Papers

Person-level psychological assessment requires aggregating meaning across many messages from the same individual, a task that document-level training objectives were not explicitly designed for. We present a systematic, empirical comparison between architecturally matched traditional (a) base-transformers and (b) document-tuned-transformers (further contrastively fine-tuned at the document-level, sometimes referred to as "sentence transformers") under otherwise identical conditions. Comparing layer-wise and overall performance across two longitudinal mental health and psychological datasets, we find document-tuned models demonstrated a consistent improvement over base representations (increase in Pearson r of 13.4%, p=.015). Robustness analyses revealed document-tuned models remained more accurate under perturbations to word deletion, synonym replacement, typo injection, and back translation. Further, hedged language (e.g., `usually') was more characteristic of outcomes in document-tuned embeddings while abundance (e.g., `lot') was more characteristic of base-transformers, suggesting document-tuned models may better capture uncertainty. These results suggest representation choice impacts mental health prediction, document-tuned models often being more adept.

Evaluating Multimodal Large Language Models (MLLMs) in Chinese Cultural Heritage (CCH) requires fine-grained reasoning over visual, textual, stylistic, and historical clues. However, existing CCH benchmarks mainly emphasize final-answer accuracy, while the accuracy and completeness of reasoning processes remain underexplored. To address this gap, we introduce CulMind and CulMind-R: a high-quality benchmark for multimodal CCH covering 50 tasks from collections of more than 100 museums, and a 24-task reasoning subset that adaptively defines task-specific dimensions for reasoning process evaluation. To evaluate reasoning quality, we propose ReaScore, a task-adaptive metric that evaluates reasoning by automatically weighting task-relevant dimensions. Experiments on 14 leading MLLMs reveal a substantial gap between answers and reasoning, especially on challenging tasks. Further analysis shows that task-adaptive dimension selection and weighting better align evaluation results with expert judgments. Overall, our benchmark and metric support a more expert-aligned assessment of CCH understanding and offer a transferable reference for broader evaluations of cultural heritage. We publicly release the data, code, and evaluation scripts at https://github.com/ZevTsao/CulMind to facilitate reproducible research.

Much work on the cognitive foundations of AI has focussed on comparisons between the ways in which Large Language Models (LLMs) and humans process information and represent it. One aspect of this comparison involves determining the extent to which LLMs can achieve or surpass human performance on a variety of cognitively interesting tasks. A second explores points of convergence and divergence between LLM and human systems for processing information. Here, I consider some recent research that has addressed both issues in two informational domains. The first is the representation of linguistic knowledge. The second is real world reasoning and planning. While LLMs frequently achieve impressive levels of performance and fluency on linguistic applications, they tend to handle linguistic content in ways that are distinct from human processing. They are also, for the most part, less efficient than humans in learning and generalisation for reasoning tasks.

The open-source ecosystem on GitHub lacks a systematic hierarchical taxonomy of software repositories. GitHub Topics, the dominant organizational mechanism, is flat, inconsistent, and covers only 67% of projects. We present ATLAS, the first framework that automatically constructs a hierarchical taxonomy for software repositories and classifies projects into it end-to-end. By combining LLM global knowledge with real repository distributions, ATLAS proposes meaningful splitting dimensions and iteratively corrects those that fail to accommodate real projects. A Designer Agent proposes splitting dimensions while a Classifier Agent assigns repositories; a self-corrective refinement loop uses classification failures to drive dimension revision through escalating strategies. We evaluate ATLAS on 54,387 GitHub repositories against six baselines spanning four paradigms, two downstream tasks, and three model families. On a stratified 2,001-repository benchmark, ATLAS achieves a Taxonomy Quality F-score (TQF) of 83.13%, outperforming the best baseline by 15 percentage points (on the full 54k corpus the approximate TQF is 73.0%, a gap driven by Path Granularity's all-or-nothing scoring on longer paths rather than lower classification accuracy). It is the only method to simultaneously achieve high structural quality and high practical applicability. On downstream tasks, ATLAS enables alternative discovery with P@1 = 85.71%, surpassing even human-curated lists (62.34%), and achieves the highest P@1 for repository retrieval. The taxonomy further reveals structural ecosystem trends that are difficult to obtain from flat tags or similarity methods: the shift from libraries to AI/ML applications (now 61% of newly community-adopted projects) becomes visible only through hierarchical, type-based categorization. An interactive taxonomy explorer is available at https://atlas-taxonomy.netlify.app/

