论文检索

Xingwei Tan, Marco Valentino, Mahmud Elahi Akhter, Maria Liakata, Nikolaos Aletras

Large language models (LLMs) have shown strong performance in many reasoning benchmarks. However, recent studies have pointed to memorization, rather than generalization, as one of the leading causes for such performance. LLMs, in fact, are susceptible to content variations, demonstrating a lack of robust planning or symbolic abstractions supporting their reasoning process. To improve reliability, many attempts have been made to combine LLMs with symbolic methods. Nevertheless, existing approaches fail to effectively leverage symbolic representations due to the challenges involved in developing reliable and scalable verification mechanisms. In this paper, we propose to overcome such limitations by synthesizing high-quality symbolic reasoning trajectories with stepwise pseudo-labels at scale via Monte Carlo estimation. A Process Reward Model (PRM) can be efficiently trained based on the synthesized data and then used to select more symbolic trajectories. The trajectories are then employed with Direct Preference Optimization (DPO) and Supervised Fine-Tuning (SFT) to improve logical reasoning and generalization. Our results on benchmarks (i.e., FOLIO and LogicAsker) show the effectiveness of the proposed method with gains on frontier and open-weight models. Moreover, additional experiments on claim verification data reveal that fine-tuning on the generated symbolic reasoning trajectories enhances out-of-domain generalizability, suggesting the potential impact of the proposed method in enhancing planning and logical reasoning.

Pittawat Taveekitworachai, Potsawee Manakul, Sarana Nutanong, Kunat Pipatanakul

This paper investigates prior prompt engineering (pPE) in the context of reinforcement fine-tuning (RFT), where language models (LMs) are incentivized to exhibit behaviors that maximize performance through reward signals. While existing RFT research has primarily focused on algorithms, reward shaping, and data curation, the design of the prior prompt–the instructions prepended to queries during training to elicit behaviors such as step-by-step reasoning–remains underexplored. We investigate whether different pPE approaches can guide LMs to internalize distinct behaviors after RFT. Inspired by inference-time prompt engineering (iPE), we translate five representative iPE strategies–reasoning, planning, code-based reasoning, knowledge recall, and null-example utilization–into corresponding pPE approaches. We experiment with Qwen2.5-7B using each of the pPE approaches, then evaluate performance on in-domain and out-of-domain benchmarks (e.g., AIME2024, HumanEval+, and GPQA-Diamond). Our results show that all pPE-trained models surpass their iPE-prompted counterparts, with the null-example pPE approach achieving the largest average performance gain and the highest improvement on AIME2024 and GPQA-Diamond, surpassing the commonly used reasoning approach. Furthermore, by adapting a behavior-classification framework, we demonstrate that different pPE strategies instill distinct behavioral styles in the resulting models. These findings position pPE as a powerful yet understudied axis for RFT.

Hammad Ayyubi, Puneet Mathur, Mehrab Tanjim, Vlad I Morariu

Editing documents and PDFs using natural language instructions is desirable for many reasons – ease of use, increasing accessibility to non-technical users, and for creativity. To do this automatically, a system needs to first understand the user’s intent and convert this to an executable plan or command, and then the system needs to identify or localize the elements that the user desires to edit. While there exist methods that can accomplish these tasks, a major bottleneck in these systems is the inability to ground the spatial edit location effectively. We address this gap through our proposed system, DELOC (Document Element LOCalizer). DELOC adapts the grounding capabilities of existing Multimodal Large Language Model (MLLM) from natural images to PDFs. This adaptation involves two novel contributions: 1) synthetically generating PDF-grounding instruction tuning data from partially annotated datasets; and 2) synthetic data cleaning via Code-NLI, an NLI-inspired process to clean data using generated Python code. The effectiveness of DELOC is apparent in the >3x zero-shot improvement it achieves over the next best Multimodal LLM, GPT-4o.

Rohit Khoja, Devanshu Gupta, Yanjie Fu, Dan Roth, Vivek Gupta

Querying tables with unstructured data is challenging due to the presence of text (or image), either embedded in the table or in external paragraphs, which traditional SQL struggles to process, especially for tasks requiring semantic reasoning. While Large Language Models (LLMs) excel at understanding context, they face limitations with long input sequences. Existing approaches that combine SQL and LLM typically rely on rigid, predefined workflows, limiting their adaptability to complex queries. To address these issues, we introduce Weaver, a modular pipeline that dynamically integrates SQL and LLM for table-based question answering (Table QA). Weaver generates a flexible, step-by-step plan that combines SQL for structured data retrieval with LLMs for semantic processing. By decomposing complex queries into manageable subtasks, Weaver improves accuracy and generalization. Our experiments show that consistently outperforms state-of-the-art methods across four Table QA datasets, reducing both API calls and error rates.

