such an approach still remains unclear, except that it can function as a probe for investigating language models' inner workings. In this work, we mathematically reformulate CPE-PLM and propose two advanced ensemble methods tailored for it, demonstrating that the new parsing paradigm can be competitive with common unsupervised parsers by introducing a set of heterogeneous PLMs combined using our techniques. Furthermore, we explore some scenarios where the trees generated by CPE-PLM are practically useful. Specifically, we show that CPE-PLM is more effective than typical supervised parsers in few-shot settings.

Youngju Joung, Taeuk Kim

We propose a unified framework that enables us to consider various aspects of contextualization at different levels to better identify the idiomaticity of multi-word expressions. Through extensive experiments, we demonstrate that our approach based on the inter- and inner-sentence context of a target MWE is effective in improving the performance of related models. We also share our experience in detail on the task of SemEval-2022 Tasks 2 such that future work on the same task can be benefited from this.

Yejin Yoon, Jungyeon Lee, Kangsan Kim et al.

Task-oriented dialogue (TOD) systems are commonly designed with the presumption that each utterance represents a single intent. However, this assumption may not accurately reflect real-world situations, where users frequently express multiple intents within a single utterance. While there is an emerging interest in multi-intent detection (MID), existing in-domain datasets such as MixATIS and MixSNIPS have limitations in their formulation. To address these issues, we present BlendX, a suite of refined datasets featuring more diverse patterns than their predecessors, elevating both its complexity and diversity. For dataset construction, we utilize both rule-based heuristics as well as a generative tool -- OpenAI's ChatGPT -- which is augmented with a similarity-driven strategy for utterance selection. To ensure the quality of the proposed datasets, we also introduce three novel metrics that assess the statistical properties of an utterance related to word count, conjunction use, and pronoun usage. Extensive experiments on BlendX reveal that state-of-the-art MID models struggle with the challenges posed by the new datasets, highlighting the need to reexamine the current state of the MID field. The dataset is available at https://github.com/HYU-NLP/BlendX.

Seunghee Kim, Bumkyu Park, Kyudan Jung et al.

Most testbeds for omni-modal models assess multimodal understanding via textual outputs, leaving it unclear whether these models can properly speak their answers. To study this, we introduce OmniACBench, a benchmark for evaluating context-grounded acoustic control in omni-modal models. Given a spoken instruction, a text script, and an image, a model must read the script aloud with an appropriate tone and manner. OmniACBench comprises 3,559 verified instances covering six acoustic features: speech rate, phonation, pronunciation, emotion, global accent, and timbre. Extensive experiments on eight models reveal their limitations in the proposed setting, despite their strong performance on prior textual-output evaluations. Our analyses show that the main bottleneck lies not in processing individual modalities, but in integrating multimodal context for faithful speech generation. Moreover, we identify three common failure modes-weak direct control, failed implicit inference, and failed multimodal grounding-providing insights for developing models that can verbalize responses effectively.

Taejun Yun, Jinhyeon Kim, Deokyeong Kang et al.

Cross-lingual transfer (XLT) is an emergent ability of multilingual language models that preserves their performance on a task to a significant extent when evaluated in languages that were not included in the fine-tuning process. While English, due to its widespread usage, is typically regarded as the primary language for model adaption in various tasks, recent studies have revealed that the efficacy of XLT can be amplified by selecting the most appropriate source languages based on specific conditions. In this work, we propose the utilization of sub-network similarity between two languages as a proxy for predicting the compatibility of the languages in the context of XLT. Our approach is model-oriented, better reflecting the inner workings of foundation models. In addition, it requires only a moderate amount of raw text from candidate languages, distinguishing it from the majority of previous methods that rely on external resources. In experiments, we demonstrate that our method is more effective than baselines across diverse tasks. Specifically, it shows proficiency in ranking candidates for zero-shot XLT, achieving an improvement of 4.6% on average in terms of NDCG@3. We also provide extensive analyses that confirm the utility of sub-networks for XLT prediction.

Juyeon Kim, Geon Lee, Dongwon Choi et al.

