COLING 2025 Papers

Large Language Models (LLMs) have demonstrated remarkable capabilities in comprehending and analyzing lengthy sequential inputs, owing to their extensive context windows that allow processing millions of tokens in a single forward pass. However, this paper uncovers a surprising limitation: LLMs fall short when handling long input sequences. We investigate this issue using three datasets and two tasks (sentiment analysis and news categorization) across various LLMs, including Claude 3, Gemini Pro, GPT 3.5 Turbo, Llama 3 Instruct, and Mistral Instruct models. To address this limitation, we propose and evaluate ad-hoc solutions that substantially enhance LLMs’ performance on long input sequences by up to 50%, while reducing API cost and latency by up to 93% and 50%, respectively.

Large language models (LLMs) have received lots of attention for their impressive performance in in-context dialogues and their potential to revolutionize service industries with a new business model, Model-as-a-Service (MaaS). Automated data labeling is a natural and promising service. However, labeling data with LLMs faces two main challenges: 1) the labels from LLMs may contain uncertainty, and 2) using LLMs for data labeling tasks can be prohibitively expensive, as the scales of datasets are usually tremendous. In this paper, we propose a hierarchical framework named LMCrowd that leverages multiple LLMs for efficient data labeling under budget constraints. The proposed LMCrowd framework first aggregates labels from multiple freely available LLMs, and then employs a large, paid MaaS LLM for relabeling selected instances. Furthermore, we formalize the core process as an optimization problem, aiming to select the optimal set of instances for relabeling by the MaaS LLM, given the current belief state. Extensive experimental evaluations across various real-world datasets demonstrate that our framework outperforms human labelers and GPT-4 in terms of both accuracy and efficiency.

With the development of multimedia technology, online social media has become a major medium for people to access news, but meanwhile, it has also exacerbated the dissemination of multi-modal fake news. An automatic and efficient multi-modal fake news detection (MFND) method is urgently needed. Existing MFND methods usually conduct cross-modal information interaction at later stage, resulting in insufficient exploration of complementary information between modalities. Another challenge lies in the differences among news data from different domains, leading to the weak generalization ability in detecting news from various domains. In this work, we propose an efficient Cross-modal Prompt Learning with Semantic enhancement method for Domain-robust fake news detection (CPLSD). Specifically, we design an efficient cross-modal prompt interaction module, which utilizes prompt as medium to realize lightweight cross-modal information interaction in the early stage of feature extraction, enabling to exploit rich modality complementary information. We design a domain-general prompt generation module that can adaptively blend domain-specific news features to generate domain-general prompts, for improving the domain generalization ability of the model. Furthermore, an image semantic enhancement module is designed to achieve image-to-text translation, fully exploring the semantic discriminative information of the image modality. Extensive experiments conducted on three MFND benchmarks demonstrate the superiority of our proposed approach over existing state-of-the-art MFND methods.

Vision-Language Models (VLMs) have recently experienced significant advancements. However, challenges persist in the accurate recognition of fine details within high resolution images, which limits performance in multiple tasks. This work introduces Pheye, a novel architecture that efficiently processes high-resolution images while training fewer parameters than similarly sized VLMs. Notably, Pheye achieves a high efficiency while maintaining strong performance, particularly in tasks that demand fine-grained image understanding and/or the handling of scene-text.

While large language models (LLMs) have shown remarkable capabilities in natural language processing, they struggle with complex, multi-step reasoning tasks involving knowledge graphs (KGs). Existing approaches that integrate LLMs and KGs either underutilize the reasoning abilities of LLMs or suffer from prohibitive computational costs due to tight coupling. To address these limitations, we propose a novel collaborative framework named EffiQA that can strike a balance between performance and efficiency via an iterative paradigm. EffiQA consists of three stages: global planning, efficient KG exploration, and self-reflection. Specifically, EffiQA leverages the commonsense capability of LLMs to explore potential reasoning pathways through global planning. Then, it offloads semantic pruning to a small plug-in model for efficient KG exploration. Finally, the exploration results are fed to LLMs for self-reflection to further improve global planning and efficient KG exploration. Empirical evidence on multiple KBQA benchmarks shows EffiQA’s effectiveness, achieving an optimal balance between reasoning accuracy and computational costs. We hope the proposed new framework will pave the way for efficient, knowledge-intensive querying by redefining the integration of LLMs and KGs, fostering future research on knowledge-based question answering.

While large language models (LLMs) have achieved remarkable success in various natural language processing tasks, their strengths have yet to be fully demonstrated in grammatical error correction (GEC). This is partly due to the misalignment between their pre-training objectives and the GEC principle of making minimal edits. In this work, we aim to bridge this gap by introducing a novel method called Edit-wise Preference Optimization (EPO). By distinguishing the importance of different tokens and assigning higher reward weights to edit tokens during preference optimization, our method captures fine-grained distinctions in GEC that traditional preference learning often overlooks. Extensive experiments on both English and Chinese datasets show that our framework consistently outperforms strong baselines, achieving state-of-the-art performance and demonstrating the advantages of LLMs in GEC.