Jaehong Yoon, Shoubin Yu, Mohit Bansal

Recent video generative models primarily rely on detailed, labor-intensive text prompts for tasks, like inpainting or style editing, limiting adaptability for personal/raw videos. This paper proposes RACCooN, a versatile and user-friendly video-to-paragraph-to-video editing method, supporting diverse video editing capabilities, such as removal, addition, and modification, through a unified pipeline. RACCooN consists of two principal stages: Video-to-Paragraph (V2P), which automatically generates structured video descriptions capturing both scene context and object details, and Paragraph-to-Video (P2V), where users (optionally) refine these descriptions to guide a video diffusion model for flexible content modifications, including removing, changing subjects, and/or adding new objects. Key contributions of RACCooN include: (1) A multi-granular spatiotemporal pooling strategy for structured video understanding, capturing both broad context and fine-grained details of major objects to enable precise text-based video editing without the need for complex human annotations. (2) A video generative model fine-tuned on our curated video-paragraph-mask dataset, enhances the editing and inpainting quality. (3) The capability to seamlessly generate new objects in videos by forecasting their movements through automatically generated mask planning. In the end, users can easily edit complex videos with RACCooN’s automatic explanations and guidance. We demonstrate its versatile capabilities in video-to-paragraph generation (up to 9.4%p absolute improvement in human evaluations) and video content editing (relative to 49.7% lower FVD), and can be integrated with SoTA video generation models for further enhancement.

Ruochun Jin, Xiyue Wang, Dong Wang, Haoqi Zheng, Yunpeng Qi, Silin Yang, Meng Zhang

Table question answering (TQA) requires accurate retrieval and reasoning over tabular data. Existing approaches attempt to retrieve query-relevant content before leveraging large language models (LLMs) to reason over long tables. However, these methods often fail to accurately retrieve contextually relevant data which results in information loss, and suffer from excessive encoding overhead. In this paper, we propose TALON, a multi-agent framework designed for question answering over long tables. TALON features a planning agent that iteratively invokes a tool agent to access and manipulate tabular data based on intermediate feedback, which progressively collects necessary information for answer generation, while a critic agent ensures accuracy and efficiency in tool usage and planning. In order to comprehensively assess the effectiveness of TALON, we introduce two benchmarks derived from the WikiTableQuestion and BIRD-SQL datasets, which contain tables ranging from 50 to over 10,000 rows. Experiments demonstrate that TALON achieves average accuracy improvements of 7.5% and 12.0% across all language models, establishing a new state-of-the-art in long-table question answering. Our code is publicly available at: https://github.com/Wwestmoon/TALON.

Julius Mayer, Mohamad Ballout, Serwan Jassim, Farbod Nosrat Nezami, Elia Bruni

Vision-Language Models (VLMs) are known to struggle with spatial reasoning and visual alignment. To help overcome these limitations, we introduce iVISPAR, an interactive multimodal benchmark designed to evaluate the spatial reasoning capabilities of VLMs acting as agents. iVISPAR is based on a variant of the sliding tile puzzle—a classic problem that demands logical planning, spatial awareness, and multi-step reasoning. The benchmark supports visual 3D, 2D, and text-based input modalities, enabling comprehensive assessments of VLMs’ planning and reasoning skills. We evaluate a broad suite of state-of-the-art open-source and closed-source VLMs, comparing their performance while also providing optimal path solutions and a human baseline to assess the task’s complexity and feasibility for humans. Results indicate that while VLMs perform better on 2D tasks compared to 3D or text-based settings, they struggle with complex spatial configurations and consistently fall short of human performance, illustrating the persistent challenge of visual alignment. This underscores critical gaps in current VLM capabilities, highlighting their limitations in achieving human-level cognition. Project website: https://microcosm.ai/ivispar.

