COLING 2025 论文列表

With the rapid growth of social network services, misinformation has spread uncontrollably. Most recent approaches to fake news detection use neural network models to predict whether the input text is fake or real. Some of them even provide explanations, in addition to veracity, generated by Large Language Models (LLMs). However, they do not utilize factual evidence, nor do they allude to it or provide evidence/justification, thereby making their predictions less credible. This paper proposes a new fake news detection method that predicts the truth or false-hood of a claim based on relevant factual evidence (if exists) or LLM’s inference mechanisms (such as common-sense reasoning) otherwise. Our method produces the final synthesized prediction, along with well-founded facts or reasoning. Experimental results on several large COVID-19 fake news datasets show that our method achieves state-of-the-art (SOTA) detection and evidence explanation performance. Our source codes are available online.

This study examines the tendency to cite older work across 20 fields of study over 43 years (1980–2023). We put NLP’s propensity to cite older work in the context of these 20 other fields to analyze whether NLP shows similar temporal citation patterns to them over time or whether differences can be observed. Our analysis, based on a dataset of ~240 million papers, reveals a broader scientific trend: many fields have markedly declined in citing older works (e.g., psychology, computer science). The trend is strongest in NLP and ML research (-12.8% and -5.5% in citation age from previous peaks). Our results suggest that citing more recent works is not directly driven by the growth in publication rates (-3.4% across fields; -5.2% in humanities; -5.5% in formal sciences) — even when controlling for an increase in the volume of papers. Our findings raise questions about the scientific community’s engagement with past literature, particularly for NLP, and the potential consequences of neglecting older but relevant research. The data and a demo showcasing our results are publicly available.

Chinese Automatic Readability Assessment (ARA) aims to classify the reading difficulty of Chinese texts. To address the issues of insufficient high-quality training data and underutilization of linguistic features in existing methods, we propose a method that combines adaptive pre-training with feature fusion based on an interactive attention mechanism. First, we enhance the model’s ability to capture different text difficulties through domain- and task-specific adaptive pre-training. Then, we propose an Adaptive Task-guided Corpus Filtering (ATCF) method, utilizing embeddings generated by the pre-trained model and applying nearest-neighbor search along with a sample balancing mechanism to ensure comprehensive learning across various difficulty levels. Finally, we propose an Interactive Attention-Driven Feature Fusion method that integrates linguistic and deep features, providing rich difficulty information to the model. Experiments on Chinese textbook dataset demonstrate that our method achieves state-of-the-art (SOTA) performance. Transfer learning experiments further indicate that our approach generalizes well to extracurricular reading and Chinese as a Foreign Language (CFL) ARA tasks.

The literature review is an indispensable step in the research process. It provides the benefit of comprehending the research problem and understanding the current research situation while conducting a comparative analysis of prior works. However, literature summary is challenging and time consuming. The previous LLM-based studies on literature review mainly focused on the complete process, including literature retrieval, screening, and summarization. However, for the summarization step, simple CoT method often lacks the ability to provide extensive comparative summary. In this work, we firstly focus on the independent literature summarization step and introduce ChatCite, an LLM agent with human workflow guidance for comparative literature summary. This agent, by mimicking the human workflow, first extracts key elements from relevant literature and then generates summaries using a Reflective Incremental Mechanism. In order to better evaluate the quality of the generated summaries, we devised a LLM-based automatic evaluation metric, G-Score, in refer to the human evaluation criteria. The ChatCite agent outperformed other models in various dimensions in the experiments. The literature summaries generated by ChatCite can also be directly used for drafting literature reviews.

We propose a novel framework that leverages Visual Question Answering (VQA) models to automate the evaluation of LLM-generated data visualizations. Traditional evaluation methods often rely on human judgment, which is costly and unscalable, or focus solely on data accuracy, neglecting the effectiveness of visual communication. By employing VQA models, we assess data representation quality and the general communicative clarity of charts. Experiments were conducted using two leading VQA benchmark datasets, ChartQA and PlotQA, with visualizations generated by OpenAI’s GPT-3.5 Turbo and Meta’s Llama 3.1 70B-Instruct models. Our results indicate that LLM-generated charts do not match the accuracy of the original non-LLM-generated charts based on VQA performance measures. Moreover, while our results demonstrate that few-shot prompting significantly boosts the accuracy of chart generation, considerable progress remains to be made before LLMs can fully match the precision of human-generated graphs. This underscores the importance of our work, which expedites the research process by enabling rapid iteration without the need for human annotation, thus accelerating advancements in this field.

