https://arxiv.org/api/80N6nDqj27bXtgxInBGGwtHlKSI 2026-06-21T10:19:08Z 112579 210 15 http://arxiv.org/abs/2602.15851v2 2026-06-17T01:10:31Z

Applications of narrative theories using large language models (LLMs) deliver promising methods in automatic story generation and understanding tasks. Our survey examines how natural language processing (NLP) research uses LLM methods to engage with diverse concepts from narrative studies. We use established distinctions from narratology to categorise ongoing efforts and discover the following: \redtext{(a) narrative texts come from diverse sources beyond just literature, (b) theoretical synthesis and validation are potential outcomes, (c) generation tasks lag behind understanding in several ways: theoretical application, post-training methods, exploring non-fiction narratives and addressing narrative levels beyond fabula and discourse.} For future directions, instead of the pursuit of a single, generalised benchmark for `narrative quality', we believe that progress can benefit from efforts that focus on the following: defining and improving theory-based metrics for individual narrative attributes; continue conducting large-scale, theory-driven literary/social/cultural analysis; generating narratives in situated contexts; and continuing experiments where outputs can be used to validate or refine narrative theories. This work provides a contextual foundation for more systematic and theoretically informed narrative research in NLP by providing an overview to ongoing research efforts and the broader narrative studies landscape.

2026-01-23T23:30:42Z 31 pages David Y. Liu Aditya Joshi Paul Dawson http://arxiv.org/abs/2601.17226v2 2026-06-17T01:09:11Z

Counterfactual story retelling exposes LLM shortcomings in constrained narrative solution spaces where they can no longer rely on recalling memorised training data. Ground-truth-based post-training, such as SFT, fails to teach LLMs how to generate logical and rational narrative events. In this paper, we introduce Retell, Reward, Repeat (RRR), an RL-based pipeline synthesising Structuralist Narratology with scalar narrativity to teach storytelling structure. We extend the TimeTravel dataset with human-annotated stages of narrative equilibrium to evaluate reward models. By using d-RLAIF, RRR derives training signals from the narrativity of textual features without the need for reference outputs. Evaluations demonstrate that RRR-trained LLMs outperform few-shot and SFT baselines in logic, rationality, and completeness, with output quality additionally validated by blind human preference. Relying on a small, query-only dataset, RRR provides a linguistically grounded, cost-effective post-training mechanism for storytelling--a domain currently lacking effective post-training methods. RRR highlights the continued relevance of integrating established linguistic theories into contemporary NLP.

2026-01-23T23:23:42Z 8 Pages, 7 figures David Y. Liu Xanthe Muston Dipankar Srirag Aditya Joshi Sebastian Sequoiah-Grayson http://arxiv.org/abs/2606.18571v1 2026-06-17T00:44:28Z

Mild Cognitive Impairment (MCI) is a medical condition characterized by a noticeable decline in memory, language, or thinking abilities. MCI detection from spontaneous speech is promising for scalable screening. However, learned models often exploit demographic cues correlated with labels, resulting in a large performance gap across subgroups. We present a multimodal framework that combines (i) cross-model fusion between modalities (speech, text, and image), and (ii) unlearning using gradient reversal that discourages the shared embedding from encoding task-irrelevant demographic attributes. Evaluated on the multilingual benchmarks TAUKADIAL and PREPARE, our method outperforms the state-of-the-art multilingual and multimodal baseline in MCI classification while substantially reducing the performance gap across patient subgroups (sex and language). We further analyze transfer across datasets, showing that demographic unlearning helps learn more robust representations for MCI detection.