Yuanzhang Lin, Zhe Zhang, He Rui, Qingao Dong, Mingyi Zhou, Jing Zhang, Xiang Gao, Hailong Sun

Mobile task automation is an emerging technology that leverages AI to automatically execute routine tasks by users’ commands on mobile devices like Android, thus enhancing efficiency and productivity. While large language models (LLMs) excel at general mobile tasks through training on massive datasets, they struggle with app-specific workflows. To solve this problem, we designed UI Map, a structured representation of target app’s UI information. We further propose a UI Map-guided LLM-based approach UICompass to automate mobile tasks. Specifically, UICompass first leverages static analysis and LLMs to automatically build UI Map from either source codes of apps or byte codes (i.e., APK packages). During task execution, UICompass mines the task-relevant information from UI Map to feed into the LLMs, generate a planned paths, and adaptively adjust the path based on the actual app state and action history. Experimental results demonstrate that UICompass achieves a 15.87% higher task executing success rate than SOTA approaches. Even when only APK is available, UICompass maintains superior performance, demonstrating its applicability to closed-source apps.

Yuhao Yang, Jiabin Tang, Lianghao Xia, Xingchen Zou, Yuxuan Liang, Chao Huang

Real-world data combines structured (e.g., graph connections) and unstructured (e.g., text, visuals) formats, capturing explicit relationships (e.g., social links) and implicit semantic interdependencies (e.g., knowledge graphs). We propose GraphAgent, an automated agent pipeline addressing both explicit and implicit graph-enhanced semantic dependencies for predictive (e.g., node classification) and generative (e.g., text generation) tasks. GraphAgent integrates three components: (i) a Graph Generator Agent creating knowledge graphs for semantic dependencies; (ii) a Task Planning Agent interpreting user queries and formulating tasks via self-planning; and (iii) a Task Execution Agent automating task execution with tool matching. These agents combine language and graph language models to reveal complex relational and semantic patterns. Extensive experiments on diverse datasets validate GraphAgent’s effectiveness in graph-related predictive and text generative tasks. GraphAgent is open-sourced at: https://anonymous.4open.science/r/GraphAgent-Submit-6F52/.

Yejin Son, Minseo Kim, Sungwoong Kim, Seungju Han, Jian Kim, Dongju Jang, Youngjae Yu, Chan Young Park

Large Language Models (LLMs) are increasingly used for decision making in embodied agents, yet existing safety evaluations often rely on coarse success rates and domain-specific setups, making it difficult to diagnose why and where these models fail. This obscures our understanding of embodied safety and limits the selective deployment of LLMs in high-risk physical environments. We introduce SAFEL, the framework for systematically evaluating the physical safety of LLMs in embodied decision making. SAFEL assesses two key competencies: (1) rejecting unsafe commands via the Command Refusal Test, and (2) generating safe and executable plans via the Plan Safety Test. Critically, the latter is decomposed into functional modules, goal interpretation, transition modeling, action sequencing enabling fine-grained diagnosis of safety failures. To support this framework, we introduce EMBODYGUARD, a PDDL-grounded benchmark containing 942 LLM-generated scenarios covering both overtly malicious and contextually hazardous instructions. Evaluation across 13 state-of-the-art LLMs reveals that while models often reject clearly unsafe commands, they struggle to anticipate and mitigate subtle, situational risks. Our results highlight critical limitations in current LLMs and provide a foundation for more targeted, modular improvements in safe embodied reasoning.

Zhenyun Deng, Yulong Chen, Andreas Vlachos

Extracting individual sentences from a document as evidence or reasoning steps is commonly done in many NLP tasks. However, extracted sentences often lack context necessary to make them understood, e.g., coreference and background information. To this end, we propose a content selection and planning framework for zero-shot decontextualisation, which determines what content should be mentioned and in what order for a sentence to be understood out of context. Specifically, given a potentially ambiguous sentence and its context, we first segment it into basic semantically-independent units. We then identify potentially ambiguous units from the given sentence, and extract relevant units from the context based on their discourse relations. Finally, we generate a content plan to rewrite the sentence by enriching each ambiguous unit with its relevant units. Experimental results demonstrate that our approach is competitive for sentence decontextualisation, producing sentences that exhibit better semantic integrity and discourse coherence, outperforming existing methods.

Ruibin Xiong, Yimeng Chen, Dmitrii Khizbullin, Mingchen Zhuge, Jürgen Schmidhuber

Long-form writing agents require flexible integration and interaction across information retrieval, reasoning, and composition. Current approaches rely on predefined workflows and rigid thinking patterns to generate outlines before writing, resulting in constrained adaptability during writing. In this paper we propose WriteHERE, a general agent framework that achieves human-like adaptive writing through recursive task decomposition and dynamic integration of three fundamental task types: retrieval, reasoning, and composition. Our methodology features: 1) a planning mechanism that interleaves recursive task decomposition and execution, eliminating artificial restrictions on writing workflow; and 2) integration of task types that facilitates heterogeneous task decomposition. Evaluations on both fiction writing and technical report generation show that our method consistently outperforms state-of-the-art approaches across all automatic evaluation metrics, demonstrating the effectiveness and broad applicability of our proposed framework. We have publicly released our code and prompts to facilitate further research.

