论文检索

Hongzhi Qi, Nan Bai, Jianqiang Li, Wei Zhai, Qing Zhao, Qi Gao, Bing Xiang Yang, Guanghui Fu

Cognitive distortion is a critical issue in psychology, with most existing studies based on Burns’ cognitive distortion theory. However, differences in annotation standards lead to variations in building analysis tools, resulting in inconsistent analyses and limiting the generalizability of findings, especially in large-scale and cross-linguistic contexts. To address this issue, we collected all publicly available datasets (four in total) and conducted a series of experiments to evaluate the generalizability of various cross-linguistic models. The results indicate that models exhibit significant performance differences across datasets, highlighting the generalization problem. To mitigate this issue, we propose two solutions. First, we propose a multi-task learning model based on teacher student architecture solution, which demonstrates improved generalization performance in our experiments. Second, we introduce a new dataset (~5,000 samples) derived from reannotating existing open datasets to ensure standardized alignment. The annotation process we provided is interpretable and grounded in psychological principles. Based on this, we constructed large language models with cognitive reasoning chains, enhancing both generalizability and interpretability. This study identifies the generalization challenge in cognitive distortion research, and our experiments show that the proposed solutions significantly improve model performance. The dataset and code are publicly available at: https://github.com/HongzhiQ/CrossLinCD.

Walter Paci, Alessandro Panunzi, Sandro Pezzelle

Implicit content plays a crucial role in political discourse, where systematically employ pragmatic strategies such as implicatures and presuppositions to influence their audiences. Large Language Models (LLMs) have demonstrated strong performance in tasks requiring complex semantic and pragmatic understanding, highlighting their potential for detecting and explaining the meaning of implicit content. However, their ability to do this within political discourse remains largely underexplored. Leveraging, for the very first time, the large IMPAQTS corpus comprising transcribed Italian political speeches with expert annotations of various types of implicit content, we propose methods to test the effectiveness of LLMs in this challenging problem. Through a multiple-choice task and an open-ended generation task, we demonstrate that all tested models struggle to interpret presuppositions and implicatures. To illustrate, the best-performing model provides a fully correct explanation in only one-fourth of cases in the open-ended generation setup. We conclude that current LLMs lack the key pragmatic capabilities necessary for accurately interpreting highly implicit language, such as that found in political discourse. At the same time, we highlight promising trends and future directions for enhancing model performance. We release our data and code at: https://github.com/WalterPaci/IMPAQTS-PID

Kyuhee Kim, Sangah Lee

As large language models (LLMs) become key advisors in various domains, their cultural sensitivity and reasoning skills are crucial in multicultural environments. We introduce Nunchi-Bench, a benchmark designed to evaluate LLMs’ cultural understanding, with a focus on Korean superstitions. The benchmark consists of 247 questions spanning 31 topics, assessing factual knowledge, culturally appropriate advice, and situational interpretation. We evaluate multilingual LLMs in both Korean and English to analyze their ability to reason about Korean cultural contexts and how language variations affect performance. To systematically assess cultural reasoning, we propose a novel verification strategy with customized scoring metrics that capture the extent to which models recognize cultural nuances and respond appropriately. Our findings highlight significant challenges in LLMs’ cultural reasoning. While models generally recognize factual information, they struggle to apply it in practical scenarios. Furthermore, explicit cultural framing enhances performance more effectively than relying solely on the language of the prompt. To support further research, we publicly release Nunchi-Bench alongside a leaderboard.

Jun Zhong, Longwei Xu, Li Kong, Xianzhuo Li, Dandan Liang, Junsheng Zhou

The assessment of children’s narrative ability is crucial for diagnosing language disorders and planning interventions. Distinct from the typical automated essay scoring, this task focuses primarily on evaluating the completeness of narrative content and the coherence of expression, as well as the interpretability of assessment results. To address these issues, we propose a novel computational assessing framework NarGINA, under which the narrative graph is introduced to provide a concise and structured summary representation of narrative text, allowing for explicit narrative measurement. To this end, we construct the first Chinese children’s narrative assessment corpus based on real children’s narrative samples, and we then design a narrative graph construction model and a narrative graph-assisted scoring model to yield accurate narrative ability assessment. Particularly, to enable the scoring model to understand narrative graphs, we propose a multi-view graph contrastive learning strategy to pre-train the graph encoder and apply instruction-tuned large language models to generate scores. The extensive experimental results show that NarGINA can achieve significant performance improvement over the baselines, simultaneously possessing good interpretability. Our findings reveal that the utilization of structured narrative graphs beyond flat text is well suited for narrative ability assessment. The model and data are publicly available at https://github.com/JlexZzz/NarGINA.