Distilling high-accuracy Graph Neural Networks (GNNs) to low-latency multilayer perceptrons (MLPs) on graph tasks has become a hot research topic. However, conventional MLP learning relies almost exclusively on graph nodes and fails to effectively capture the graph structural information. Previous methods address this issue by processing graph edges into extra inputs for MLPs, but such graph structures may be unavailable for various scenarios. To this end, we propose Prototype-Guided Knowledge Distillation (PGKD), which does not require graph edges (edge-free setting) yet learns structure-aware MLPs. Our insight is to distill graph structural information from GNNs. Specifically, we first employ the class prototypes to analyze the impact of graph structures on GNN teachers, and then design two losses to distill such information from GNNs to MLPs. Experimental results on popular graph benchmarks demonstrate the effectiveness and robustness of the proposed PGKD.

Recently, there has been a growing trend of employing large language models (LLMs) to judge the quality of other LLMs. Many studies have adopted closed-source models, mainly using GPT-4 as the evaluator. However, due to the closed-source nature of the GPT-4 model, employing it as an evaluator has resulted in issues including transparency, controllability, and cost-effectiveness. Some researchers have turned to using fine-tuned open-source LLMs as evaluators. However, existing open-source evaluation LLMs generally lack a user-friendly visualization tool, and they have not been optimized for accelerated model inference, which causes inconvenience for researchers with limited resources and those working across different fields. This paper presents EasyJudge, a model developed to evaluate significant language model responses. It is lightweight, precise, efficient, and user-friendly, featuring an intuitive visualization interface for ease of deployment and use. EasyJudge uses detailed datasets and refined prompts for model optimization, achieving strong consistency with human and proprietary model evaluations. The model optimized with quantitative methods enables EasyJudge to run efficiently on consumer-grade GPUs or even CPUs.

Research on Large Language Models (LLMs) has recently witnessed an increasing interest in extending the models’ context size to better capture dependencies within long documents. While benchmarks have been proposed to assess long-range abilities, existing efforts primarily considered generic tasks that are not necessarily aligned with real-world applications. In contrast, we propose a new benchmark for long-context LLMs focused on a practical meeting assistant scenario in which the long contexts consist of transcripts obtained by automatic speech recognition, presenting unique challenges for LLMs due to the inherent noisiness and oral nature of such data. Our benchmark, ELITR-Bench, augments the existing ELITR corpus by adding 271 manually crafted questions with their ground-truth answers, as well as noisy versions of meeting transcripts altered to target different Word Error Rate levels. Our experiments with 12 long-context LLMs on ELITR-Bench confirm the progress made across successive generations of both proprietary and open models, and point out their discrepancies in terms of robustness to transcript noise. We also provide a thorough analysis of our GPT-4-based evaluation, including insights from a crowdsourcing study. Our findings indicate that while GPT-4’s scores align with human judges, its ability to distinguish beyond three score levels may be limited.

We propose ELAINE (EngLish-jApanese-chINesE)-medLLM, a trilingual (English, Japanese, Chinese) large language model adapted for the bio-medical domain based on Llama-3-8B. The training dataset was carefully curated in terms of volume and diversity to adapt to the biomedical domain and endow trilingual capability while preserving the knowledge and abilities of the base model. The training follows 2-stage paths: continued pre-training and supervised fine-tuning (SFT). Our results demonstrate that ELAINE-medLLM exhibits superior trilingual capabilities compared to existing bilingual or multilingual medical LLMs without severely sacrificing the base model’s capability.

Personality is a fundamental construct in psychology, reflecting an individual’s behavior, thinking, and emotional patterns. While previous researches have made progress in personality detection, their designed methods generally overlook the important connection between psychological knowledge “emotion regulation” and personality traits. Based on this, we propose a new personality detection method called EERPD. This method introduces the use of emotion regulation, a psychological concept highly correlated with personality, for personality prediction. By combining this concept with emotion features, EERPD retrieves few-shot examples and provides process CoTs for inferring labels from text. This approach enhances the understanding of LLM for personality implicit within text and improves the performance in personality detection. Experimental results demonstrate that EERPD significantly enhances the accuracy and robustness of personality detection, outperforming previous SOTA by 15.05/4.29 in average F1 on the two benchmark datasets.

Individuals facing financial difficulties often make decisions driven by emotions rather than rational analysis. EDAR, a pipeline for Emotion and Dialogue Act Recognition, is designed specifically for the debt collection process in France. By integrating EDAR into decision-making systems, debt collection outcomes could be improved. The pipeline employs Machine Learning and Deep Learning models, demonstrating that smaller models with fewer parameters can achieve high performance, offering an efficient alternative to large language models.

Empathy improves human-machine dialogue systems by enhancing the user’s experience. While traditional models have aimed to detect and express users’ emotions from dialogue history, they neglect the crucial and complex interactions among emotion, emotion causes, and commonsense. To address this, we introduce the ECC (Emotion, Cause, and Commonsense) framework, which leverages specialized encoders to capture the key features of emotion, cause, and commonsense and collaboratively models these through a Conditional Variational Auto-Encoder. ECC further employs novel loss functions to refine the interplay of three factors and generates empathetic responses using an energy-based model supported by ODE sampling. Empirical results on the EmpatheticDialogues dataset demonstrate that ECC outperforms existing baselines, offering a robust solution for empathetic dialogue generation.

