Papers

We present a fully automated pipeline that transforms large collections of court decisions into legal commentaries for statutes - without providing any handcrafted doctrinal framework. Using 4.555 decisions of the German Federal Court of Justice that cite sections 242, 280, 812 and 823 of the German Civil Code (BGB), we extract paragraph-level chunks, summarize their reasoning, and derive keywords, which are embedded and clustered. For each cluster, an LLM generates headings and synthesizes citation-rich sections, which are then merged into coherent commentaries by four state-of-the-art LLMs. We evaluate along five dimensions - topical relevance, heading-match, citation faithfulness, cluster distinction and logical ordering - using both a human expert and an LLM-judge. Our results show that commentary-like argument mining from court decisions to generate reports that can be refreshed within minutes at minimal cost is feasible, yet they highlight limitations arising from restricted sources and the normativity of legal reasoning.

Document retrieval in real-world scenarios faces significant challenges due to diverse document formats and modalities. Traditional text-based approaches rely on tailored parsing techniques that disregard layout information and are prone to errors, while recent parsing-free visual methods often struggle to capture fine-grained textual semantics in text-rich scenarios. To address these limitations, we propose Unveil, a novel visual-textual embedding framework that effectively integrates textual and visual features for robust document representation. Through knowledge distillation, we transfer the semantic understanding capabilities from the visual-textual embedding model to a purely visual model, enabling efficient parsing-free retrieval while preserving semantic fidelity. Experimental results demonstrate that our visual-textual embedding method surpasses existing approaches, while knowledge distillation successfully bridges the performance gap between visual-textual and visual-only methods, improving both retrieval accuracy and efficiency.

Children expertly construct mental models under uncertainty despite limited world knowledge. Which computational principles make this possible? We study this question in an interactive physical Box Task in which participants infer a latent cause through interaction with a stochastic and partially observed environment. The Box Task was originally designed to study children's puzzle-solving qualitatively. We contribute a computational analysis of behavior that formalizes the cognitive process by which children generate new hypotheses and respond to environmental uncertainty as inference unfolds. Our modeling framework is based on Bayesian particle-based inference, which maintains a small set of working hypotheses and continually generates new ones. We give two complementary implementations of hypothesis representations within this framework: (1) as constraints on evidence, and (2) as executable programs evaluated against evidence. Using the constraint-based formulation, we show that the model needs to account both for subjective reliability of evidence and online hypothesis generation to explain key signatures of children's behavior. Using the programmatic formulation, we show that hypotheses may take the form of partial compositional rules to guide information gathering when evidence interpretation is hard. The programmatic hypotheses are implemented using LLM-based program synthesis, evaluated across three prompt variants and several LLM backends, demonstrating robustness of our approach. Together, these results extend current models of children's inference to account for subjective reliability of evidence and open-ended hypothesis spaces, and provide a computational analysis of the form generated hypotheses may take in practice.

In non-invasive neural language decoding, results can be inflated by sources that are not stimulus-evoked neural evidence: decoder priors, embedding-based metrics, and non-neural structural nuisances such as signal duration. The methodological challenge is therefore attribution: a reported gain is more informative when it can be traced to a specific source. We recast stimulus-locked MEG-to-audio retrieval as an auditing framework that separates apparent performance into three sources - structural shortcuts, window-level stimulus-locked evidence, and cross-window contextual aggregation - and provides a diagnostic for each. Signal-blind Gaussian noise reaches 66.3% Rank@1 (R@1) under variable-length decoding but collapses to near chance once fixed-duration windows and stimulus-identity splits are enforced, isolating structural leakage. Under these controls, fixed-window retrieval recovers measurable MEG-audio discriminability, while an oracle sentence-bucket diagnostic shows that 95.7% of Top-1 errors select the wrong sentence, localising the residual bottleneck to sentence-level competition. We audit this contextual source with Group Context Bias (GCB), an inference-time additive logit bias that pools sentence-consistent evidence across windows while leaving the base retrieval scores and candidate pool fixed. Used as a score-space intervention, GCB makes the contextual source measurable: R@1 shifts from 44% to 52% on Gwilliams and from 22% to 29% on MOUS under the same fixed setting. GCB is auditable under this design: its effect collapses under random-grouping perturbations and vanishes when local evidence is attenuated in MEG or is near chance in EEG, supporting its use as a controlled source-attribution intervention. These results suggest that brain-to-language performance should be source-attributed, not merely reported.