Retrieval over visually rich documents is essential for tasks such as legal discovery, scientific search, and enterprise knowledge management. Existing approaches fall into two paradigms: single-vector retrieval, which is efficient but coarse, and multi-vector retrieval, which is accurate but computationally expensive. To address this trade-off, we propose HEAVEN, a two-stage hybrid-vector framework. In the first stage, HEAVEN efficiently retrieves candidate pages using a single-vector method over Visually-Summarized Pages (VS-Pages), which assemble representative visual layouts from multiple pages. In the second stage, it reranks candidates with a multi-vector method while filtering query tokens by linguistic importance to reduce redundant computations. To evaluate retrieval systems under realistic conditions, we also introduce ViMDOC, the first benchmark for visually rich, multi-document, and long-document retrieval. Across four benchmarks, HEAVEN attains 99.87% of the Recall@1 performance of multi-vector models on average while reducing per-query computation by 99.82%, achieving efficiency and accuracy. Our code and datasets are available at: https://github.com/juyeonnn/HEAVEN

Jungyeon Lee, Kangmin Lee, Taeuk Kim

Knowledge conflict often arises in retrieval-augmented generation (RAG) systems, where retrieved documents may be inconsistent with one another or contradict the model's parametric knowledge. Existing benchmarks for investigating the phenomenon have notable limitations, including a narrow focus on the question answering setup, heavy reliance on entity substitution techniques, and a restricted range of conflict types. To address these issues, we propose a knowledge graph (KG)-based framework that generates varied and subtle conflicts between two similar yet distinct contexts, while ensuring interpretability through the explicit relational structure of KGs. Experimental results on our benchmark, MAGIC, provide intriguing insights into the inner workings of LLMs regarding knowledge conflict: both open-source and proprietary models struggle with conflict detection -- especially when multi-hop reasoning is required -- and often fail to pinpoint the exact source of contradictions. Finally, we present in-depth analyses that serve as a foundation for improving LLMs in integrating diverse, sometimes even conflicting, information.

Juyeon Kim, Geon Lee, Taeuk Kim et al.

Entity linking (EL) aligns textual mentions with their corresponding entities in a knowledge base, facilitating various applications such as semantic search and question answering. Recent advances in multimodal entity linking (MEL) have shown that combining text and images can reduce ambiguity and improve alignment accuracy. However, most existing MEL methods overlook the rich structural information available in the form of knowledge-graph (KG) triples. In this paper, we propose KGMEL, a novel framework that leverages KG triples to enhance MEL. Specifically, it operates in three stages: (1) Generation: Produces high-quality triples for each mention by employing vision-language models based on its text and images. (2) Retrieval: Learns joint mention-entity representations, via contrastive learning, that integrate text, images, and (generated or KG) triples to retrieve candidate entities for each mention. (3) Reranking: Refines the KG triples of the candidate entities and employs large language models to identify the best-matching entity for the mention. Extensive experiments on benchmark datasets demonstrate that KGMEL outperforms existing methods. Our code and datasets are available at: https://github.com/juyeonnn/KGMEL.

Choonghyun Park, Hyuhng Joon Kim, Junyeob Kim et al.

AI Generated Text (AIGT) detectors are developed with texts from humans and LLMs of common tasks. Despite the diversity of plausible prompt choices, these datasets are generally constructed with a limited number of prompts. The lack of prompt variation can introduce prompt-specific shortcut features that exist in data collected with the chosen prompt, but do not generalize to others. In this paper, we analyze the impact of such shortcuts in AIGT detection. We propose Feedback-based Adversarial Instruction List Optimization (FAILOpt), an attack that searches for instructions deceptive to AIGT detectors exploiting prompt-specific shortcuts. FAILOpt effectively drops the detection performance of the target detector, comparable to other attacks based on adversarial in-context examples. We also utilize our method to enhance the robustness of the detector by mitigating the shortcuts. Based on the findings, we further train the classifier with the dataset augmented by FAILOpt prompt. The augmented classifier exhibits improvements across generation models, tasks, and attacks. Our code will be available at https://github.com/zxcvvxcz/FAILOpt.