Yi Feng, Jiaqi Wang, Wenxuan Zhang, Zhuang Chen, Shen Yutong, Xiyao Xiao, Minlie Huang, Liping Jing, Jian Yu

Recent progress in large language models (LLMs) has opened new possibilities for mental health support, yet current approaches lack realism in simulating specialized psychotherapy and fail to capture therapeutic progression over time. Narrative therapy, which helps individuals transform problematic life stories into empowering alternatives, remains underutilized due to limited access and social stigma. We address these limitations through a comprehensive framework with two core components. First, **INT** (Interactive Narrative Therapist) simulates expert narrative therapists by planning therapeutic stages, guiding reflection levels, and generating contextually appropriate responses through retrieval-augmentation. Second, **IMA** (Innovative Moment Assessment) provides a therapy-centric evaluation method that quantifies effectiveness by tracking “Innovative Moments” (IMs), critical narrative shifts in client speech signaling therapy progress. Experimental results on 260 simulated clients and 230 human participants reveal that **INT** consistently outperforms standard methods in therapeutic quality and depth. We further demonstrate the effectiveness of **INT** in synthesizing high-quality support conversations to facilitate social applications.

Erik Arakelyan, Pasquale Minervini, Patrick Lewis, Pat Verga, Isabelle Augenstein

Modern Question Answering (QA) and Reasoning approaches with Large Language Models (LLMs) commonly use Chain-of-Thought (CoT) prompting but struggle with generating outputs faithful to their intermediate reasoning chains. While neuro-symbolic methods like Faithful CoT (F-CoT) offer higher faithfulness through external solvers, they require code-specialized models and struggle with ambiguous tasks.We introduce \textbf{F}aithful \textbf{L}ogic-\textbf{A}ided \textbf{R}easoning and \textbf{E}xploration (\textbf{FLARE}), which uses LLMs to plan solutions, formalize queries into logic programs, and simulate code execution through multi-hop search without external solvers. Our method achieves SOTA results on \mathbf{7} out of \mathbf{9} diverse reasoning benchmarks and 3 out of 3 logic inference benchmarks while enabling measurement of reasoning faithfulness. We demonstrate that model faithfulness correlates with performance and that successful reasoning traces show an 18.1% increase in unique emergent facts, 8.6% higher overlap between code-defined and execution-trace relations, and 3.6% reduction in unused relations.

Jaewoo Ahn, Junseo Kim, Heeseung Yun, Jaehyeon Son, Dongmin Park, Jaewoong Cho, Gunhee Kim

GUI agents powered by LLMs show promise in interacting with diverse digital environments. Among these, video games offer a valuable testbed due to their varied interfaces, with adventure games posing additional challenges through complex, narrative-driven interactions. Existing game benchmarks, however, lack diversity and rarely evaluate agents on completing entire storylines. To address this, we introduce FlashAdventure, a benchmark of 34 Flash-based adventure games designed to test full story arc completion and tackle the observation-behavior gap—the challenge of remembering and acting on earlier gameplay information. We also propose CUA-as-a-judge, an automated gameplay evaluator, and COAST, an agentic framework leveraging long-term clue memory to better plan and solve sequential tasks. Experiments show current GUI agents struggle with full story arcs, while COAST improves milestone completion by bridging the observation-behavior gap. Nonetheless, a marked discrepancy between humans and best-performing agents warrants continued research efforts to narrow this divide.

Yangfan Wang, Jie Liu, Chen Tang, Lian Yan, Jingchi Jiang

Multi-hop question answering faces substantial challenges due to data sparsity, which increases the likelihood of language models learning spurious patterns. To address this issue, prior research has focused on diversifying question generation through content planning and varied expression. However, these approaches often emphasize generating simple questions and neglect the integration of essential knowledge, such as relevant sentences within documents. This paper introduces the **Knowledge Composition Sampling (KCS)**, an innovative framework designed to expand the diversity of generated multi-hop questions by sampling varied knowledge compositions within a given context. KCS models the knowledge composition selection as a sentence-level conditional prediction task and utilizes a probabilistic contrastive loss to predict the next most relevant piece of knowledge. During inference, we employ a stochastic decoding strategy to effectively balance accuracy and diversity. Compared to competitive baselines, our KCS improves the overall accuracy of knowledge composition selection by 3.9%, and its application for data augmentation yields improvements on HotpotQA and 2WikiMultihopQA datasets. Our code is available at: https://github.com/yangfanww/kcs.