Visual decoding from brain signals is a key challenge at the intersection of computer vision and neuroscience, requiring methods that bridge neural representations and computational models of vision. We introduce a tri-modal contrastive framework for EEG-based visual decoding that aligns EEG, visual, and textual representations within a unified latent space. Our approach follows a two-stage design. First, we pre-train an EEG encoder via masked reconstruction on unlabeled trials, learning spatio-temporal regularities that transfer robustly to downstream tasks. Second, we jointly align EEG, image, and LLM-generated textual descriptions through contrastive learning, where text supervision acts as a semantic regularizer that injects linguistic structure into the shared space without overwhelming the primary EEG-image signal. The encoder integrates subject-specific adaptation, graph-attention over channels, and temporal-spatial convolutional embeddings. On the Things-EEG2 200-way zero-shot benchmark, our framework achieves 54.1% Top-1 and 83.4% Top-5 accuracy, substantially exceeding the strongest prior baseline (32.4% / 64.0%), with paired Wilcoxon tests confirming significance (p < 0.01) over all in-subject baselines. We validate generalization on Things-MEG. Analysis reveals that compact embedding geometries (CN-CLIP) outperform much larger backbones, and that decoding aligns with established neurophysiology of visual processing. This work is a critical step towards robust, semantically-grounded visual decoding from non-invasive temporal neural signals. The source code is publicly available in https://github.com/anon-eeg/eeg_image_decoding.

The quadratic complexity of self-attention in Transformer models remains a significant bottleneck for processing long sequences and deploying large language models efficiently. For this approach, there has been significant research into Sparse Attention, and Deepseek Sparse Attention has combined various methods of creating segments of tokens to reduce the time complexity. This paper introduces a novel approach, Grammatically-Guided Sparse Attention, which constrains attention computations based on the grammatical roles of tokens. By leveraging Parts-of-Speech (POS) tags, attention masks are dynamically generated that enforce linguistically coherent connections between tokens, reducing the computational graph without sacrificing essential linguistic dependencies. Two masking strategies are proposed and evaluated: a hard mask that strictly allows only predefined grammatical interactions, and a soft mask that biases attention towards these interactions. The experiments, conducted on the SST-2 sentiment classification task using a DistilBERT-like architecture, demonstrate that Grammatically-Guided Sparse Attention maintains comparable accuracy to full attention while significantly reducing the theoretical computational overhead. Preliminary results show accuracy values of 0.8200 for hard masking and 0.8165 for soft masking, closely matching the 0.8200 of full attention, providing a path towards more efficient, interpretable, and linguistically-informed Transformer architectures.

CLI agents are the closest thing language models have to an embodied setting: the model emits commands, the terminal executes them, and the returned stream – stdout, errors, files, logs, and traces – records the consequences. We argue that this stream is a supervision signal, but standard agent RL discards it: GRPO-style training updates action tokens with sparse outcome-level rewards while ignoring environment responses already in the rollout. Failed rollouts provide little policy-gradient signal despite containing rich evidence about how the environment responds. We introduce ECHO (Environment Cross-entropy Hybrid Objective), a hybrid objective that combines the standard policy-gradient loss on action tokens with an auxiliary loss that trains the policy to predict environment observation tokens resulting from its own actions. ECHO reuses the same forward pass as GRPO, requires no additional rollouts, and turns terminal feedback into dense supervision for all rollouts. ECHO doubles GRPO pass@1 on TerminalBench-2.0: Qwen3-8B improves from 2.70% to 5.17%, and Qwen3-14B from 5.17% to 10.79%. ECHO also produces policies that better predict terminal dynamics, even on trajectories they did not generate: across held-out rollouts, it sharply reduces environment-token cross-entropy while GRPO alone barely changes it. From base Qwen3-8B, ECHO matches expert-SFT-then-GRPO performance on held-out terminal tasks without expert demonstrations, and recovers roughly half of the expert-SFT initialization benefit on TerminalBench-2.0. In some settings, the environment prediction loss alone enables verifier-free self-improvement, allowing policies to improve on unseen OOD tasks by learning only from environment interactions. Together, these results suggest that environment observations are not merely context for future actions, but a dense, on-policy supervision signal already present in every rollout.