Yejin Yoon, Minseo Kim, Taeuk Kim

Users often omit essential details in their requests to LLM-based agents, resulting in under-specified inputs for tool use. This poses a fundamental challenge for tool-augmented agents, as API execution typically requires complete arguments, highlighting the need for personalized tool calling. To study this problem, we introduce MPT, a benchmark comprising 265 multi-session dialogues that cover three challenges: Preference Recall, Preference Induction, and Preference Transfer. We also propose PRefine, a test-time memory-augmented method that represents user preferences as evolving hypotheses. Through a generate--verify--refine loop, it extracts reusable constraints from history and improves tool-calling accuracy while using only 1.24% of the tokens required by full-history prompting. These results indicate that robust personalization in agentic systems depends on memory that captures the reasons behind user choices, not just the choices themselves.

Deokyeong Kang, Ki Jung Seo, Taeuk Kim

Modular programming, which aims to construct the final program by integrating smaller, independent building blocks, has been regarded as a desirable practice in software development. However, with the rise of recent code generation agents built upon large language models (LLMs), a question emerges: is this traditional practice equally effective for these new tools? In this work, we assess the impact of modularity in code generation by introducing a novel metric for its quantitative measurement. Surprisingly, unlike conventional wisdom on the topic, we find that modularity is not a core factor for improving the performance of code generation models. We also explore potential explanations for why LLMs do not exhibit a preference for modular code compared to non-modular code.

Yejin Yoon, Yuri Son, Namyoung So et al.

Conversational agents have traditionally been developed for either task-oriented dialogue (TOD) or open-ended chitchat, with limited progress in unifying the two. Yet, real-world conversations naturally involve fluid transitions between these modes. To address this gap, we introduce TACT (TOD-And-Chitchat Transition), a dataset designed for transition-aware dialogue modeling that incorporates structurally diverse and integrated mode flows. TACT supports both user- and agent-driven mode switches, enabling robust modeling of complex conversational dynamics. To evaluate an agent's ability to initiate and recover from mode transitions, we propose two new metrics -- Switch and Recovery. Models trained on TACT outperform baselines in both intent detection and mode transition handling. Moreover, applying Direct Preference Optimization (DPO) to TACT-trained models yields additional gains, achieving 75.74\% joint mode-intent accuracy and a 70.1\% win rate against GPT-4o in human evaluation. These results demonstrate that pairing structurally diverse data with DPO enhances response quality and transition control, paving the way for more proactive and transition-aware conversational agents.

Namyoung So, Seokgyu Jang, Taeuk Kim

Adaptive multi-agent systems (MAS) are increasingly adopted to tackle complex problems.However, the narrow task coverage of their optimization raises the question of whether they can function as general-purpose systems.To address this gap, we conduct an extensive empirical study of adaptive MAS, revealing two key findings: (1) topological overfitting -- they fail to generalize across different domains; and (2) illusory coordination -- they achieve reasonable surface-level accuracy while the underlying agent interactions diverge from ideal MAS behavior, raising concerns about their practical utility.These findings highlight the pressing need to prioritize generalization in MAS development and motivate evaluation protocols that extend beyond simple final-answer correctness.

Seunghee Kim, Changhyeon Kim, Taeuk Kim

Real-world decision-making often requires integrating and reasoning over information from multiple modalities. While recent multimodal large language models (MLLMs) have shown promise in such tasks, their ability to perform multi-hop reasoning across diverse sources remains insufficiently evaluated. Existing benchmarks, such as MMQA, face challenges due to (1) data contamination and (2) a lack of complex queries that necessitate operations across more than two modalities, hindering accurate performance assessment. To address this, we present Financial Cross-Modal Multi-Hop Reasoning (FCMR), a benchmark created to analyze the reasoning capabilities of MLLMs by urging them to combine information from textual reports, tables, and charts within the financial domain. FCMR is categorized into three difficulty levels-Easy, Medium, and Hard-facilitating a step-by-step evaluation. In particular, problems at the Hard level require precise cross-modal three-hop reasoning and are designed to prevent the disregard of any modality. Experiments on this new benchmark reveal that even state-of-the-art MLLMs struggle, with the best-performing model (Claude 3.5 Sonnet) achieving only 30.4% accuracy on the most challenging tier. We also conduct analysis to provide insights into the inner workings of the models, including the discovery of a critical bottleneck in the information retrieval phase.