Zhipeng Yang, Junzhuo Li, Siyu Xia, Xuming Hu

We show that large language models (LLMs) exhibit an \textit{internal chain-of-thought}: they sequentially decompose and execute composite tasks layer-by-layer. Two claims ground our study: (i) distinct subtasks are learned at different network depths, and (ii) these subtasks are executed sequentially across layers. On a benchmark of 15 two-step composite tasks, we employ layer-from context-masking and propose a novel cross-task patching method, confirming (i). To examine claim (ii), we apply LogitLens to decode hidden states, revealing a consistent layerwise execution pattern. We further replicate our analysis on the real-world \text{TRACE} benchmark, observing the same stepwise dynamics. Together, our results enhance LLMs transparency by showing their capacity to internally plan and execute subtasks (or instructions), opening avenues for fine-grained, instruction-level activation steering.

Zifeng Ding, Sikuan Yan, Moy Yuan, Xianglong Hu, Fangru Lin, Andreas Vlachos

Temporal reasoning and planning are essential capabilities for large language models (LLMs), yet most existing benchmarks evaluate them in isolation and under limited forms of complexity. To address this gap, we introduce the Temporal Constraint-based Planning (TCP) benchmark, that jointly assesses both capabilities. Each instance in TCP features a naturalistic dialogue around a collaborative project, where diverse and interdependent temporal constraints are explicitly or implicitly expressed, and models must infer an optimal schedule that satisfies all constraints. To construct TCP, we generate abstract problem prototypes that are then paired with realistic scenarios from various domains and enriched into dialogues using an LLM. A human quality check is performed on a sampled subset to confirm the reliability of our benchmark. We evaluate state-of-the-art LLMs and find that even the strongest models may struggle with TCP, highlighting its difficulty and revealing limitations in LLMs’ temporal constraint-based planning abilities. We analyze underlying failure cases, open source our benchmark, and hope our findings can inspire future research.

Mihir Parmar, Palash Goyal, Xin Liu, Yiwen Song, Mingyang Ling, Chitta Baral, Hamid Palangi, Tomas Pfister

Recently, decomposing complex problems into simple subtasks–a crucial part of human-like natural planning–to solve the given problem has significantly boosted the performance of large language models (LLMs). However, leveraging such planning structures during post-training to boost the performance of smaller open-source LLMs remains underexplored. Motivated by this, we introduce PLAN-TUNING, a unified post-training framework that (i) distills synthetic task decompositions (termed “planning trajectories”) from large-scale LLMs and (ii) fine-tunes smaller models via supervised and reinforcement-learning objectives designed to mimic these planning processes to improve complex reasoning. On GSM8k and the MATH benchmarks, plan-tuned models outperform strong baselines by an average ~7%. Furthermore, plan-tuned models show better generalization capabilities on out-of-domain datasets, with average ~10% and ~12% performance improvements on OlympiadBench and AIME 2024, respectively. Our detailed analysis demonstrates how planning trajectories improves complex reasoning capabilities, showing that PLAN-TUNING is an effective strategy for improving task-specific performance of smaller LLMs.

Mihir Parmar, Xin Liu, Palash Goyal, Yanfei Chen, Long Le, Swaroop Mishra, Hossein Mobahi, Jindong Gu, Zifeng Wang, Hootan Nakhost 等

Recent agent frameworks and inference-time algorithms often struggle with natural planning problems due to limitations in verifying generated plans or reasoning and varying complexity of instances within a single task. Many existing methods for these tasks either perform task-level verification without considering constraints or apply inference-time algorithms without adapting to instance-level complexity. To address these limitations, we propose PlanGEN, a model-agnostic and easily scalable agent framework with three key components: constraint, verification, and selection agents. Specifically, our approach proposes constraint-guided iterative verification to enhance performance of inference-time algorithms–Best of \mathcal{N}, Tree-of-Thought, and REBASE. In PlanGEN framework, the selection agent optimizes algorithm choice based on instance complexity, ensuring better adaptability to complex planning problems. Experimental results demonstrate significant improvements over the strongest baseline across multiple benchmarks, achieving state-of-the-art results on NATURAL PLAN (~8%\uparrow), OlympiadBench (~4%\uparrow), DocFinQA (~7%\uparrow), and GPQA (~1%\uparrow). Our key finding highlights that constraint-guided iterative verification improves inference-time algorithms, and adaptive selection further boosts performance on complex planning and reasoning problems.