论文检索

Haixia Han, Jiaqing Liang, Jie Shi, Qianyu He, Yanghua Xiao

Generative Language Models (LMs) such as ChatGPT have exhibited remarkable performance across various downstream tasks. Nevertheless, one of their most prominent drawbacks is generating inaccurate or false information with a confident tone. Previous studies have devised sophisticated pipelines and prompts to induce large LMs to exhibit the capability for self-correction. However, large LMs are explicitly prompted to verify and modify their answers separately rather than completing all steps spontaneously like humans. Moreover, these complex prompts are extremely challenging for small LMs to follow. In this paper, we introduce the Intrinsic Self-Correction (ISC) in generative language models, aiming to correct the initial output of LMs in a self-triggered manner, even for those small LMs with 6 billion parameters. Specifically, we devise a pipeline for constructing self-correction data and propose Partial Answer Masking (PAM), aiming to endow the model with the capability for intrinsic self-correction through fine-tuning. We conduct experiments using LMs with parameters sizes ranging from 6 billion to 13 billion in two tasks, including commonsense reasoning and factual knowledge reasoning. Our experiments demonstrate that the outputs generated using ISC outperform those generated without self-correction. We believe that the output quality of even small LMs can be further improved by empowering them with the ability to intrinsic self-correct.

Xinyan Guan, Yanjiang Liu, Hongyu Lin, Yaojie Lu, Ben He, Xianpei Han, Le Sun

Incorporating factual knowledge in knowledge graph is regarded as a promising approach for mitigating the hallucination of large language models (LLMs). Existing methods usually only use the user's input to query the knowledge graph, thus failing to address the factual hallucination generated by LLMs during its reasoning process. To address this problem, this paper proposes Knowledge Graph-based Retrofitting (KGR), a new framework that incorporates LLMs with KGs to mitigate factual hallucination during the reasoning process by retrofitting the initial draft responses of LLMs based on the factual knowledge stored in KGs. Specifically, KGR leverages LLMs to extract, select, validate, and retrofit factual statements within the model-generated responses, which enables an autonomous knowledge verifying and refining procedure without any additional manual efforts. Experiments show that KGR can significantly improve the performance of LLMs on factual QA benchmarks especially when involving complex reasoning processes, which demonstrates the necessity and effectiveness of KGR in mitigating hallucination and enhancing the reliability of LLMs.

Yuheng Chen, Pengfei Cao, Yubo Chen, Kang Liu, Jun Zhao

Pre-trained language models (PLMs) contain vast amounts of factual knowledge, but how the knowledge is stored in the parameters remains unclear. This paper delves into the complex task of understanding how factual knowledge is stored in multilingual PLMs, and introduces the Architecture-adapted Multilingual Integrated Gradients method, which successfully localizes knowledge neurons more precisely compared to current methods, and is more universal across various architectures and languages. Moreover, we conduct an in-depth exploration on knowledge neurons, leading to the following two important discoveries: (1) The discovery of Language-Independent Knowledge Neurons, which store factual knowledge in a form that transcends language. We design cross-lingual knowledge editing experiments, demonstrating that the PLMs can accomplish this task based on language-independent neurons; (2) The discovery of Degenerate Knowledge Neurons, a novel type of neuron showing that different knowledge neurons can store the same fact. Its property of functional overlap endows the PLMs with a robust mastery of factual knowledge. We design fact-checking experiments, proving that the degenerate knowledge neurons can help the PLMs to detect wrong facts. Experiments corroborate these findings, shedding light on the mechanisms of factual knowledge storage in multilingual PLMs, and contribute valuable insights to the field. The code is available at https://github.com/heng840/AMIG.