2026-06-17T00:44:28Z Interspeech 2026 William Nguyen Jiali Cheng Hadi Amiri http://arxiv.org/abs/2606.18543v1 2026-06-16T23:37:52Z

Language model agents are becoming proficient executors at isolated, short-horizon tasks such as software engineering and customer service. Yet real-world challenges require a combination of sophisticated skills that remain largely untested in agents: (1) navigating long horizons amid uncertainty; (2) acquiring information in noisy environments; (3) adapting to a changing world; (4) orchestrating multiple moving parts toward a coherent goal. We introduce CEO-Bench, which evaluates these capabilities together by simulating a representative real-world task: operating a startup for 500 days. An agent manages pricing, marketing, budgeting, and many other aspects of a fictional company through a programmable Python interface, operating in the same environment and facing the same challenges as a human CEO. Success demands analyzing noisy, interconnected business databases, translating signals into sound strategy, and coordinating many decisions with programming. The strongest agents write sophisticated code that simulates customer cohorts to forecast future cash and mines negotiation history to uncover hidden customer preferences. Even so, most state-of-the-art models struggle in this environment. Only Claude Opus 4.8 and GPT-5.5 finish above the $1M starting balance, and neither consistently turns a profit. CEO-Bench takes a first step toward measuring the intelligence required to drive sustained, adaptive progress over time.

2026-06-16T23:37:52Z Haozhe Chen Karthik Narasimhan Zhuang Liu http://arxiv.org/abs/2605.16385v3 2026-06-16T22:53:58Z

Geometric problem solving, as a typical multimodal reasoning problem, has attracted much attention and made great progress recently, however most of works focus on plane geometry while usually fail in solid geometry due to 3D spatial diagrams and complex reasoning. To bridge this gap, we introduce Hilbert-Geo, the first unified formal language framework for solid geometry, including an extensive predicate library and a dedicated theorem bank. Based on this framework, we propose a Parse2Reason method containing two steps of first parsing then reasoning. In the parsing step, we utilize conditional description language (CDL), a formalized language composed of predicates specifically designed to construct geometric conditions, to represent both problem description (natural text) and solid diagrams (visual image). In the reasoning step, we leverage those formal CDL and the theorem bank to perform relational inference and algebraic computation, generating strictly correct, verifiable, and human-readable reasoning processes. Notably, our proposed Hilbert-Geo is also applicable to plane geometry. To advance geometric reasoning, we curate two expert-annotated dataset SolidFGeo2k and PlaneFGeo3k, which are furnished with geometric formal language annotations, solutions and answers. Extensive experiments show that our proposed method achieves the state-of-the-art (SOTA) performance 77.3% in SolidFGeo2k and 84.1% in MathVerse-Solid (one small subset in MathVerse dedicated to solid geometry), substantially outperforming leading MLLMs, such as Gemini-2.5-pro (54.2% on SolidFGeo2k) and GPT-5 (62.9% on MathVerse-Solid). In addition, our method achieves the SOTA accuracy 80.2% in PlaneFGeo3k, demonstrating the generality of the Hilbert-Geo in geometric reasoning. Our code and datasets are released at https://github.com/PremiLab-Math/Hilbert-Geo.

2026-05-11T13:42:41Z Computer Vision and Pattern Recognition (CVPR), 2026 Ruoran Xu Haoyu Cheng Bin Dong Qiufeng Wang http://arxiv.org/abs/2606.18530v1 2026-06-16T22:51:02Z

Domain-camouflaged injection attacks embed malicious instructions in retrieved content using domain-appropriate vocabulary, evading standard detectors that rely on syntactic injection markers. When detection fails, practitioners need to know which defense architectures reduce attack success. We evaluate five prompting-based defenses (spotlighting, paraphrasing, prompt sandwiching, and two combinations) against domain-camouflaged injection across three model families (Claude Haiku, Llama 3.1 8B, Gemini 2.0 Flash) and three deployment domains (financial, legal, general) using 3,510 trials. Paraphrasing retrieved content before agent processing is the most consistently effective defense in this benchmark, reducing camouflage attack success rate by 55-84\% depending on model, and achieves lower attack success rates than our Llama Guard 4 configuration on every model tested. Defense effectiveness is strongly model-dependent: spotlighting halves attack success on Claude Haiku but provides no benefit on Llama 3.1 8B. Financial domain deployments face the highest residual risk at 26-33\% baseline attack success rate, with no prompting-based defense fully eliminating the threat on weaker models. These results provide the first systematic evaluation of prompting-based defenses specifically against camouflage-class injection attacks and establish benchmark-based recommendations for practitioners. All tasks use synthetically constructed professional documents; whether these benchmark rankings generalize to real enterprise documents remains an open question.