Given the ubiquity of charts as a data analysis, visualization, and decision-making tool across industries and sciences, there has been a growing interest in developing pre-trained foundation models as well as general purpose instruction-tuned models for chart understanding and reasoning. However, existing methods suffer crucial drawbacks across two critical axes affecting the performance of chart representation models: they are trained on data generated from underlying data tables of the charts, ignoring the visual trends and patterns in chart images, and use weakly aligned vision-language backbone models for domain-specific training, limiting their generalizability when encountering charts in the wild. We address these important drawbacks and introduce ChartGemma, a novel chart understanding and reasoning model developed over PaliGemma. Rather than relying on underlying data tables, ChartGemma is trained on instruction-tuning data generated directly from chart images, thus capturing both high-level trends and low-level visual information from a diverse set of charts. Our simple approach achieves state-of-the-art results across 5 benchmarks spanning chart summarization, question answering, and fact-checking, and our elaborate qualitative studies on real-world charts show that ChartGemma generates more realistic and factually correct summaries compared to its contemporaries. We release the code, model checkpoints, dataset, and demos at https://github.com/vis-nlp/ChartGemma.

This paper introduces CharMoral, a dataset designed to analyze the moral evolution of characters in long-form narratives. CharMoral, built from 1,337 movie synopses, includes annotations for character actions, context, and morality labels. To automatically construct CharMoral, we propose a four-stage framework, utilizing Large Language Models, to automatically classify actions as moral or immoral based on context. Human evaluations and various experiments confirm the framework’s effectiveness in moral reasoning tasks in multiple genres. Our code and the CharMoral dataset are publicly available at https://github.com/BaeSuyoung/CharMoral.

Large Language Models (LLMs) exhibit remarkable generative capabilities, enabling the generation of valuable information. Despite these advancements, previous research found that LLMs sometimes struggle with adhering to specific constraints, such as being in a specific place or at a specific time, and at times even overlook them, which leads to responses that are either too generic or not fully satisfactory. Existing approaches attempted to address this issue by decomposing and rewriting input instructions or reflecting on prior failings, yet they fall short in adequately emphasizing specific constraints and unlocking the underlying knowledge, such as programming within the context of software development. In response, this paper proposes a simple yet effective method called Chain-of-Specificity (CoS). Specifically, CoS emphasizes the specific constraints in the input instructions, unlocks knowledge within LLMs, and refines responses. Experiments conducted on publicly available and self-built complex datasets demonstrate that CoS outperforms existing methods in enhancing generated content, especially in terms of specificity. Additionally, as the number of specific constraints increases, other baselines falter, while CoS still performs well. Moreover, we show that distilling responses generated by CoS effectively enhances the ability of smaller models to follow constrained instructions.

Open-ended question answering requires mod- els to find appropriate evidence to form well-reasoned, comprehensive and helpful answers. In practical applications, models also need to engage in extended discussions on potential scenarios closely relevant to the question. With augmentation of retrieval module, open-source Large Language Models (LLMs) can produce coherent answers often with different focuses, but are still sub-optimal in terms of reliable ev- idence selection and in-depth question analysis. In this paper, we propose a novel Chain-of- Discussion framework to leverage the synergy among multiple open-source LLMs aiming to provide more correct and more comprehensive answers for open-ended QA, although they are not strong enough individually. Our exper- iments show that discussions among multiple LLMs play a vital role in enhancing the quality of answers.

Automatically evaluating the quality of responses in dialogue systems is a challenging yet crucial task. Current metrics often fail to align with human judgments, especially when assessing responses that are grammatically correct. To address this issue, we propose a novel metric, called CausalScore, which assesses the relevance of responses by measuring the causal strength between dialogue histories and responses. The causal strength is estimated by utilizing both unconditional dependence and conditional dependencies from dialogue histories to responses. We compare our metric with the existing competitive metrics in terms of their alignment with human judgements. Our experimental results demonstrate that CausalScore significantly surpasses existing state-of-the-art metrics by aligning better with human judgements. Additionally, we collect a dialogue dataset CGDIALOG+ with human-annotated causal relations and a set of pairwise human judgements to facilitate the development of automatic metrics.