Recent progress on long-horizon agentic tasks has been driven largely by scaling up individual agents through stronger models, better tools, and more effective scaffolding. In contrast, much less is understood about scaling out: whether multiple peer agents, all targeting the same task, can become an additional source of capability without relying on explicit role specialization or workflow orchestration. We study this question and propose AgentFugue, a collective reasoning framework built around a shared reasoning hub. As peer agents explore the same task in parallel, the hub records concise notes on what each agent has established, attempted, or ruled out, and enables each agent to selectively access what other agents have discovered in a form useful for its current search. This design turns otherwise isolated trajectories into a connected ecology of reusable intermediate reasoning without requiring centralized planning. We instantiate the hub as a plug-in communication layer, trained with supervised fine-tuning and end-to-end reinforcement learning. Across the challenging long-horizon settings we study, AgentFugue improves over strong baselines. Our results suggest that collective reasoning can turn scaling out peer agent systems into a distinct source of capability gains, rather than merely a way of spending more compute.

Large language models (LLMs) have shown strong performance in the legal domain, demonstrating notable potential in Legal Question Answering (LQA). However, unlike general QA, LQA requires answers that are not only accurate but also rigorously grounded in explicit legal authority. In statutory LQA, many questions require multi-hop reasoning across multiple legal issues, substantially increasing the risk of hallucination, thereby making accurate retrieval of supporting statutory provisions a critical prerequisite. Despite recent progress in multi-hop QA, existing approaches often rely on reasoning in natural language or retrieval without explicit query reformulation, leaving the vocabulary gap between user questions and statutory text largely unaddressed. To address this challenge, we propose Decompose-and-Refine (DaR), a statute-grounded LQA framework that tightly integrates step-wise question decomposition with parametric knowledge-based query refinement. DaR progressively decomposes a complex legal question into atomic sub-questions and generates statute-aligned parametric queries for each sub-question, enabling the selection of a single most central statutory provision corresponding to each legal issue. We evaluate DaR on KoBLEX, a Korean multi-hop LQA benchmark grounded in statutory law, using Qwen3-32B and Gemma3-27B. Experimental results demonstrate that DaR consistently improves both retrieval accuracy and final answer quality over existing approaches. Moreover, by explicitly separating sub-questions and their corresponding statutory provisions, DaR facilitates transparent, issue-level verification of complex legal reasoning processes.

Legal NLP benchmarks evaluate models on randomly split data, implicitly assuming that legal language is stationary. We test this assumption by fine-tuning four transformer encoders – XLM-RoBERTa (base and large) and their legal-domain variants – on Ukrainian court decisions from three temporal epochs defined by geopolitical disruptions: pre-war (2008-2013), hybrid war (2014-2021), and full-scale invasion (2022-2026). Each model is trained on one epoch and evaluated on all three, producing a 3x3 cross-temporal generalization matrix. Four findings emerge. (1) Forward degradation is severe: models trained on pre-war data lose up to 27.2 percentage points of macro-F1 when applied to full-scale invasion era decisions. (2) The degradation is asymmetric: backward transfer (full-scale to pre-war) is substantially more robust than forward transfer, consistent with the hypothesis that legal language is additive. (3) Legal-domain pretraining (Legal-XLM-R) does not improve absolute performance but reduces forward degradation magnitude and asymmetry. (4) Chronological continual learning eliminates catastrophic forgetting for general XLM-R: pre-war knowledge is fully retained (+1.8 to +6.2 pp) while full-scale performance gains +16.5 to +19.0 pp; reverse-chronological training causes severe forgetting. Cross-jurisdictional pretraining on Swiss Judgment Prediction data improves absolute performance but does not reduce temporal degradation magnitude, confirming that temporal drift is an intrinsic property of legal language evolution. The dataset (428K decisions across three epochs) is publicly available as a LEXTREME contribution.