Seong Hoon Lim, Taejun Yun, Jinhyeon Kim et al.

The successful adaptation of multilingual language models (LMs) to a specific language-task pair critically depends on the availability of data tailored for that condition. While cross-lingual transfer (XLT) methods have contributed to addressing this data scarcity problem, there still exists ongoing debate about the mechanisms behind their effectiveness. In this work, we focus on one of promising assumptions about inner workings of XLT, that it encourages multilingual LMs to place greater emphasis on language-agnostic or task-specific features. We test this hypothesis by examining how the patterns of XLT change with a varying number of source languages involved in the process. Our experimental findings show that the use of multiple source languages in XLT-a technique we term Multi-Source Language Training (MSLT)-leads to increased mingling of embedding spaces for different languages, supporting the claim that XLT benefits from making use of language-independent information. On the other hand, we discover that using an arbitrary combination of source languages does not always guarantee better performance. We suggest simple heuristics for identifying effective language combinations for MSLT and empirically prove its effectiveness.

Hyuhng Joon Kim, Youna Kim, Sang-goo Lee et al.

Large Language Models (LLMs) demonstrate exceptional performance across diverse tasks by leveraging pre-trained (i.e., parametric) and external (i.e., contextual) knowledge. While substantial efforts have been made to enhance the utilization of both forms of knowledge, situations in which models lack relevant information remain underexplored. To investigate this challenge, we first present a controlled testbed featuring four distinct knowledge access scenarios, including the aforementioned edge case, revealing that conventional LLM usage exhibits insufficient robustness in handling all instances. Addressing this limitation, we propose Contrastive Decoding with Abstention (CDA), a novel training-free decoding method that allows LLMs to generate responses when relevant knowledge is available and to abstain otherwise. CDA estimates the relevance of both knowledge sources for a given input, adaptively deciding which type of information to prioritize and which to exclude. Through extensive experiments, we demonstrate that CDA can effectively perform accurate generation and abstention simultaneously, enhancing reliability and preserving user trust.

Hwiyeong Lee, Uiji Hwang, Hyelim Lim et al.

Large language models often retain unintended content, prompting growing interest in knowledge unlearning. Recent approaches emphasize localized unlearning, restricting parameter updates to specific regions in an effort to remove target knowledge while preserving unrelated general knowledge. However, their effectiveness remains uncertain due to the lack of robust and thorough evaluation of the trade-off between the competing goals of unlearning. In this paper, we begin by revisiting existing localized unlearning approaches. We then conduct controlled experiments to rigorously evaluate whether local parameter updates causally contribute to unlearning. Our findings reveal that the set of parameters that must be modified for effective unlearning is not strictly determined, challenging the core assumption of localized unlearning that parameter locality is inherently indicative of effective knowledge removal.

Jisu Kim, Youngwoo Shin, Uiji Hwang et al.

Idioms have long posed a challenge due to their unique linguistic properties, which set them apart from other common expressions. While recent studies have leveraged large language models (LLMs) to handle idioms across various tasks, e.g., idiom-containing sentence generation and idiomatic machine translation, little is known about the underlying mechanisms of idiom processing in LLMs, particularly in multilingual settings. To this end, we introduce MIDAS, a new large-scale dataset of idioms in six languages, each paired with its corresponding meaning. Leveraging this resource, we conduct a comprehensive evaluation of LLMs' idiom processing ability, identifying key factors that influence their performance. Our findings suggest that LLMs rely not only on memorization, but also adopt a hybrid approach that integrates contextual cues and reasoning, especially when processing compositional idioms. This implies that idiom understanding in LLMs emerges from an interplay between internal knowledge retrieval and reasoning-based inference.