Mingshan Chang, Min Yang, Qingshan Jiang, Ruifeng Xu

Despite having achieved notable success for aspect-based sentiment analysis (ABSA), deep neural networks are susceptible to spurious correlations between input features and output labels, leading to poor robustness. In this paper, we propose a novel Counterfactual-Enhanced Information Bottleneck framework (called CEIB) to reduce spurious correlations for ABSA. CEIB extends the information bottleneck (IB) principle to a factual-counterfactual balancing setting by integrating augmented counterfactual data, with the goal of learning a robust ABSA model. Concretely, we first devise a multi-pattern prompting method, which utilizes the large language model (LLM) to generate high-quality counterfactual samples from the original samples. Then, we employ the information bottleneck principle and separate the mutual information into factual and counterfactual parts. In this way, we can learn effective and robust representations for the ABSA task by balancing the predictive information of these two parts. Extensive experiments on five benchmark ABSA datasets show that our CEIB approach achieves superior prediction performance and robustness over the state-of-the-art baselines. Code and data to reproduce the results in this paper is available at: https://github.com/shesshan/CEIB.

Bo Xiong, Mojtaba Nayyeri, Linhao Luo, Zihao Wang, Shirui Pan, Steffen Staab

Reasoning with knowledge graphs (KGs) has primarily focused on triple-shaped facts. Recent advancements have been explored to enhance the semantics of these facts by incorporating more potent representations, such as hyper-relational facts. However, these approaches are limited to atomic facts, which describe a single piece of information. This paper extends beyond atomic facts and delves into nested facts, represented by quoted triples where subjects and objects are triples themselves (e.g., ((BarackObama, holds_position, President), succeed_by, (DonaldTrump, holds_position, President))). These nested facts enable the expression of complex semantics like situations over time and logical patterns} over entities and relations. In response, we introduce NestE, a novel KG embedding approach that captures the semantics of both atomic and nested factual knowledge. NestE represents each atomic fact as a 1*3 matrix, and each nested relation is modeled as a 3*3 matrix that rotates the 1*3 atomic fact matrix through matrix multiplication. Each element of the matrix is represented as a complex number in the generalized 4D hypercomplex space, including (spherical) quaternions, hyperbolic quaternions, and split-quaternions. Through thorough analysis, we demonstrate the embedding's efficacy in capturing diverse logical patterns over nested facts, surpassing the confines of first-order logic-like expressions. Our experimental results showcase NestE's significant performance gains over current baselines in triple prediction and conditional link prediction. The code and pre-trained models are open available at https://github.com/xiongbo010/NestE.

Zhipeng Qian, Yiwei Ma, Jiayi Ji, Xiaoshuai Sun

Referring 3D instance segmentation is a challenging task aimed at accurately segmenting a target instance within a 3D scene based on a given referring expression. However, previous methods have overlooked the distinct roles played by different words in referring expressions. Additionally, they have failed to incorporate the positional relationship within referring expressions with the spatial correlations in 3D scenes. To alleviate these issues, we present a novel model called X-RefSeg3D, which constructs a cross-modal graph for the input 3D scene and unites textual and spatial relationships for reasoning via graph neural networks. Our approach begins by capturing object-specific text features, which are then fused with the instance features to construct a comprehensive cross-modal scene graph. Subsequently, we integrate the obtained cross-modal features into graph neural networks, leveraging the K-nearest algorithm to derive explicit instructions from expressions and factual relationships in scenes. This enables the effective capture of higher-order relationships among instances, thereby enhancing feature fusion and facilitating reasoning. Finally, the refined feature undergoes a matching module to compute the ultimate matching score. Experimental results on ScanRefer demonstrate the effectiveness of our method, surpassing previous approaches by a substantial margin of +3.67% in terms of mIOU.

Chengen Lai, Shengli Song, Shiqi Meng, Jingyang Li, Sitong Yan, Guangneng Hu

Natural language explanation in visual question answer (VQA-NLE) aims to explain the decision-making process of models by generating natural language sentences to increase users' trust in the black-box systems. Existing post-hoc methods have achieved significant progress in obtaining a plausible explanation. However, such post-hoc explanations are not always aligned with human logical inference, suffering from the issues on: 1) Deductive unsatisfiability, the generated explanations do not logically lead to the answer; 2) Factual inconsistency, the model falsifies its counterfactual explanation for answers without considering the facts in images; and 3) Semantic perturbation insensitivity, the model can not recognize the semantic changes caused by small perturbations. These problems reduce the faithfulness of explanations generated by models. To address the above issues, we propose a novel self-supervised Multi-level Contrastive Learning based natural language Explanation model (MCLE) for VQA with semantic-level, image-level, and instance-level factual and counterfactual samples. MCLE extracts discriminative features and aligns the feature spaces from explanations with visual question and answer to generate more consistent explanations. We conduct extensive experiments, ablation analysis, and case study to demonstrate the effectiveness of our method on two VQA-NLE benchmarks.