Vietnamese exhibits substantial dialectal phonetic variation across Northern, Central, and Southern regions, where identical lexical items may be realized with markedly different pronunciations. Such variation poses challenges for automatic speech recognition (ASR) and remains difficult to model computationally due to the complex relationship between Vietnamese orthography and phonology. Existing approaches typically address dialect variability at the word level, assuming dialect-invariant mappings between spelling and pronunciation, which limits their ability to capture systematic phonetic differences. We propose a dialect-aware phonetic framework that explicitly models Vietnamese phonological structure and dialectal variation at both the vocabulary and decoding levels. The framework introduces a phonetic vocabulary that decomposes each syllable into structured phonetic components and maps them to dialect-specific IPA representations, together with a phonetic-structure decoder that jointly predicts these components. Experiments on the UIT-ViMD, a only-available dataset for multi-dialect in Vietnamese, show that the proposed approach outperforms various pre-trained baselines, especially matches the performance of the strongest pretrained wav2ve2-base-vi-250h across dialects while using substantially fewer parameters and no external pretraining. Code for experimental reproducibility will be publicly available upon the acceptance of this paper.

Large Language Model (LLM) interactions are typically underspecified, with users clarifying all necessary details across multiple conversational turns. Yet recent work shows that LLMs perform far worse in this multi-turn setting than in a single turn with same information being available at once, a phenomenon termed "Lost-in-Conversation." However, bridging this gap effectively remains an open problem. Here we introduce Found in Conversation (FiC), a training framework where a model teaches itself to find and recover its single-turn competence given underspecified multi-turn prompts. We develop View-Asymmetric Self-Distillation, which distills across two views of the same task information–single-turn view for the teacher, multi-turn view for the student–transferring strong single-turn behavior into weak multi-turn behavior. This requires no stronger external teacher, which is unavailable as even frontier LLMs exhibit this gap. Across model families (Llama, Qwen, Phi, and OLMo) and sizes (3B-14B), FiC recovers at least 92% of single-turn performance and reaches 100% on two Llama backbones, yielding more efficient and helpful multi-turn conversations with single-turn capabilities intact.

Large language model (LLM) agents increasingly learn through interaction, yet their training process often evolves unevenly: the policy changes with experience while the learning environment continues to provide largely fixed forms of support. As training progresses, this creates a growing mismatch between what the agent currently struggles with and what its learning process emphasizes. We call this phenomenon Agent-Learning Environment Misalignment (ALEM). We introduce SEAL, a framework for synergistic co-evolution of agents and learning environments. SEAL turns verifier-grounded execution failures into a shared diagnostic signal that coordinates two complementary forms of adaptation: it reshapes the support available in subsequent interactions and adjusts how the policy learns from collected trajectories. In this way, the learning environment follows the agent's evolving failure profile while both sides remain grounded in the same executable feedback. Across three backbones on BFCL V3, SEAL consistently improves over matched Vanilla RL, reaching 40.25% versus 30.75% with Qwen2.5-7B. A matched factorial ablation further yields a +5.00-point interaction estimate for the coupled configuration. The same trained checkpoints also show positive transfer to the evaluated BFCL V4 and $\tau^2$-bench domains. These results suggest that coordinating how an agent learns with the environment it learns from is a promising direction for self-evolving tool-use agents.

The residual update of a pre-norm Transformer layer admits an interpretation as one step of a first-order optimizer acting on a surrogate token energy, wherein the attention and MLP sublayers function as gradient oracles. Based on this observation, we build a family of optimizer-inspired Transformers (triple-momentum, Adam/AdamW, Muon, SOAP) and compare them under matched compute. In our main pretraining experiment, the triple-momentum TMMFormer achieves the lowest validation loss, outperforming the vanilla Transformer and prior architectural variants. A controlled ablation and supporting theory show that momentum, not preconditioning, is the main source of the gain. We further show that TMMFormer and other momentum-based designs reach flatter minima than the vanilla Transformer, which leads to less forgetting and better generalization.