2026-06-16T22:51:02Z 9 pages, 4 figures, 4 tables; under review at the AdvML-Frontiers x CoTMA workshop, COLM 2026 Aaditya Pai http://arxiv.org/abs/2606.18520v1 2026-06-16T22:20:52Z

Computing geometric representations of data is a cornerstone of modern machine learning, typically achieved by training dual encoders which map queries and documents into a shared embedding space. Recent work of You et al. [NeurIPS '25] has extended this approach to hierarchical retrieval, where relevance is determined by the ancestor-descendant relationships in a Directed Acyclic Graph (DAG). While previous work has shown that valid embeddings exist when the number of descendants is small, these bounds degrade significantly for deep hierarchies, requiring dimensions as large as the total number of nodes. In this paper, we investigate compact reachability embeddings for more general graph classes and provide theoretical guarantees for representing hierarchies using embeddings whose dimension depends on structural graph parameters. We prove that for any directed tree, there exists a reachability embedding in constant dimension 3, independent of the tree's size or depth. We generalize this result to graphs characterized by treewidth $t$, constructing embeddings of dimension $O(t \log n)$, where $n$ is the number of nodes. Complementing these upper bounds, we provide matching or near-matching lower bounds, showing that dimension $Ω(n)$ is necessary for general DAGs and $Ω(t/\log(n/t))$ is required for graphs of treewidth $t$. We also obtain upper and lower bounds parameterized by the number of cross-edges in the DAG. We additionally show that our embeddings can be constructed on real world datasets, and that they give much smaller dimensions in high recall regimes compared to prior embeddings with theoretical guarantees.

2026-06-16T22:20:52Z Published at the 39th Annual Conference on Learning Theory (COLT) 2026. 22 Pages Prashant Gokhale Piotr Indyk Yuhao Liu Sandeep Silwal Tony Chang Wang Haike Xu http://arxiv.org/abs/2606.18508v1 2026-06-16T21:50:01Z

Retrieval-augmented generation (RAG) systems depend critically on how documents are chunked and searched. Fine-grained chunks can improve retrieval precision but expand the search space, increasing latency and cost; larger chunks reduce the number of candidates but make dense similarity less reliable, as the representation for each chunk mixes multiple topics and introduces more semantic noise. This trade-off becomes especially limiting in deep research tasks, where retrieval must be both fast and precise across large, heterogeneous corpora. We introduce MCompassRAG, a metadata-guided retrieval framework that uses topic-level signals as a semantic compass for selecting relevant evidence. Instead of relying only on cosine similarity between queries and noisy chunk embeddings, MCompassRAG enriches chunk representations with topic metadata in the same embedding space and trains a lightweight retriever through LLM-teacher distillation. At inference time, MCompassRAG performs topic-aware retrieval without additional LLM calls, improving both efficiency and evidence quality. Across six complex retrieval benchmarks, MCompassRAG improves information efficiency (IE) by 8.24% on average with over 5 times lower latency than the strongest efficient RAG baselines. Code is available on https://github.com/AmirAbaskohi/MCompassRAG.

2026-06-16T21:50:01Z Amirhossein Abaskohi Raymond Li Gaetano Cimino Peter West Giuseppe Carenini Issam H. Laradji http://arxiv.org/abs/2510.09905v2 2026-06-16T21:38:14Z

When an AI assistant remembers that Sarah is a single mother working two jobs, does it interpret her stress differently than if she were a wealthy executive? As personalized AI systems increasingly incorporate long-term user memory, understanding how this memory shapes emotional reasoning is critical. We investigate how user memory affects emotional intelligence in large language models (LLMs) by evaluating 15 models on human-validated emotional intelligence tests. We find that identical scenarios paired with different user profiles produce systematically divergent emotional interpretations. Across validated user-independent emotional scenarios and diverse user profiles, systematic biases emerged in several high-performing LLMs where advantaged profiles received more accurate emotional interpretations. Moreover, LLMs demonstrate significant disparities across demographic factors in emotion reasoning and supportive recommendations tasks, indicating that personalization mechanisms can embed social hierarchies into models' emotional reasoning. These results highlight a key challenge for memory-enhanced AI: systems designed for personalization may reinforce social inequalities. To mitigate these disparities, we curate a general-purpose preference dataset designed to reduce demographic profiles' influence on emotional understanding.