Yuuki Yamanaka, Hiroshi Takahashi, Tomoya Yamashita

Large language models (LLMs) are widely used for conversational systems, but they face significant challenges in interpretability of dialogue flow and reproducibility of expert knowledge. To address this, we propose a novel method that extracts flowcharts from dialogue data and incorporates them into LLMs. This approach not only makes the decision-making process more interpretable through visual representation, but also ensures the reproducibility of expert knowledge by explicitly modeling structured reasoning flows. By evaluating on dialogue datasets, we demonstrate that our method effectively reconstructs expert decision-making paths with high precision and recall scores. These findings underscore the potential of flowchart-based decision making to bridge the gap between flexibility and structured reasoning, making chatbot systems more interpretable for developers and end-users.

Yifan Sun, Danding Wang, Qiang Sheng, Juan Cao, Jintao Li

Concept-based explainable approaches have emerged as a promising method in explainable AI because they can interpret models in a way that aligns with human reasoning. However, their adaption in the text domain remains limited. Most existing methods rely on predefined concept annotations and cannot discover unseen concepts, while other methods that extract concepts without supervision often produce explanations that are not intuitively comprehensible to humans, potentially diminishing user trust. These methods fall short of discovering comprehensible concepts automatically. To address this issue, we propose ECO-Concept, an intrinsically interpretable framework to discover comprehensible concepts with no concept annotations. ECO-Concept first utilizes an object-centric architecture to extract semantic concepts automatically. Then the comprehensibility of the extracted concepts is evaluated by large language models. Finally, the evaluation result guides the subsequent model fine-tuning to obtain more understandable explanations using relatively comprehensible concepts. Experiments show that our method achieves superior performance across diverse tasks. Further concept evaluations validate that the concepts learned by ECO-Concept surpassed current counterparts in comprehensibility.

Yile Wang, Zhanyu Shen, Hui Huang

Semantic text representation is a fundamental task in the field of natural language processing. Existing text embedding (e.g., SimCSE and LLM2Vec) have demonstrated excellent performance, but the values of each dimension are difficult to trace and interpret. Bag-of-words, as classic sparse interpretable embeddings, suffers from poor performance. Recently, Benara et al. (2024) propose interpretable text embeddings using large language models, which forms ”0/1” embeddings based on responses to a series of questions. These interpretable text embeddings are typically high-dimensional (larger than 10,000). In this work, we propose Low-dimensional (lower than 500) Dense and Interpretable text embeddings with Relative representations (LDIR). The numerical values of its dimensions indicate semantic relatedness to different anchor texts through farthest point sampling, offering both semantic representation as well as a certain level of traceability and interpretability. We validate LDIR on multiple semantic textual similarity, retrieval, and clustering tasks. Extensive experimental results show that LDIR performs close to the black-box baseline models and outperforms the interpretable embeddings baselines with much fewer dimensions.

Junying Chen, Chi Gui, Anningzhe Gao, Ke Ji, Xidong Wang, Xiang Wan, Benyou Wang

The field of AI healthcare has undergone a significant transformation with the advent of large language models (LLMs), yet the challenges of interpretability within these models remain largely unaddressed. This study introduces **Chain-of-Diagnosis (CoD)** to enhance the interpretability of medical automatic diagnosis. CoD transforms the diagnostic process into a diagnostic chain that mirrors a physician’s thought process, providing a transparent reasoning pathway. Additionally, CoD outputs the disease confidence distribution to ensure transparency in decision-making. This interpretability makes model diagnostics controllable and aids in identifying critical symptoms for inquiry through the entropy reduction of confidences. With CoD, we developed **DiagnosisGPT**, capable of diagnosing 9,604 diseases for validating CoD. Experimental results demonstrate that DiagnosisGPT outperforms other LLMs on automatic diagnostic tasks across three real-world benchmarks. Moreover, DiagnosisGPT provides interpretability while ensuring controllability in diagnostic rigor.