Ki Jung Seo, Sehun Lim, Taeuk Kim

Recent progress in large language models (LLMs) has enabled them to express their confidence in natural language, enhancing transparency and reliability. However, their confidence often exhibits overconfidence, the cause of which remains poorly understood. In this work, we conduct a detailed analysis of the dynamics underlying verbalized confidence and identify answer-independence as a key factor, defined as the model's failure to condition confidence on its own answer. To address this, we propose ADVICE (Answer-Dependent Verbalized Confidence Estimation), a fine-tuning framework that facilitates answer-grounded confidence estimation. Extensive experiments show that ADVICE substantially improves confidence calibration while preserving task performance. Further analyses confirm that ADVICE strengthens answer-groundedness, leading to more balanced and well-calibrated confidence distributions. Our findings shed light on the origin of overconfidence and establish a framework for more trustworthy confidence verbalization.

Seunghee Kim, Ingyu Bang, Seokgyu Jang et al.

Cross-modal multi-hop reasoning (CMR) is a valuable yet underexplored capability of multimodal large language models (MLLMs), entailing the integration of information from multiple modalities to produce a coherent output for a given context. We argue that existing benchmarks for evaluating this ability have critical shortcomings: (1) they largely overlook the speech modality, and (2) they exhibit heavily biased reasoning path distributions, which can severely undermine fair evaluation. To address these limitations, we introduce a novel benchmark -- Cross-Modal Multi-Hop Reasoning over Text, Image and Speech with Path Balance (CMR-SPB) -- designed to assess tri-modal multi-hop reasoning while ensuring both unbiased and diverse reasoning paths. Our experiments with the new dataset reveal consistent model failures in specific reasoning sequences and show that biased benchmarks risk misrepresenting model performance. Finally, based on our extensive analysis, we propose a new ECV (Extract, Connect, Verify) prompting technique that effectively mitigates the performance gap across different reasoning paths. Overall, we call for more careful evaluation in CMR to advance the development of robust multimodal AI.

Minhae Oh, Jeonghye Kim, Nakyung Lee et al.

Scientific reasoning requires not only long-chain reasoning processes, but also knowledge of domain-specific terminologies and adaptation to updated findings. To deal with these challenges for scientific reasoning, we introduce RAISE, a step-by-step retrieval-augmented framework which retrieves logically relevant documents from in-the-wild corpus. RAISE is divided into three steps: problem decomposition, logical query generation, and logical retrieval. We observe that RAISE consistently outperforms other baselines on scientific reasoning benchmarks. We analyze that unlike other baselines, RAISE retrieves documents that are not only similar in terms of the domain knowledge, but also documents logically more relevant.

Youna Kim, Hyuhng Joon Kim, Minjoon Choi et al.

Language models often benefit from external knowledge beyond parametric knowledge. While this combination enhances performance, achieving reliable knowledge utilization remains challenging, as it requires assessing the state of each knowledge source based on the presence of relevant information. Yet, prior work on knowledge integration often overlooks this challenge by assuming ideal conditions and provides limited coverage of knowledge scenarios. To address this gap, we introduce UniKnow, a Unified framework for reliable LM behavior across parametric and external Knowledge. UniKnow enables controlled evaluation across knowledge scenarios such as knowledge conflict, distraction, and absence conditions that are rarely addressed together. Beyond evaluating existing methods under this setting, we extend our work by introducing UniKnow-Aware methods to support comprehensive evaluation. Experiments on UniKnow reveal that existing methods struggle to generalize across a broader range of knowledge configurations and exhibit scenario-specific biases. UniKnow thus provides a foundation for systematically exploring and improving reliability under knowledge scenarios.

Young Hyun Yoo, Jii Cha, Changhyeon Kim et al.

While the introduction of contrastive learning frameworks in sentence representation learning has significantly contributed to advancements in the field, it still remains unclear whether state-of-the-art sentence embeddings can capture the fine-grained semantics of sentences, particularly when conditioned on specific perspectives. In this paper, we introduce Hyper-CL, an efficient methodology that integrates hypernetworks with contrastive learning to compute conditioned sentence representations. In our proposed approach, the hypernetwork is responsible for transforming pre-computed condition embeddings into corresponding projection layers. This enables the same sentence embeddings to be projected differently according to various conditions. Evaluation on two representative conditioning benchmarks, namely conditional semantic text similarity and knowledge graph completion, demonstrates that Hyper-CL is effective in flexibly conditioning sentence representations, showcasing its computational efficiency at the same time. We also provide a comprehensive analysis of the inner workings of our approach, leading to a better interpretation of its mechanisms.