Language models (LMs) can exhibit biases based on variations in their dialects, even in the absence of a dialect label, a behavior known as covert dialect bias. In this work, we quantify covert dialect bias in online discourse by evaluating how LMs associate stereotypical traits (derived from social psychology research on racial bias) with intent-equivalent tweets in Standard American English (SAE) and African-American Vernacular English (AAVE). While prior work shows that LMs associate more negative stereotypes with AAVE when evaluating tweets in isolation, we are surprised to find that this bias is significantly exacerbated when SAE / AAVE tweet pairs are compared side by side, a setting that more closely reflects high-impact decision making contexts in which models are used to rank candidates. The bias only worsens when dialect labels are explicitly specified. This is striking, given the extensive efforts from commercial developers to mitigate bias in their LMs. Encouragingly, we show that counterfactual fairness finetuning can mitigate covert dialect bias for some stereotypical traits, reducing average disparities when evaluating tweets in isolation, however, these improvements do not consistently hold across traits when evaluating SAE / AAVE tweets side by side. Our findings show that existing evaluation settings for covert dialect bias may underestimate its severity, specifically in contrastive settings. Additionally, overt dialect bias remains pronounced even after safety aligned finetuning, indicating that it remains an unresolved problem, and motivates the need for more robust evaluation and mitigation frameworks.

CT report generation (CTRG) requires models to summarize three-dimensional anatomical context and pathological findings from hundreds of axial slices. Existing methods typically learn a direct image-to-text mapping, providing limited mechanisms for modeling how CT evidence evolves across slices or how reports respond to controlled changes in latent lesion-related factors. We propose SliceWorld, a CT-specific world-state framework that treats an axial CT scan as an ordered sequence along the z-axis. SliceWorld encodes prefix CT evidence into factor-aware latent states containing anatomy, lesion, and uncertainty components, and projects these states into world tokens used for multi-step future-slice feature prediction, lesion-factor intervention, and LLM-based report generation. The model is first pretrained on CT slice sequences with predictive, factor-aware, and counterfactual objectives, and is then fine-tuned on paired CT-report data. Experiments on M3D-Cap and CT-RATE show that SliceWorld improves natural language generation metrics and clinically oriented automatic evaluation. Further analyses demonstrate multi-horizon future-slice prediction, measurable factor alignment, reduced-slice robustness, and selective lesion-sensitive report modulation.

Large Language Models (LLMs) have been widely adopted in conversational applications. However, their reliance on parametric knowledge limits reliability in real-world scenarios that require dynamic or domain-specific information. Retrieval-Augmented Generation (RAG) addresses this limitation by incorporating external knowledge during generation, but existing text-based and graph-based RAG methods often struggle with noisy or irrelevant contexts. In this work, we propose Structure-aware Retrieval Augmented Generation (SA-RAG), which uses tables as an intermediate structured representation to provide a compact and controllable interface that reduces noise while preserving essential information. We introduce a quality-aware table metadata generation framework that models metadata normalization and effectiveness, improving metadata quality and downstream performance. Furthermore, we explore both training-free and training-based table generation methods. Generation validation and direct preference optimization further improve table quality while maintaining semantic and structural consistency. Experiments on two noisy real-world datasets show that SA-RAG significantly outperforms existing RAG baselines. Our code is publicly available at a public repository.

Large language models (LLMs) can support scientific literature synthesis, but remain prone to hallucinated references, uneven coverage, and weakly grounded thematic organization. We evaluate whether bibliometric structure improves LLM-assisted synthesis by comparing six pipelines for generating cluster descriptions under different levels of evidence and structure. Using 100 published bibliometric analyses, we reconstruct Scopus corpora, extract human-written cluster descriptions, and assess outputs by human alignment, semantic coverage, clustering quality, graph quality, and reference grounding. Results show that LLMs produce descriptions semantically close to human-written ones, but are unreliable when asked to infer bibliometric structure from scratch. Performance improves when bibliometric algorithms define the clusters and the LLM interprets them. Overall, LLM-assisted bibliometric synthesis is most promising as a hybrid workflow in which algorithms provide auditable structure and LLMs generate readable descriptions.

Research on harmful meme detection has garnered significant attention, resulting in the development of numerous datasets and methods. However, progress in detecting Chinese harmful memes lags considerably, primarily due to two challenges: first, accurately assessing a meme's harmfulness depends heavily on understanding deep cultural context; second, many memes are semantically ambiguous, making harmfulness highly subjective. To address these issues, we focus on the interpretable detection of Chinese harmful memes by constructing the first Chinese harmful meme explanation dataset, Ex-ToxiCN-MM. This dataset offers opposing interpretations, categorized as "harmful" and "non-harmful", for each meme, aiming to rigorously evaluate a model's ability to discern and comprehend ambiguous, culturally grounded content. We built a specialized knowledge base of Chinese cultural concepts and offensive vocabulary to supply models with essential prior knowledge (C-HarmKB). To address the ambiguity and lack of background knowledge in meme attribution, we have developed a comprehensive attribution analysis framework, RIKE, which includes an Attribution Knowledge Enhancement module (AKE) and a Relative Intent Reasoning module (RIR). Extensive quantitative and qualitative experiments demonstrate that our method outperforms mainstream baseline models across multiple metrics in the task of attributing harmful memes in Chinese. The code, Ex-ToxiCN-MM dataset, and Chinese Harmful Semantic Knowledge Base (C-HarmKB) involved in this study have been open-sourced at https://github.com/wimiw123/Ex-ToxiCN-MM