Shitian Zhao, Zhuowan Li, Yadong Lu, Alan Yuille, Yan Wang

While Multi-modal Language Models (MLMs) demon strate impressive multimodal ability they still struggle on providing factual and precise responses for tasks like vi sual question answering (VQA). In this paper we address this challenge from the perspective of contextual informa tion. We propose Causal Context Generation Causal-CoG which is a prompting strategy that engages contextual infor mation to enhance precise VQA during inference. Specifi cally we prompt MLMs to generate contexts i.e text de scription of an image and engage the generated contexts for question answering. Moreover we investigate the ad vantage of contexts on VQA from a causality perspective introducing causality filtering to select samples for which contextual information is helpful. To show the effective ness of Causal-CoG we run extensive experiments on 10 multimodal benchmarks and showconsistent improvements e.g. +6.30% on POPE +13.69% on Vizwiz and +6.43% on VQAv2 compared to direct decoding surpassing exist ing methods. We hope Casual-CoG inspires explorations of context knowledge in multimodal models and serves as a plug-and-play strategy for MLM decoding.

Qifan Yu, Juncheng Li, Longhui Wei, Liang Pang, Wentao Ye, Bosheng Qin, Siliang Tang, Qi Tian, Yueting Zhuang

Multi-modal Large Language Models (MLLMs) tuned on machine-generated instruction-following data have demonstrated remarkable performance in various multimodal understanding and generation tasks. However the hallucinations inherent in machine-generated data which could lead to hallucinatory outputs in MLLMs remain under-explored. This work aims to investigate various hallucinations (i.e. object relation attribute hallucinations) and mitigate those hallucinatory toxicities in large-scale machine-generated visual instruction datasets. Drawing on the human ability to identify factual errors we present a novel hallucination detection and elimination framework HalluciDoctor based on the cross-checking paradigm. We use our framework to identify and eliminate hallucinations in the training data automatically. Interestingly HalluciDoctor also indicates that spurious correlations arising from long-tail object co-occurrences contribute to hallucinations. Based on that we execute counterfactual visual instruction expansion to balance data distribution thereby enhancing MLLMs' resistance to hallucinations. Comprehensive experiments on hallucination evaluation benchmarks show that our method successfully mitigates 44.6% hallucinations relatively and maintains competitive performance compared to LLaVA. The data and code for this paper are publicly available.

Dingkang Yang, Kun Yang, Mingcheng Li, Shunli Wang, Shuaibing Wang, Lihua Zhang

Context-aware emotion recognition (CAER) has recently boosted the practical applications of affective computing techniques in unconstrained environments. Mainstream CAER methods invariably extract ensemble representations from diverse contexts and subject-centred characteristics to perceive the target person's emotional state. Despite advancements the biggest challenge remains due to context bias interference. The harmful bias forces the models to rely on spurious correlations between background contexts and emotion labels in likelihood estimation causing severe performance bottlenecks and confounding valuable context priors. In this paper we propose a counterfactual emotion inference (CLEF) framework to address the above issue. Specifically we first formulate a generalized causal graph to decouple the causal relationships among the variables in CAER. Following the causal graph CLEF introduces a non-invasive context branch to capture the adverse direct effect caused by the context bias. During the inference we eliminate the direct context effect from the total causal effect by comparing factual and counterfactual outcomes resulting in bias mitigation and robust prediction. As a model-agnostic framework CLEF can be readily integrated into existing methods bringing consistent performance gains.