Taeuk Kim, Kang Min Yoo, Sang-goo Lee

Although BERT and its variants have reshaped the NLP landscape, it still remains unclear how best to derive sentence embeddings from such pre-trained Transformers. In this work, we propose a contrastive learning method that utilizes self-guidance for improving the quality of BERT sentence representations. Our method fine-tunes BERT in a self-supervised fashion, does not rely on data augmentation, and enables the usual [CLS] token embeddings to function as sentence vectors. Moreover, we redesign the contrastive learning objective (NT-Xent) and apply it to sentence representation learning. We demonstrate with extensive experiments that our approach is more effective than competitive baselines on diverse sentence-related tasks. We also show it is efficient at inference and robust to domain shifts.

Bowen Li, Taeuk Kim, Reinald Kim Amplayo et al.

Transformer-based pre-trained language models (PLMs) have dramatically improved the state of the art in NLP across many tasks. This has led to substantial interest in analyzing the syntactic knowledge PLMs learn. Previous approaches to this question have been limited, mostly using test suites or probes. Here, we propose a novel fully unsupervised parsing approach that extracts constituency trees from PLM attention heads. We rank transformer attention heads based on their inherent properties, and create an ensemble of high-ranking heads to produce the final tree. Our method is adaptable to low-resource languages, as it does not rely on development sets, which can be expensive to annotate. Our experiments show that the proposed method often outperform existing approaches if there is no development set present. Our unsupervised parser can also be used as a tool to analyze the grammars PLMs learn implicitly. For this, we use the parse trees induced by our method to train a neural PCFG and compare it to a grammar derived from a human-annotated treebank.

Jaeyoul Shin, Taeuk Kim, Sang-goo Lee

We propose a novel method that enables us to determine words that deserve to be emphasized from written text in visual media, relying only on the information from the self-attention distributions of pre-trained language models (PLMs). With extensive experiments and analyses, we show that 1) our zero-shot approach is superior to a reasonable baseline that adopts TF-IDF and that 2) there exist several attention heads in PLMs specialized for emphasis selection, confirming that PLMs are capable of recognizing important words in sentences.

Taeuk Kim, Bowen Li, Sang-goo Lee

As it has been unveiled that pre-trained language models (PLMs) are to some extent capable of recognizing syntactic concepts in natural language, much effort has been made to develop a method for extracting complete (binary) parses from PLMs without training separate parsers. We improve upon this paradigm by proposing a novel chart-based method and an effective top-K ensemble technique. Moreover, we demonstrate that we can broaden the scope of application of the approach into multilingual settings. Specifically, we show that by applying our method on multilingual PLMs, it becomes possible to induce non-trivial parses for sentences from nine languages in an integrated and language-agnostic manner, attaining performance superior or comparable to that of unsupervised PCFGs. We also verify that our approach is robust to cross-lingual transfer. Finally, we provide analyses on the inner workings of our method. For instance, we discover universal attention heads which are consistently sensitive to syntactic information irrespective of the input language.

Taeuk Kim, Jihun Choi, Daniel Edmiston et al.

With the recent success and popularity of pre-trained language models (LMs) in natural language processing, there has been a rise in efforts to understand their inner workings. In line with such interest, we propose a novel method that assists us in investigating the extent to which pre-trained LMs capture the syntactic notion of constituency. Our method provides an effective way of extracting constituency trees from the pre-trained LMs without training. In addition, we report intriguing findings in the induced trees, including the fact that pre-trained LMs outperform other approaches in correctly demarcating adverb phrases in sentences.

Sanghwan Bae, Taeuk Kim, Jihoon Kim et al.

As an attempt to combine extractive and abstractive summarization, Sentence Rewriting models adopt the strategy of extracting salient sentences from a document first and then paraphrasing the selected ones to generate a summary. However, the existing models in this framework mostly rely on sentence-level rewards or suboptimal labels, causing a mismatch between a training objective and evaluation metric. In this paper, we present a novel training signal that directly maximizes summary-level ROUGE scores through reinforcement learning. In addition, we incorporate BERT into our model, making good use of its ability on natural language understanding. In extensive experiments, we show that a combination of our proposed model and training procedure obtains new state-of-the-art performance on both CNN/Daily Mail and New York Times datasets. We also demonstrate that it generalizes better on DUC-2002 test set.