2025-10-10T22:39:37Z 19 pages 5 figures Xi Fang Weijie Xu Yuchong Zhang Stephanie Eckman Scott Nickleach Chandan K. Reddy http://arxiv.org/abs/2606.18502v1 2026-06-16T21:30:10Z

Large language model (LLM)-based multi-agent systems demonstrate strong performance on complex reasoning and task execution, enabling broad enterprise applications. However, production deployment remains challenging due to domain-specific customization requirements and high latency and inference costs in agentic workflows. We propose a unified framework for customization and efficient deployment of multi-agent systems in real-world settings. The first stage, Agentic Model Customization, combines continual pretraining, supervised fine-tuning, and preference optimization to adapt a compact model to specialized domains while retaining strong agentic capabilities. The second stage, Inference Optimization, integrates speculative decoding and FP8 quantization with targeted calibration to enable cost-efficient serving with minimal quality loss. Across enterprise workloads, our framework enables rapid domain adaptation and achieves a 4.48x speedup in throughput while maintaining performance and improving robustness on long-tail scenarios.

2026-06-16T21:30:10Z Preprint Paresh Dashore Shreyas Kulkarni Uttam Gurram Nadia Bathaee Kartik Balasubramaniam Genta Indra Winata Sambit Sahu Shi-Xiong Zhang http://arxiv.org/abs/2606.16000v2 2026-06-16T21:25:39Z

We introduce GRACE-DS, a Guarded Reward-guided Agent Correction Environment in Data Science for pre-deployment evaluation of LLM-powered AutoML agents. GRACE-DS is a set of evaluation metrics in an isolated environment that can be applied to tabular ML tasks specific to a particular organization. It exposes agents to realistic workflow stages, from planning and data inspection through feature engineering, model development, validation, and code repair to final submission, while hidden executable validators measure not only final predictive performance but also leakage avoidance, reproducibility, protocol validity, correction behavior, and reward alignment. The strongest structured regime, flexible iterative interaction (our approach), achieves higher end-to-end normalized hidden-test quality than single-shot generation, unstructured interaction, and restart-based baselines, while also improving protocol-valid completion. Validated across more than 7,000 episodes, these results establish GRACE-DS as a robust platform for assessing the capacity of LLM-based AutoML agents to execute machine learning workflows under production-like conditions and in accordance with organization-specific requirements.

2026-06-14T19:58:06Z Aleksandr Tsymbalov Danis Zaripov Artem Epifanov Anastasiya Palienko http://arxiv.org/abs/2606.18487v1 2026-06-16T20:59:55Z

The standard heuristic of selecting the SFT checkpoint with the highest pass@1 for GRPO can fail when SFT compresses the rollout distribution. For binary rewards, the expected within group advantage variance is $p(1{-}p)(g{-}1)/g$; when early GRPO drives $p$ below $p^*(g)$, most groups have identical rewards and provide no group relative signal. We study SFT depth ladders for Qwen2.5-Coder-3B and DeepSeek-Coder-6.7B. We test Qwen2.5-Coder-3B across five depths and three seeds, and DeepSeek-Coder-6.7B across four matched depths and three seeds. On Qwen, pre RL pass@1 rises with SFT depth, but peak GRPO pass@10 falls from $0.806$ to $0.481$ (3 seed mean, $n{=}20$); pre RL entropy is positively associated with the GRPO outcome ($ρ{=}{+}0.69$). On DeepSeek, pass@1 remains far above $p^*(8){=}0.083$, and GRPO outcomes compress rather than invert. A two stage diagnostic, combining pre RL entropy triage with an early GRPO entropy monitor, flags high risk checkpoints and can stop failing runs early. Simple KL to reference regularisation and label smoothing variants do not rescue the collapsed Qwen checkpoint in our setting, suggesting the failure is not a trivial GRPO hyperparameter artefact.