Modern large language models are sensitive to prompts, and another synonymous expression or a typo may lead to unexpected results for the model. Composing an optimal prompt for a specific demand lacks theoretical support and relies entirely on human experimentation, which poses a considerable obstacle to popularizing generative artificial intelligence. However, there is no systematic analysis of the stability of large language models to resist prompt perturbations. In this work, we propose to evaluate the ease-of-use of large language models and construct E-Bench, simulating the actual situation of human use from synonymous perturbation (including paraphrasing, simplification, and colloquialism) and typographical perturbation. Besides we also discuss the combination of these two types of perturbation and analyze the main reasons for performance degradation. Experimental results indicate that with the increase of model size, although the ease-of-use could be significantly improved, there is still a long way to go to build a sufficiently user-friendly model.

Instruction tuning is a burgeoning method to elicit the general intelligence of Large Language Models (LLMs). While numerous studies have examined the impact of factors such as data volume and model size on English models, the scaling properties of instruction tuning in other languages remain largely unexplored. In this work, we systematically investigate the effects of data quantity, model size, and data construction methods on instruction tuning for Chinese LLMs. We utilize a newly curated dataset, DoIT, which includes over 40,000 high-quality instruction instances covering ten underlying abilities, such as creative writing, code generation, and logical reasoning. Our experiments, conducted on models ranging from 7b to 33b parameters, yield three key findings: (i) While these factors directly affect overall model performance, some abilities are more responsive to scaling, whereas others demonstrate significant resistance. (ii) The scaling sensitivity of different abilities to these factors can be explained by two features: Complexity and Transference. (iii) By tailoring training strategies to their varying sensitivities, specific abilities can be efficiently learned, enhancing performance on two public benchmarks.

Continual Few-shot Relation Extraction (CFRE) aims to continually learn new relations from limited labeled data while preserving knowledge about previously learned relations. Facing the inherent issue of catastrophic forgetting, previous approaches predominantly rely on memory replay strategies. However, they often overlook task interference in continual learning and the varying memory requirements for different relations. To address these shortcomings, we propose a novel framework, DPC-FT, which features: 1) a lightweight relation encoder for each task to mitigate negative knowledge transfer across tasks; 2) a dynamic prototype module to allocate less memory for easier relations and more memory for harder relations. Additionally, we introduce the None-Of-The-Above (NOTA) detection in CFRE and propose a threshold criterion to identify relations that have never been learned. Extensive experiments demonstrate the effectiveness and efficiency of our method in CFRE, making our approach more practical and comprehensive for real-world scenarios.

Multimodal emotion recognition in conversation (MERC) refers to identifying and classifying human emotional states by combining data from multiple different modalities (e.g., audio, images, text, video, etc.). Specifically, human emotional expressions are often complex and diverse, and these complex emotional expressions can be captured and understood more comprehensively through the fusion of multimodal information. Most existing graph-based multimodal emotion recognition methods can only use shallow GCNs to extract emotion features and fail to capture the temporal dependencies caused by dynamic changes in emotions. To address the above problems, we propose a Dynamic Graph Neural Ordinary Differential Equation Network (DGODE) for multimodal emotion recognition in conversation, which combines the dynamic changes of emotions to capture the temporal dependency of speakers’ emotions. Technically, the key idea of DGODE is to use the graph ODE evolution network to characterize the continuous dynamics of node representations over time and capture temporal dependencies. Extensive experiments on two publicly available multimodal emotion recognition datasets demonstrate that the proposed DGODE model has superior performance compared to various baselines. Furthermore, the proposed DGODE can also alleviate the over-smoothing problem, thereby enabling the construction of a deep GCN network.

This paper presents DynRank, a novel framework for enhancing passage retrieval in open-domain question-answering systems through dynamic zero-shot question classification. Traditional approaches rely on static prompts and pre-defined templates, which may limit model adaptability across different questions and contexts. In contrast, DynRank introduces a dynamic prompting mechanism, leveraging a pre-trained question classification model that categorizes questions into fine-grained types. Based on these classifications, contextually relevant prompts are generated, enabling more effective passage retrieval. We integrate DynRank into existing retrieval frameworks and conduct extensive experiments on multiple QA benchmark datasets.

Euphemisms are a linguistic device used to soften discussions of sensitive or uncomfortable topics, with death being a prominent example. In this paper, we present a study on the detection of death-related euphemisms in historical literary texts from a corpus containing Danish and Norwegian novels from the late 19th century. We introduce an annotated dataset of euphemistic and literal references to death, including both common and rare euphemisms, ranging from well-established terms to more culturally nuanced expressions. We evaluate the performances of state-of-the-art pre-trained language models fine-tuned for euphemism detection. Our findings show that fixed, literal expressions of death became less frequent over time, while metaphorical euphemisms grew in prevalence. Additionally, euphemistic language was more common in historical novels, whereas contemporary novels tended to refer to death more literally, reflecting the rise of secularism. These results shed light on the shifting discourse on death during a period when the concept of death as final became prominent.