Yuxi Sun, Aoqi Zuo, Wei Gao, Jing Ma

Large Language Models (LLMs) often exhibit knowledge disparities across languages. Encouraging LLMs to \textit{abstain} when faced with knowledge gaps is a promising strategy to reduce hallucinations in multilingual settings. Current abstention strategies for multilingual scenarios primarily rely on generating feedback in various languages using LLMs and performing self-reflection. However, these methods can be adversely impacted by inaccuracies and biases in the generated feedback. To address this, from a causal perspective, we introduce \textit{CausalAbstain}, a method that helps LLMs determine whether to utilize multiple generated feedback responses and how to identify the most useful ones. Extensive experiments demonstrate that \textit{CausalAbstain} effectively selects helpful feedback and enhances abstention decisions with interpretability in both native language (\textit{Casual-native}) and multilingual (\textit{Causal-multi}) settings, outperforming strong baselines on two benchmark datasets covering encyclopedic and commonsense knowledge QA tasks.

Xiaofei Wen, Wenxuan Zhou, Wenjie Jacky Mo, Muhao Chen

Ensuring the safety of large language models (LLMs) is critical as they are deployed in real-world applications. Existing guardrails rely on rule-based filtering or single-pass classification, limiting their ability to handle nuanced safety violations. To address this, we propose ThinkGuard, a critique-augmented guardrail model that distills knowledge from high-capacity LLMs by generating structured critiques alongside safety labels. Fine-tuned on critique-augmented data, the captured deliberative thinking ability drastically enhances the guardrail’s cautiousness and interpretability. Evaluated on multiple safety benchmarks, ThinkGuard achieves the highest average F1 and AUPRC, outperforming all baselines. Compared to LLaMA Guard 3, ThinkGuard improves accuracy by 16.1% and macro F1 by 27.0%. Moreover, it surpasses label-only fine-tuned models, confirming that structured critiques enhance both classification precision and nuanced safety reasoning while maintaining computational efficiency.

Jiamu Zhang, Jiayi Yuan, Andrew Wen, Hoang Anh Duy Le, Yu-Neng Chuang, Soo-Hyun Choi, Rui Chen, Xia Hu

Large Language Models (LLMs) are increasingly adopted across real-world applications, yet traditional evaluations rely on expensive, domain-specific ground-truth labels that are often unavailable or infeasible. We introduce a ground-truth-free evaluation framework focused on reasoning consistency and instruction following, shifting the emphasis from correctness—which is elusive without labels—to transparent, coherent, evidence-based reasoning. Each model response must include a direct answer, a structured multi-step explanation, and supporting evidence, all assessed via semantic similarity and output adherence checks. We further propose TopK-ReRank, which refines rankings by constructing a consensus answer from the most reliable models, reducing ambiguity across diverse reasoning styles. Experiments show that our framework outperforms existing label-free methods, including majority voting, triplet ranking, and peer-review approaches, providing a more interpretable and efficient alternative for evaluating LLMs in the absence of ground-truth labels.

Chris Samarinas, Alexander Krubner, Alireza Salemi, Youngwoo Kim, Hamed Zamani

This paper presents ICAT, an evaluation framework for measuring coverage of diverse factual information in long-form text generation. ICAT breaks down a long output text into a list of atomic claims and not only verifies each claim through retrieval from a (reliable) knowledge source, but also computes the alignment between the atomic factual claims and various aspects expected to be presented in the output. We study three implementations of the ICAT framework, each with a different assumption on the availability of aspects and alignment method. By adopting data from the diversification task in the TREC Web Track and the ClueWeb corpus, we evaluate the ICAT framework. We demonstrate strong correlation with human judgments and provide comprehensive evaluation across multiple state-of-the-art LLMs. Our framework further offers interpretable and fine-grained analysis of diversity and coverage. Its modular design allows for easy adaptation to different domains and datasets, making it a valuable tool for evaluating the qualitative aspects of long-form responses produced by LLMs.