Yushi Hu, Otilia Stretcu, Chun-Ta Lu, Krishnamurthy Viswanathan, Kenji Hata, Enming Luo, Ranjay Krishna, Ariel Fuxman

Solving complex visual tasks such as "Who invented the musical instrument on the right?" involves a composition of skills: understanding space recognizing instruments and also retrieving prior knowledge. Recent work shows promise by decomposing such tasks using a large language model (LLM) into an executable program that invokes specialized vision models. However generated programs are error-prone: they omit necessary steps include spurious ones and are unable to recover when the specialized models give incorrect outputs. Moreover they require loading multiple models incurring high latency and computation costs. We propose Visual Program Distillation (VPD) an instruction-tuning framework that produces a vision-language model (VLM) capable of solving complex visual tasks with a single forward pass. VPD distills the reasoning ability of LLMs by using them to sample multiple candidate programs which are then executed and verified to identify the correct one. It translates each correct program into a language description of the reasoning steps which are then distilled into a VLM. Extensive experiments show that VPD improves the VLM's ability to count understand spatial relations and reason compositionally. Our VPD-trained PaLI-X outperforms all prior VLMs achieving state-of-the-art performance across complex vision tasks including MMBench OK-VQA A-OKVQA TallyQA POPE and Hateful Memes. An evaluation with human annotators also confirms that VPD improves model response factuality and consistency. Finally experiments on content moderation demonstrate that VPD is also helpful for adaptation to real-world applications with limited data.

Tianyu Yu, Yuan Yao, Haoye Zhang, Taiwen He, Yifeng Han, Ganqu Cui, Jinyi Hu, Zhiyuan Liu, Hai-Tao Zheng, Maosong Sun 等

Multimodal Large Language Models (MLLMs) have recently demonstrated impressive capabilities in multimodal understanding reasoning and interaction. However existing MLLMs prevalently suffer from serious hallucination problems generating text that is not factually grounded in associated images. The problem makes existing MLLMs untrustworthy and thus impractical in real-world (especially high-stakes) applications. To address the challenge we present RLHF-V which enhances MLLM trustworthiness via behavior alignment from fine-grained correctional human feedback. Specifically RLHF-V collects human preference in the form of segment-level corrections on hallucinations and performs dense direct preference optimization over the human feedback. Comprehensive experiments on five benchmarks in both automatic and human evaluation show that RLHF-V can enable substantially more trustworthy MLLM behaviors with promising data and computation efficiency. Remarkably using 1.4k annotated data samples RLHF-V significantly reduces the hallucination rate of the base MLLM by 34.8% outperforming the concurrent LLaVA-RLHF trained on 10k annotated data. The final model achieves state-of-the-art performance in trustworthiness among open-source MLLMs and shows better robustness than GPT-4V in preventing hallucinations aroused from over-generalization. All the data code and model weights will be released to facilitate future research.

Andong Wang, Bo Wu, Sunli Chen, Zhenfang Chen, Haotian Guan, Wei-Ning Lee, Li Erran Li, Chuang Gan

Reasoning from visual dynamics scenes has many real world applications. However existing video reasoning benchmarks are still inadequate since they were mainly designed for factual or situated reasoning and rarely involve broader knowledge in the real world. Our work aims to delve deeper into reasoning evaluations specifically within dynamic open-world and structured context knowledge. We propose a new benchmark (SOK-Bench) consisting of 44K questions and 10K situations with instance-level annotations depicted in the videos. The reasoning process is required to understand and apply situated knowledge and general knowledge for problem-solving. To create such a dataset we propose an automatic and scalable generation method to generate question-answer pairs knowledge graphs and rationales by instructing the combinations of LLMs and MLLMs. Concretely we first extract observable situated entities relations and processes from videos for situated knowledge and then extend to open-world knowledge beyond the visible content. The task generation is facilitated through multiple dialogues as iterations and subsequently corrected and refined by our designed self-promptings and demonstrations. With a corpus of both explicit situated facts and implicit commonsense we generate associated question-answer pairs and reasoning processes finally followed by manual reviews for quality assurance. We evaluated recent mainstream large vision language models on the benchmark and found several insightful conclusions. For more information please refer to our benchmark at www.bobbywu.com/SOKBench.