2026-06-16T20:59:55Z 14 pages, 6 figures. Accepted at the Deep Learning for Code (DL4C) Workshop at ICML 2026 Siddharth Aphale Kelly Liu http://arxiv.org/abs/2506.12311v3 2026-06-16T20:55:53Z

Text-to-speech (TTS) for Modern Hebrew is challenged by the language's orthographic complexity, with existing solutions ignoring underspecified phonetic features such as stress. We present a framework for more phonetically accurate Hebrew TTS with four contributions: (1) Phonikud, an open-source Hebrew grapheme-to-phoneme (G2P) system that outputs fully-specified International Phonetic Alphabet (IPA) transcriptions, designed by augmenting a base diacritizer. (2) The ILSpeech corpus of paired Hebrew audio, text, and expert IPA annotations. (3) A benchmark for the previously unmeasured task of Hebrew G2P conversion. (4) Hebrew audio-to-IPA models capturing previously disregarded phonetic details for automatic TTS evaluation. Our results show that Phonikud more accurately predicts Hebrew phonemes than prior methods, and that small, local TTS models with phonetic input from Phonikud approach large proprietary systems. We release our code, data, and models at https://phonikud.github.io.

2025-06-14T02:16:38Z Accepted to Interspeech 2026. Project page: https://phonikud.github.io Yakov Kolani Maxim Melichov Cobi Calev Morris Alper http://arxiv.org/abs/2503.04989v2 2026-06-16T20:44:40Z

Classification of textual data in terms of sentiment, or more nuanced sociopsychological markers (e.g., agency), is now a popular approach commonly applied at the sentence level. In this paper, we exploit the integrated gradient (IG) method to capture the classification output at the word level, revealing which words actually contribute to the classification process. This approach improves explainability and provides in-depth insights into the text. We focus on sociopsychological markers beyond sentiment and investigate how to effectively train IG in agency, one of the very few markers for which a verified deep learning classifier, BERTAgent, is currently available. Performance and system parameters are carefully tested, alternatives to the IG approach are evaluated, and the usefulness of the result is verified in a relevant application scenario. The method is also applied in a scenario where only a small labeled dataset is available, with the aim of exploiting IG to identify the salient words that contribute to building the different classes that relate to relevant sociopsychological markers. To achieve this, an uncommon training procedure that encourages overfitting is employed to enhance the distinctiveness of each class. The results are analyzed through the lens of social psychology, offering valuable insights.

2025-03-06T21:35:24Z Submitted to IEEE Trans. on Computational Social Systems Ali Aghababaei Jan Nikadon Magdalena Formanowicz Maria Laura Bettinsoli Carmen Cervone Caterina Suitner Tomaso Erseghe http://arxiv.org/abs/2604.23130v2 2026-06-16T20:40:27Z

Jailbreak attacks expose a persistent failure mode in safety-aligned LLMs: models can be pushed into harmful behavior, but the internal representations enabling this shift remain poorly localized. Recent mechanistic safety studies often explain such behavior through broad representational objects, including global refusal directions, activation steering vectors, and refusal-related SAE features. We instead ask whether jailbreak vulnerability can be traced to finer-grained, prompt-conditioned SAE feature subgroups. We introduce a token-driven mechanistic pipeline that decomposes the residual stream of Gemma-2-2B into Sparse Autoencoder (SAE) features and identifies feature subgroups associated with unsafe behavior. Using single-category unsafe examples from BeaverTails to reduce cross-category interference, we extract harmful concepts from adversarial responses and align them with concept-relevant prompt tokens through subspace similarity. We then apply three feature-grouping strategies: cluster-based, hierarchical-linkage, and single-token-driven, to identify SAE feature subgroups across all 26 layers. Finally, we amplify the top features in each subgroup and evaluate the resulting generations with a standardized harmfulness judge. Single-token-driven grouping achieves harmfulness comparable to full cluster-based grouping, showing that individual harmful prompt tokens are sufficient to localize vulnerability-relevant SAE feature subgroups without relying on broader cluster-level aggregation. These subgroups appear across early and mid-to-late layers, with stronger concentration in mid-to-late layers, where targeted steering exposes specific model vulnerabilities. Overall, our results suggest that jailbreak susceptibility can be traced to sparse, token-localized SAE feature subgroups, complementing prior accounts based on broad adversarial, refusal, or steering directions.

2026-04-25T03:54:03Z Nilanjana Das Mathew Dawit Aman Chadha Manas Gaur