Chi-Jane Chen, Yuhang Chen, Sukwon Yun, Natalie Stanley, Tianlong Chen

Image mass cytometry (IMC) enables high-dimensional spatial profiling by combining mass cytometry’s analytical power with spatial distributions of cell phenotypes. Recent studies leverage large language models (LLMs) to extract cell states by translating gene or protein expression into biological context. However, existing single-cell LLMs face two major challenges: (1) Integration of spatial information—they struggle to generalize spatial coordinates and effectively encode spatial context as text, and (2) Treating each cell independently—they overlook cell-cell interactions, limiting their ability to capture biological relationships. To address these limitations, we propose Spatial2Sentence, a novel framework that integrates both single-cell expression and spatial information into natural language using a multi-sentence approach. Given an expression matrix and spatial coordinates, Spatial2Sentence constructs expression similarity and distance matrices, pairing spatially adjacent and expressionally similar cells as positive pairs while using distant and dissimilar cells as negatives. These multi-sentence representations are processed by LLMs, enabling them to learn cellular interactions in both expression and spatial contexts. Equipped with multi-task learning, Spatial2Sentence outperforms existing single-cell LLMs on preprocessed IMC datasets for diabetes and brain tumors, improving cell-type classification by 5.98% and clinical status prediction by 4.18% on the diabetes dataset while enhancing interpretability. The source code can be found here: https://github.com/UNITES-Lab/Spatial2Sentence.

Ala N. Tak, Amin Banayeeanzade, Anahita Bolourani, Mina Kian, Robin Jia, Jonathan Gratch

Large language models (LLMs) show promising capabilities in predicting human emotions from text. However, the mechanisms through which these models process emotional stimuli remain largely unexplored. Our study addresses this gap by investigating how autoregressive LLMs infer emotions, showing that emotion representations are functionally localized to specific regions in the model. Our evaluation includes diverse model families and sizes, and is supported by robustness checks. We then show that the identified representations are psychologically plausible by drawing on cognitive appraisal theory—a well-established psychological framework positing that emotions emerge from evaluations (appraisals) of environmental stimuli. By causally intervening on construed appraisal concepts, we steer the generation and show that the outputs align with theoretical and intuitive expectations. This work highlights a novel way to causally intervene and control emotion inference, potentially benefiting safety and alignment in sensitive affective domains.

Pedro Calais, Gabriel Franco, Zilu Tang, Themistoklis Nikas, Wagner Meira Jr., Evimaria Terzi, Mark Crovella

Do LLMs process text and mathematics as a unified skill, or do these components rely on distinct underlying mechanisms? We investigate this question by disentangling the textual interpretation and mathematical solving steps in word problems drawn from Brazil’s largest college entrance exam (ENEM) and GSM8K, a popular grade school-level benchmark. Using the symbolic solver SymPy, we transform word problems into equivalent purely mathematical representations, isolating equation formulation from textual comprehension. Our extended benchmarks enable a structured analysis of LLM performance across these two dimensions. Through empirical evaluations, we find that small-scale LLMs struggle significantly more with text interpretation than with equation solving, with accuracy dropping by a factor of 2 to 7 when solving full word problems compared to their math-only counterparts. Exploratory factor analysis confirms a bidimensional structure in LLM reasoning, where models exhibit distinct proficiencies in textual and mathematical components, underscoring the need for targeted improvements in language comprehension. By analyzing the latent factors associated with each model, our findings provide a framework for researchers and practitioners to make informed choices when selecting models based on computational costs and the nature of their tasks.

Ahmad Dawar Hakimi, Ali Modarressi, Philipp Wicke, Hinrich Schuetze

Understanding how large language models (LLMs) acquire and store factual knowledge is crucial for enhancing their interpretability, reliability, and efficiency. In this work, we analyze the evolution of factual knowledge representation in the OLMo-7B model by tracking the roles of its Attention Heads and Feed Forward Networks (FFNs) over training. We classify these components into four roles—general, entity, relation-answer, and fact-answer specific—and examine their stability and transitions. Our results show that LLMs initially depend on broad, general-purpose components, which later specialize as training progresses. Once the model reliably predicts answers, some components are repurposed, suggesting an adaptive learning process. Notably, answer-specific attention heads display the highest turnover, whereas FFNs remain stable, continually refining stored knowledge. These insights offer a mechanistic view of knowledge formation in LLMs and have implications for model pruning, optimization, and transparency.