Anastasiia Sedova, Robert Litschko, Diego Frassinelli, Benjamin Roth, Barbara Plank

One of the major aspects contributing to the striking performance of large language models (LLMs) is the vast amount of factual knowledge accumulated during pre-training. Yet, many LLMs suffer from self-inconsistency, which raises doubts about their trustworthiness and reliability. This paper focuses on entity type ambiguity, analyzing the proficiency and consistency of state-of-the-art LLMs in applying factual knowledge when prompted with ambiguous entities. To do so, we propose an evaluation protocol that disentangles knowing from applying knowledge, and test state-of-the-art LLMs on 49 ambiguous entities. Our experiments reveal that LLMs struggle with choosing the correct entity reading, achieving an average accuracy of only 85%, and as low as 75% with underspecified prompts. The results also reveal systematic discrepancies in LLM behavior, showing that while the models may possess knowledge, they struggle to apply it consistently, exhibit biases toward preferred readings, and display self-inconsistencies. This highlights the need to address entity ambiguity in the future for more trustworthy LLMs.

Evgeniia Razumovskaia, Ivan Vulić, Pavle Marković, Tomasz Cichy, Qian Zheng, Tsung-Hsien Wen, Paweł Budzianowski

Factual faithfulness is a crucial requirement in information-seeking dialogue: the system should respond to the user queries so that the responses are meaningful and aligned with the knowledge provided to the system. However, most modern large language models (LLMs) suffer from hallucinations, that is, they generate responses not supported by or even contradicting the knowledge source. To mitigate the issue and increase faithfulness of information-seeking dialogue systems supported by the LLMs, we introduce BeInfo, a simple yet effective method that applies ‘behavioural tuning’ on the LLMs to aid information-seeking dialogue. Relying on three standard information seeking dialogue datasets, we show that models tuned with BeInfo become considerably more faithful to the knowledge source both for datasets and domains seen during BeInfo-tuning, as well as on unseen domains, when applied in a zero-shot manner. In addition, we present a ‘real-life’ case study on conversations with real users, showcasing that the models with 3B parameters (e.g., Flan-T5) tuned with BeInfo demonstrate strong performance on data from real ‘production’ conversations: when tuned on a limited amount of such realistic in-domain dialogues, they surpass much larger LLMs used ‘off-the-shelf’, both on automatic and human evaluation metrics.

Chao-Wei Huang, Yun-Nung Chen

Large language models have demonstrated significant potential as the next-generation information access engines. However, their reliability is hindered by issues of hallucination and generating non-factual content. This is particularly problematic in long-form responses, where assessing and ensuring factual accuracy is complex. In this paper, we address this gap by proposing FactAlign, a novel alignment framework designed to enhance the factuality of LLMs’ long-form responses while maintaining their helpfulness. We introduce fKTO, a fine-grained, sentence-level alignment algorithm that extends the Kahneman-Tversky Optimization (KTO) alignment method. Leveraging recent advances in automatic factuality evaluation, FactAlign utilizes fine-grained factuality assessments to guide the alignment process. Our experiments on open-domain prompts and information-seeking questions demonstrate that FactAlign significantly improves the factual accuracy of LLM responses while also improving their helpfulness. Further analyses identify that FactAlign is capable of training LLMs to provide more information without losing factual precision, thus improving the factual F1 score. Our source code, datasets, and trained models are publicly available at https://github.com/MiuLab/FactAlign

Alexander Spangher, James Youn, Matt DeButts, Nanyun Peng, Emilio Ferrara, Jonathan May

Human writers plan, _then_ write. For large language models (LLMs) to play a role in longer-form article generation, we must understand the planning steps humans make before writing. We explore one kind of planning, source-selection in news, as a case-study for evaluating plans in long-form generation. We ask: why do _specific_ stories call for _specific_ kinds of sources? We imagine a generative process for story writing where a source-selection schema is first selected by a journalist, and then sources are chosen based on categories in that schema. Learning the article’s _plan_ means predicting the schema initially chosen by the journalist. Working with professional journalists, we adapt five existing schemata and introduce three new ones to describe journalistic plans for the inclusion of sources in documents. Then, inspired by Bayesian latent-variable modeling, we develop metrics to select the most likely plan, or schema, underlying a story, which we use to compare schemata. We find that two schemata: _stance_ and _social affiliation_ best explain source plans in most documents. However, other schemata like _textual entailment_ explain source plans in factually rich topics like “Science”. Finally, we find we can predict the most suitable schema given just the article’s headline with reasonable accuracy. We see this as an important case-study for human planning, and provides a framework and approach for evaluating other kinds of plans, like discourse or plot-oriented plans. We release a corpora, _NewsSources_, with annotations for 4M articles, for further study.