Entity Alignment (EA) is essential for integrating Knowledge Graphs (KGs) by matching equivalent entities across diverse KGs. With the rise of multi-modal KGs, which emerged to better depict real-world KGs by integrating visual, textual, and structured data, Multi-Modal Entity Alignment (MMEA) has become crucial in enhancing EA. However, existing MMEA methods often neglect the inherent semantic category information of entities, limiting alignment precision and robustness. To address this, we propose Category-enhanced Entity Alignment (CateEA), which combines implicit entity category information into multi-modal representations. By generating pseudo-category labels from entity embeddings and integrating them into a multi-task learning framework, CateEA captures latent category semantics, enhancing entity representations. CateEA allows for adaptive adjustments of similarity measures, leading to improved alignment precision and robustness in multi-modal contexts. Experiments on benchmark datasets demonstrate that CateEA outperforms state-of-the-art methods in various settings.

Large Language Models (LLMs) have shown outstanding breakthroughs in code generation. Recent work improves code LLMs by training on synthetic data generated by some powerful LLMs, which can be challenging to scale due to the dependence on a teacher model and high generation costs. In this paper, we focus on synthesizing code data at scale and propose a Case2Code task by exploiting the expressiveness and correctness of programs. Case2Code is an inductive inference task that aims to infer underlying code implementations by observing input-output examples or program behaviors, By incorporating LLMs to generate program inputs, and executing the program with these inputs to obtain the program outputs, we can synthesize diverse and high-quality Case2Code data at scale for training and evaluating code LLMs. Experimental results show that case-to-code induction is challenging for current representative LLMs if they are untrained. Models trained with Case2Code improve performance not only on distribution case-to-code induction but also various coding-generation tasks, demonstrating the great potential of large-scale synthetic data and inductive learning.

In today’s assistant landscape, personalisation enhances interactions, fosters long-term relationships, and deepens engagement. However, many systems struggle with retaining user preferences, leading to repetitive user requests and disengagement. Furthermore, the unregulated and opaque extraction of user preferences in industry applications raises significant concerns about privacy and trust, especially in regions with stringent regulations like Europe. In response to these challenges, we propose a long-term memory system for voice assistants, structured around predefined categories. This approach leverages Large Language Models to efficiently extract, store, and retrieve preferences within these categories, ensuring both personalisation and transparency. We also introduce a synthetic multi-turn, multi-session conversation dataset (CarMem), grounded in real industry data, tailored to an in-car voice assistant setting. Benchmarked on the dataset, our system achieves an F1-score of .78 to .95 in preference extraction, depending on category granularity. Our maintenance strategy reduces redundant preferences by 95% and contradictory ones by 92%, while the accuracy of optimal retrieval is at .87. Collectively, the results demonstrate the system’s suitability for industrial applications.

As prompt engineering research rapidly evolves, evaluations beyond accuracy are crucial for developing cost-effective techniques. We present the Economical Prompting Index (EPI), a novel metric that combines accuracy scores with token consumption, adjusted by a user-specified cost concern level to reflect different resource constraints. Our study examines 6 advanced prompting techniques, including Chain-of-Thought, Self-Consistency, and Tree of Thoughts, across 10 widely-used language models and 4 diverse datasets. We demonstrate that approaches such as Self-Consistency often provide statistically insignificant gains while becoming cost-prohibitive. For example, on high-performing models like Claude 3.5 Sonnet, the EPI of simpler techniques like Chain-of-Thought (0.72) surpasses more complex methods like Self-Consistency (0.64) at slight cost concern levels. Our findings suggest a reevaluation of complex prompting strategies in resource-constrained scenarios, potentially reshaping future research priorities and improving cost-effectiveness for end-users.