Pratik Kayal, Pascal Mettes, Nima Dehmamy, Minsu Park

Predicting video popularity is often framed as a supervised learning task, relying heavily on meta-information and aggregated engagement data. However, video popularity is shaped by complex cultural and social factors that such approaches often overlook. We argue that Large Language Models (LLMs), with their deep contextual awareness, can better capture these nuances. To bridge the gap between pixel-based video data and token-based LLMs, we convert frame-level visuals into sequential text representations using Vision-Language Models. This enables LLMs to process multimodal content—titles, frame-based descriptions, and captions—capturing both engagement intensity (view count) and geographic spread (number of countries where a video trends). On 13,639 popular videos, a supervised neural network using content embeddings achieves 80% accuracy, while our LLM-based approach reaches 82% without fine-tuning. Combining the neural network’s predictions with the LLM further improves accuracy to 85.5%. Moreover, the LLM generates interpretable, attribute-based explanations for its predictions. Manual validations confirm the quality of these hypotheses and address concerns about hallucinations in the video-to-text conversion process. Overall, our findings suggest that LLMs, equipped with text-based multimodal representations, offer a powerful, interpretable, and data-efficient solution for tasks requiring rich contextual insight, such as video popularity prediction.

Junling Wang, Anna Rutkiewicz, April Wang, Mrinmaya Sachan

Visuals are valuable tools for teaching math word problems (MWPs), helping young learners interpret textual descriptions into mathematical expressions before solving them.However, creating such visuals is labor-intensive and there is a lack of automated methods to support this process. In this paper, we present Math2Visual, an automatic framework for generating pedagogically meaningful visuals from MWP text descriptions. Math2Visual leverages a pre-defined visual language and a design space grounded in interviews with math teachers, to illustrate the core mathematical relationships in MWPs.Using Math2Visual, we construct an annotated dataset of 1,903 visuals and evaluate Text-to-Image (TTI) models for their ability to generate visuals that align with our design. We further fine-tune several TTI models with our dataset, demonstrating improvements in educational visual generation. Our work establishes a new benchmark for automated generation of pedagogically meaningful visuals and offers insights into key challenges in producing multimodal educational content, such as the misrepresentation of mathematical relationships and the omission of essential visual elements.

Jiachun Li, Pengfei Cao, Yubo Chen, Jiexin Xu, Huaijun Li, Xiaojian Jiang, Kang Liu, Jun Zhao

Chain-of-thought (CoT) prompting demonstrates varying performance under different reasoning tasks.Previous work attempts to evaluate it but falls short in providing an in-depth analysis of patterns that influence the CoT. In this paper, we study the CoT performance from the perspective of effectiveness and faithfulness. For the former, we identify key factors that influence CoT effectiveness on performance improvement, including problem difficulty, information gain, and information flow. For the latter, we interpret the unfaithful CoT issue by conducting a joint analysis of the information interaction among the question, CoT, and answer. The result demonstrates that, when the LLM predicts answers, it can recall correct information missing in the CoT from the question, leading to the problem. Finally, we propose a novel algorithm to mitigate this issue, in which we recall extra information from the question to enhance the CoT generation and evaluate CoTs based on their information gain. Extensive experiments demonstrate that our approach enhances both the faithfulness and effectiveness of CoT.

Yongqi Fan, Yating Wang, Guandong Wang, Zhai Jie, Jingping Liu, Qi Ye, Tong Ruan

Open-ended question answering (QA) is a key task for evaluating the capabilities of large language models (LLMs). Compared to closed-ended QA, it demands longer answer statements, more nuanced reasoning processes, and diverse expressions, making refined and interpretable automatic evaluation both crucial and challenging. Traditional metrics like ROUGE and BERTScore struggle to capture semantic similarities due to different patterns between model responses and reference answers. Current LLM-based evaluation approaches, such as pairwise or listwise comparisons of candidate answers, lack intuitive interpretability. While pointwise scoring of each response provides some descriptions, it fails to adapt across different question contents. Most notably, existing methods overlook the distinction between factoid and non-factoid questions. To address these challenges, we propose MinosEval, a novel evaluation method that first distinguishes open-ended questions and then ranks candidate answers using different evaluation strategies. For factoid questions, it applies an adaptive key-point scoring strategy, while for non-factoid questions, it uses an instance-aware listwise ranking strategy. Experiments on multiple open-ended QA datasets, including self-built ones with more candidate responses to complement community resources, show that MinosEval better aligns with human annotations and offers more interpretable results.