Kang Min Yoo, Taeuk Kim, Sang-goo Lee

We propose a simple yet effective approach for improving Korean word representations using additional linguistic annotation (i.e. Hanja). We employ cross-lingual transfer learning in training word representations by leveraging the fact that Hanja is closely related to Chinese. We evaluate the intrinsic quality of representations learned through our approach using the word analogy and similarity tests. In addition, we demonstrate their effectiveness on several downstream tasks, including a novel Korean news headline generation task.

Jihun Choi, Taeuk Kim, Sang-goo Lee

We present a latent variable model for predicting the relationship between a pair of text sequences. Unlike previous auto-encoding--based approaches that consider each sequence separately, our proposed framework utilizes both sequences within a single model by generating a sequence that has a given relationship with a source sequence. We further extend the cross-sentence generating framework to facilitate semi-supervised training. We also define novel semantic constraints that lead the decoder network to generate semantically plausible and diverse sequences. We demonstrate the effectiveness of the proposed model from quantitative and qualitative experiments, while achieving state-of-the-art results on semi-supervised natural language inference and paraphrase identification.

Taeuk Kim, Jihun Choi, Daniel Edmiston et al.

Most existing recursive neural network (RvNN) architectures utilize only the structure of parse trees, ignoring syntactic tags which are provided as by-products of parsing. We present a novel RvNN architecture that can provide dynamic compositionality by considering comprehensive syntactic information derived from both the structure and linguistic tags. Specifically, we introduce a structure-aware tag representation constructed by a separate tag-level tree-LSTM. With this, we can control the composition function of the existing word-level tree-LSTM by augmenting the representation as a supplementary input to the gate functions of the tree-LSTM. In extensive experiments, we show that models built upon the proposed architecture obtain superior or competitive performance on several sentence-level tasks such as sentiment analysis and natural language inference when compared against previous tree-structured models and other sophisticated neural models.

Jihun Choi, Taeuk Kim, Sang-goo Lee

We propose a method of stacking multiple long short-term memory (LSTM) layers for modeling sentences. In contrast to the conventional stacked LSTMs where only hidden states are fed as input to the next layer, the suggested architecture accepts both hidden and memory cell states of the preceding layer and fuses information from the left and the lower context using the soft gating mechanism of LSTMs. Thus the architecture modulates the amount of information to be delivered not only in horizontal recurrence but also in vertical connections, from which useful features extracted from lower layers are effectively conveyed to upper layers. We dub this architecture Cell-aware Stacked LSTM (CAS-LSTM) and show from experiments that our models bring significant performance gain over the standard LSTMs on benchmark datasets for natural language inference, paraphrase detection, sentiment classification, and machine translation. We also conduct extensive qualitative analysis to understand the internal behavior of the suggested approach.

Taeuk Kim, Jihun Choi, Sang-goo Lee

We present a novel neural architecture for the Argument Reasoning Comprehension task of SemEval 2018. It is a simple neural network consisting of three parts, collectively judging whether the logic built on a set of given sentences (a claim, reason, and warrant) is plausible or not. The model utilizes contextualized word vectors pre-trained on large machine translation (MT) datasets as a form of transfer learning, which can help to mitigate the lack of training data. Quantitative analysis shows that simply leveraging LSTMs trained on MT datasets outperforms several baselines and non-transferred models, achieving accuracies of about 70% on the development set and about 60% on the test set.

Sanghyuk Choi, Taeuk Kim, Jinseok Seol et al.

Word embedding has become a fundamental component to many NLP tasks such as named entity recognition and machine translation. However, popular models that learn such embeddings are unaware of the morphology of words, so it is not directly applicable to highly agglutinative languages such as Korean. We propose a syllable-based learning model for Korean using a convolutional neural network, in which word representation is composed of trained syllable vectors. Our model successfully produces morphologically meaningful representation of Korean words compared to the original Skip-gram embeddings. The results also show that it is quite robust to the Out-of-Vocabulary problem.