Estimating the difficulty of multiple-choice questions would be great help for educators who must spend substantial time creating and piloting stimuli for their tests, and for learners who want to practice. Supervised approaches to difficulty estimation have yielded to date mixed results. In this contribution we leverage an aspect of generative large models which might be seen as a weakness when answering questions, namely their uncertainty. Specifically, we exploit model uncertainty towards exploring correlations between two different metrics of uncertainty, and the actual student response distribution. While we observe some present but weak correlations, we also discover that the models’ behaviour is different in the case of correct vs wrong answers, and that correlations differ substantially according to the different question types which are included in our fine-grained, previously unused dataset of 451 questions from a Biopsychology course. In discussing our findings, we also suggest potential avenues to further leverage model uncertainty as an additional proxy for item difficulty.

Utilizing Large Language Models (LLMs) as evaluators to assess the performance of other LLMs has garnered attention. However, this evaluation approach is affected by potential biases within LLMs, raising concerns about the accuracy and reliability of the evaluation results of LLMs. To address this issue, we propose and explore two many-shot In-Context Learning (ICL) prompt templates to help LLM evaluators mitigate potential biases: Many-Shot with Reference (MSwR) and Many-Shot without Reference (MSoR). Specifically, the former utilizes in-context examples with model-generated rationales as references, while the latter does not include these references. Using these prompt designs, we investigate the impact of increasing the number of in-context examples on the consistency and quality of the evaluation results. Experimental results show that advanced LLMs, such as GPT-4, perform better in the many-shot regime than in the zero-shot regime. Furthermore, in most cases, MSwR performs significantly better than MSoR.

Relation-based Argument Mining (RbAM) is the process of automatically determining agreement (support) and disagreement (attack) relations amongst textual arguments (in the binary prediction setting), or neither relation (in the ternary prediction setting). As the number of platforms supporting online debate increases, the need for RbAM becomes ever more urgent, especially in support of downstream tasks. RbAM is a challenging classification task, with existing state-of-the-art methods, based on Language Models (LMs), failing to perform satisfactorily across different datasets. In this paper, we show that general-purpose Large LMs (LLMs), appropriately primed and prompted, can significantly outperform the best performing (RoBERTa-based) baseline. Specifically, we experiment with two open-source LLMs (Llama-2 and Mistral) and with GPT-3.5-turbo on several datasets for (binary and ternary) RbAM, as well as with GPT-4o-mini on samples (to limit costs) from the datasets.

The Myers-Briggs Type Indicator (MBTI) is one of the most influential personality theories reflecting individual differences in thinking, feeling, and behaving. MBTI personality detection has garnered considerable research interest and has evolved significantly over the years. However, this task tends to be overly optimistic, as it currently does not align well with the natural distribution of population personality traits. Specifically, the self-reported labels in existing datasets result in data quality issues and the hard labels fail to capture the full range of population personality distributions. In this paper, we identify the task by constructing MBTIBench, the first manually annotated MBTI personality detection dataset with soft labels, under the guidance of psychologists. Our experimental results confirm that soft labels can provide more benefits to other psychological tasks than hard labels. We highlight the polarized predictions and biases in LLMs as key directions for future research.

We address the task of hierarchical multi-label classification (HMC) of scientific documents at an industrial scale, where hundreds of thousands of documents must be classified across thousands of dynamic labels. The rapid growth of scientific publications necessitates scalable and efficient methods for classification, further complicated by the evolving nature of taxonomies—where new categories are introduced, existing ones are merged, and outdated ones are deprecated. Traditional machine learning approaches, which require costly retraining with each taxonomy update, become impractical due to the high overhead of labelled data collection and model adaptation. Large Language Models (LLMs) have demonstrated great potential in complex tasks such as multi-label classification. However, applying them to large and dynamic taxonomies presents unique challenges as the vast number of labels can exceed LLMs’ input limits. In this paper, we present novel methods that combine the strengths of LLMs with dense retrieval techniques to overcome these challenges. Our approach avoids frequent retraining by leveraging zero-shot and few-shot learning for real-time label assignment. We evaluate the effectiveness of our methods on SSRN, a large repository of preprints spanning multiple disciplines, and demonstrate significant improvements in both classification accuracy and cost-efficiency. By developing a tailored evaluation framework for dynamic taxonomies and publicly releasing our code, this research provides critical insights into applying LLMs for document classification, where the number of classes corresponds to the number of nodes in a large taxonomy, at an industrial scale, significantly contributing to both data science and natural language processing.