Shashank Sonkar, Naiming Liu, Richard Baraniuk

The pursuit of personalized education has led to the integration of Large Language Models (LLMs) in developing intelligent tutoring systems. To better understand and adapt to individual student needs, including their misconceptions, LLMs need to be trained on extensive datasets of student-tutor dialogues. Our research uncovers a fundamental challenge in this approach: the “Student Data Paradox”. This paradox emerges when LLMs, trained on student data to understand learner behavior, inadvertently compromise their own factual knowledge and reasoning abilities. We investigate this paradox by training state-of-the-art language models on student-tutor dialogue datasets and evaluating their performance across multiple benchmarks. These benchmarks assess various aspects of language model capabilities, including reasoning, truthfulness, and common sense understanding. Our findings reveal significant declines in the models’ performance across these diverse benchmarks, indicating a broad impact on their capabilities when trained to model student behavior. Our research makes two primary contributions: (1) empirical demonstration of the Student Data Paradox through quantitative analysis of model performance, and (2) introduction of “hallucination tokens” as a mitigation strategy. These tokens, while improving performance, highlight the persistent challenge of balancing accurate student behavior modeling with maintaining the LLM’s integrity as an educational tool.This study emphasizes the need for innovative solutions to reconcile the conflicting goals of faithfully understanding diverse student cognition while preserving the model’s ability to provide accurate information and guidance.

Subramanian Chidambaram, Li Erran Li, Min Bai, Xiaopeng Li, Kaixiang Lin, Xiong Zhou, Alex C. Williams

Large Language Models (LLMs) are increasingly used for generating code solutions, empowered by features like self-debugging and self-reflection. However, LLMs often struggle with complex programming problems without human guidance. This paper investigates the strategies employed by expert programmers to steer code-generating LLMs toward successful outcomes. Through a study involving experts using natural language to guide GPT-4, Gemini Ultra, and, Claude 3.5 Sonnet on highly difficult programming challenges, we frame our analysis using the “Socratic Feedback” paradigm for understanding effective steering strategies. By analyzing 30 conversational transcripts across all three models, we map observed feedback strategies to five stages of Socratic Questioning: Definition, Elenhus, Maieutic, Dialectic, and Counter-factual reasoning. We find evidence that by employing a combination of different Socratic feedback strategies across multiple turns, programmers successfully guided the models to solve 74% of the problems that the models initially failed to solve on their own.

Hengrui Gu, Kaixiong Zhou, Yili Wang, Ruobing Wang, Xin Wang

During pre-training, the Text-to-Image (T2I) diffusion models encode factual knowledge into their parameters. These parameterized facts enable realistic image generation, but they may become obsolete over time, thereby misrepresenting the current state of the world. Knowledge editing techniques aim to update model knowledge in a targeted way. However, facing the dual challenges posed by inadequate editing datasets and unreliable evaluation criterion, the development of T2I knowledge editing encounter difficulties in effectively generalizing injected knowledge. In this work, we design a T2I knowledge editing framework by comprehensively spanning on three phases: First, we curate a dataset CAKE, comprising paraphrase and multi-object test, to enable more fine-grained assessment on knowledge generalization. Second, we propose a novel criterion, adaptive CLIP threshold, to effectively filter out false successful images under the current criterion and achieve reliable editing evaluation. Finally, we introduce MPE, a simple but effective approach for T2I knowledge editing. Instead of tuning parameters, MPE precisely recognizes and edits the outdated part of the conditioning text-prompt to accommodate the up-to-date knowledge. A straightforward implementation of MPE (Based on in-context learning) exhibits better overall performance than previous model editors. We hope these efforts can further promote faithful evaluation of T2I knowledge editing methods.