Current high-performing intracortical speech neuroprostheses achieve low word error rates but typically rely on external language models during inference, increasing memory, computation, and latency. In this work, we investigate whether meaningful character-level decoding is achievable without such models. We propose an end-to-end Conformer-based neural decoder trained directly on intracortical recordings from a participant with amyotrophic lateral sclerosis (ALS). Without any external language model, the system achieves a character error rate (CER) of 23.80% on held-out validation data. Analysis shows that performance variability is driven by inter-session signal degradation, while dominant errors arise from incorrect word boundary segmentation. These results demonstrate that effective character-level decoding is possible in a fully end-to-end framework, providing a strong neural signal for downstream linguistic processing.

Lexical gaps are words that do not exist in certain languages. They pose challenges for building multilingual lexical resources, for machine translation, and for cross-lingual transfer. Existing lexical gap detection relies on human judgments or fixed conceptual taxonomies. We propose a data-driven framework for identifying cross-lingual lexical gaps. We extracted contextualized embeddings from Korean-English bilingual LLMs for Korean-to-English and English-to-Korean translation pairs. Combinations of LLMs, embedding types, dimensionality, and orthogonal transformations across 100 train-test splits yielded 4000 distinct embedding spaces in each source language. In each space, we computed the semantic similarity between each source word and its nearest neighbor in the target language, and compared their distribution for gap words versus non-gap words. In 94% (Korean-to-English) and 97% (English-to-Korean) of embedding spaces, gap words showed weaker cross-lingual semantic alignment than non-gap words. Logistic classifiers trained on unaligned embedding spaces can reliably separate gap words from non-gap words, achieving AUCs of 0.81 (Korean-to-English) and 0.76 (English-to-Korean) and retrieving 18/19 Korean and 26/27 English gap words. This approach provides a language-agnostic and taxonomy-free method for scalable lexical gap identification.

Post-training of large language models optimizes only parameters, while inference-time procedural scaffolds are typically designed independently of parameter training. This disconnect makes it difficult to automatically acquire and internalize complex strategies. We propose scaffold-mediated post-training: procedural scaffolds are organized into an evolvable graph structure that co-evolves with model parameters through discovery, distillation, and dynamic recompilation. We instantiate this paradigm as Skill Training. On FeatureBench, automatically discovered skills improve the passed rate by 8.1pp, and after progressive distillation the model still achieves a 27.7% passed rate without any external scaffold (distillation retention rate 85.2%, defined as post-distillation / with-skill passed rate), significantly outperforming standard SFT on the same data.

Traditional sentiment analysis (SA) models, while effective for polarity classification, provide limited insight into the rhetorical, ideological, and framing dimensions of political discourse – dimensions that are central to research in the social sciences and humanities (SSH). In this paper, we present a comparative study of RoBERTa-based sentiment analysis and an LLM-based multi-dimensional framing analysis platform applied to a corpus of 50 political news articles from 17 international media outlets. The results reveal a critical limitation we term neutral collapse: RoBERTa classifies 70% of articles as neutral, effectively flattening substantively rich political content into an analytically uninformative category. We find that 23% of neutral-classified articles exhibit negative probability scores above 0.30. By contrast, the LLM-based approach captures political bias direction and intensity, sensationalism, emotional appeal, and political framing – yielding multi-dimensional analytical outputs aligned with SSH epistemologies. We argue that for political media analysis, traditional SA alone is insufficient, and that LLM-based multi-dimensional frameworks offer a more epistemologically adequate computational lens for SSH research needs.

Large Language Models (LLMs) are temporally overexposed: trained on vast contemporary corpora, they encode present-day concepts that make them unreliable narrators of the past. We present TimeCapsule, a 1.2B-parameter LLaMA-style causal model trained exclusively on Victorian texts (1800-1875) as an epistemologically isolated generative archive. Quantitative evaluation shows a 45.4% perplexity reduction over a GPT-2 baseline on held-out Victorian prose, while larger contemporary causal models achieve lower raw perplexity through broader pretraining but lack temporal isolation. TimeCapsule exhibits computational sensemaking, generating historically plausible analogical explanations for unfamiliar modern concepts (e.g., describing a computer as a "hypertrophied lung"). A qualitative hermeneutic probe with two humanities scholars revealed a crisis of authenticity, as both misclassified approximately 40% of genuine Victorian excerpts as machine-produced. We argue that structural ignorance of the future transforms hallucinations into interpretive probes of nineteenth-century ontologies.

LLMs are increasingly used with external knowledge sources like the internet. Do they weigh information appropriately – updating more for reliable sources (source discernment) and more when claims bring priors closer to the truth (truth discernment)? We formalize this as information discernment and introduce Learn2Discern (L2D), an experimental framework and benchmark grounded in three normative axioms with interpretable metrics. To establish external validity, a pre-registered, quota-matched user study (n=299) confirms that real LLM users endorse all three axioms and report that violations reduce their trust and usage intent. Across 13 models and nearly 670K trials, we find consistent failures across both dimensions: models perform near chance on source and truth discernment, rely on source popularity twice as much as source reliability, and update roughly equally whether a claim improves or worsens their position relative to the ground truth. Models integrate external knowledge most effectively on datasets where their priors are already the most accurate. Newer and larger models improve truth discernment but not source discernment, a blind spot that model complexity does not address. We identify simple inference-time interventions that improve both forms of discernment. We release our dataset and survey as a testbed for a core alignment property that scales in importance as LLMs replace traditional search.

Online travel platforms generate vast volumes of user-generated hotel reviews, offering rich opportunities to understand traveler experiences at scale. However, transforming unstructured textual feedback into structured, actionable insights remains a challenging task. This paper presents SentimentLens, a scalable analysis system based on Aspect-Based Sentiment Analysis that performs knowledge extraction from unstructured hotel reviews and organizes them into interpretable service categories. SentimentLens integrates aspect term extraction, aspect sentiment classification, semantic category assignment, and multi-level analytical modules to support region-level, hotel-level, and category-level evaluation. The system is designed to operate across different geographic contexts and hospitality settings. To demonstrate its practical utility, we apply SentimentLens to a large real-world dataset of over 10,000 publicly available hotel reviews. Through extensive analysis, the framework reveals how traveler sentiment varies across regions, service categories, and hotel archetypes. We further implement a cross-modal reconciliation of textual sentiment and numerical ratings to identify latent operational conflicts, structural inconsistencies in service quality, and high-impact improvement opportunities using importance–performance and entropy-based analyses. The results show that SentimentLens effectively transforms large-scale unstructured reviews into actionable intelligence, supporting data-driven decision-making for hospitality management and tourism policy. While demonstrated using a national case study, the proposed system is generalizable to other destinations and review-driven service domains.

The Cognitive Categorical Transformer (CCT) is a 306M-parameter architecture that augments a pretrained GPT-2 Small backbone with cognitively grounded components derived from category theory and several inspirations from cognitive science. Under a matched-step protocol (215,000 optimizer steps, matched data, matched optimizer and schedule) on WikiText-103, CCT reaches 21.27 validation perplexity, compared with 24.19 for an identically fine-tuned GPT-2 Small baseline. The architecture therefore contributes a 2.92 PPL (12% relative) reduction beyond what in-domain fine-tuning alone provides. A retrain-from-scratch ablation that holds GT-Full simplicial message passing bypassed across the entire seven-phase activation schedule reaches 23.72 PPL, localizing 84% of the architectural improvement (2.45 of 2.92 PPL) to GT-Full. We present the first ablation-validated evidence that simplicial message passing improves language-model perplexity at the 306M-parameter scale on WikiText-103. Published GPT-2 Large reaches 22.05 zero-shot PPL on WikiText-103 with 6.2x more parameters than GPT-2 Small; this paper treats that number as an external published reference, not as the architectural benchmark. Three negative results on consistency-style categorical priors (sheaf smoothing, adjunction round-trip, curvature regularization) and the joint structural-prior result for GT-Full and PrecisionWeightedPP together support an empirical pattern termed the *structure/consistency distinction*, in which categorical priors that add new topology improve language modeling and those that enforce a consistency identity do not.

We consider the conversion of musical recordings into human-readable sheet music annotated with timestamps. Such output lets a listener clearly visualize rubato (temporally expressive playing), a learner diagnose ensemble precision and timing choices against the written music, and a musicology scholar compare performance styles across recordings of the same work. We introduce (1) a prompt-conditioned encoder-decoder model, named Rubato, trained to output (2) a new textual representation for polyphonic music, named InterMo, which we designed for compatibility with sequence-to-sequence training. Our experiments demonstrate that Rubato produces timestamped piano sheet music from audio with higher notational accuracy than the best existing approaches, which are based on cascades. We find that even if the cascade is given ground-truth MIDI instead of audio, Rubato performs better, suggesting that the ceiling of existing approaches is primarily representational, not acoustic. Further, because Rubato is trained on several related tasks (with prompts), it competes with or outperforms the best single-task systems on related but simpler tasks like MIDI note grounding and beat/downbeat detection. A demo is available at https://nctamer.github.io/rubato-transcription .

Chain-of-thought (CoT) reasoning is useful for monitoring language models only when the reasoning trace faithfully reflects the computation that produces the final answer. However, models can rely on prompt-to-answer shortcuts that bypass the CoT, making the visible reasoning trace misleading even when it appears plausible. We study CoT faithfulness through a structural information-flow perspective: faithful reasoning should route answer-relevant information through the mediated path from prompt to CoT to answer, rather than through a direct prompt-to-answer shortcut. This perspective yields a task-agnostic framework based on three complementary properties, sufficiency, completeness, and necessity, which we instantiate with entropy-based, masked-KL, and gradient-based diagnostics. We show that these metrics recover externally judged faithfulness differences in hinted reasoning, and identify a low-entropy failure mode of KL-based diagnostics where gradient-based measures remain more stable. Building on this analysis, we introduce update-time interventions for verifier-based on-policy RL, including attention masking, backward-only gradient masking, CoT gradients, and adversarial perturbations of prompt representations. Across hinted arithmetic, reward-hackable code repair, and DAPO-Math models trained without hints but evaluated under wrong-hint injection, our interventions shift behavioral and structural indicators toward stronger CoT mediation. In particular, they make shortcut and reward-hacking behavior more transparent in the CoT and improve task-agnostic faithfulness metrics, while in some settings also reducing wrong-hint susceptibility. Our results suggest that controlling information flow during training is a practical route toward more faithful and monitorable CoT reasoning. Code is available at https://github.com/safety-research/faithful-cot.

Frontier language models are trained to exhibit stable programming assistant personas, yet those personas may drift over long agentic coding sessions in production. After a long real-world coding session, a model that hedges preferences in a fresh chat ("I don't have preferences") may instead express them directly ("Python – the feedback loop is instant..."). Existing persona-stability studies focus on scripted chat of tens of turns, leaving long production coding sessions involving thousands of tool-using agent steps, repeated compaction, and hours of interaction largely uncharacterized. We introduce ContextEcho, a benchmark and reusable harness with a 25-probe identity suite, a snapshot-then-probe protocol that evaluates conversation state without perturbing the main session, judge-based and judge-free metrics, and a corpus of 57 long agentic-coding sessions collected with consent and redacted for release across Claude Code and Codex CLI. We use one representative session for a 23-model cross-provider study and all 57 sessions for corpus-scale trajectory analysis across frontier and open-weight models. Results show that persona drift occurs across providers and emerges early in real agentic interaction, while in-session compaction neither reliably resets nor amplifies it. A single-shot anchor restores the original assistant persona, though placebo controls suggest a generic register-reset effect. On same-task continuations, drifted prefixes improve tool-call fidelity – which we attribute to retained task context rather than drift itself. In tool-free chat, however, the same prefixes break output-format contracts and inflate output length. These findings motivate ContextEcho as an open-source framework for auditing persona drift in long coding sessions without retraining.