AI-mediated answer systems increasingly determine how brands and organizations are represented to users. Existing approaches reduce visibility to mention rate or citation frequency. This paper argues that aggregate metrics are insufficient because entities exhibit systematically different AI visibility error profiles. We introduce Per-Entity Bias Mapping (PEBM): a ten-dimensional framework distinguishing raw from verified mentions. Three failure modes are identified: (1) underrepresented entities suffer invisibility due to weak knowledge graph presence; (2) large entities suffer the Brand Hallucination Paradox – model familiarity creates stronger surfaces for plausible but incorrect completions; (3) CEE entities face a structural infrastructure gap across knowledge graphs, NER, and entity linking. A fourth dimension, Parametric-Retrieval Lag Asymmetry, describes divergence between retrieval-augmented and parametric memory update cycles. A full-scale empirical study (n=100 Hungarian B2B entities, 1,400 probe runs, 2,062 sources) finds Tier 1 brands produce 52.69% fabricated citations versus 37.87% for Tier 3 entities (+14.82 pp; p=1.67e-11), supporting the Brand Hallucination Paradox. Regulatory-framed queries elevate fabrication to 56.77% versus 37.59% baseline (+19.2 pp). We identify rejection-induced confabulation escalation: agentic quality filters function as hallucination accelerators in compliance contexts. We introduce ghost cartography as a unifying mechanism: entities in sparse latent regions produce confident output interpolated from neighboring dense regions, yielding a two-dimensional confabulation space (fabricated presence vs. frozen representation).

Large language models (LLMs) have shown potential for reference-free span-level translation quality estimation (QE), yet existing approaches based on Multidimensional Quality Metrics (MQM) typically rely on fixed rubric configurations shared across translation instances. However, translation instances often differ substantially in error complexity, ambiguity, and required evaluation granularity, making static rubric allocation suboptimal for span-level error detection. We find that larger MQM subtype spaces improve error coverage but also introduce more false positives, while different translation instances prefer different rubric granularities, suggesting that evaluation spaces should be allocated dynamically for each case. Therefore, we propose a case-specific dynamic rubric framework that adaptively constructs MQM evaluation spaces for individual translation instances. Unlike methods that generate fully free-form rubrics, our framework remains grounded in the predefined MQM taxonomy while dynamically selecting suitable subtype spaces and evaluation granularity for different cases. Experiments on span-level QE benchmarks from the Conference on Machine Translation (WMT) across multiple model scales demonstrate that the proposed framework consistently improves F1 and Matthews Correlation Coefficient (MCC).

Collaborative Problem Solving (CPS) is a core skill in education, where the process of peer interaction is highly important. However, existing educational dialogue datasets mostly focus on classroom instruction or tutoring (i.e., teacher/tutor-student interaction), yet datasets centering small-group, student-student interaction are limited. This thus leaves research with limited resources for studying how students interact, coordinate, and solve problems together in real educational settings. To address this, we introduce PeerMathDial, the first dataset of peer CPS dialogues collected from authentic middle school math classrooms. It contains 55 dialogues from 27 students, totaling 6,406 turns. To facilitate research on CPS discourse analysis, we further build a corpus-grounded dialogue act taxonomy assisted by LLMs. Using the dataset and the dialogue act taxonomy, we demonstrate the practical applications of PeerMathDial across three use cases. First, we track how dialogues evolve over time and measure the impact of teacher interventions. Second, we align dialogue actions with student surveys to reveal the connection between students' traits (e.g., confidence, leadership) and their actual behaviors. Third, by evaluating LLMs on dialogue act prediction, we glimpse at the potential of LLMs for student simulation in educational applications. Our dataset and source code will be released to the community.

Agentic retrieval-augmented generation (RAG) systems combine iterative reasoning loops, query decomposition, and adaptive retrieval to tackle multi-hop question answering. However, the contribution of each component remains poorly understood, particularly under resource-constrained settings using only local language models. Many agentic designs add adaptive retrieval routing and deeper retrieval loops on the assumption that the added complexity helps. To test whether it does, we run a controlled ablation study of a full agentic RAG pipeline evaluated on 5,000 questions from the HotpotQA distractor development set using a local 7B parameter model (Qwen2.5-7B-Instruct). Our full pipeline achieves EM=53.2% and F1=61.6%, compared to a single-pass dense-retrieval baseline of EM=43.1% and F1=54.0%. Across eight ablation conditions, we find that: (1) fixed hybrid retrieval via reciprocal rank fusion consistently outperforms rule-based adaptive routing (+1.8 EM, +1.9 F1), as the routing heuristic over-routes to BM25 by firing on named entities present in nearly all multi-hop sub-questions; (2) two retrieval iterations over the decomposed sub-questions capture 95% of the gains of five, with no meaningful benefit from deeper loops; and (3) query decomposition and cross-encoder reranking each contribute statistically significant but smaller gains (p<0.01 and p<0.001 respectively). Taken together, on a fixed local-model budget, the simpler and fixed choices turn out to be competitive with or better than their adaptive versions: most of the gain comes from running a short retrieval loop, not from adaptive routing or from many iterations. We use no proprietary APIs or large-scale compute.

Detecting hallucinations in clinical text is increasingly framed as an explainability problem: systems should not merely flag an unreliable response but point to the offending span. Architectures built around knowledge-graph (KG) triple decomposition are marketed for exactly this auditability, yet their localization ability is typically assumed rather than measured. We introduce MedHal-Loc, a benchmark and metric for localization faithfulness – whether a detector's top-ranked error unit actually overlaps the erroneous span. The controlled subset comprises 300 PubMedQA-derived statements with single, span-level errors injected across four localizable types (entity substitution, relation error, mechanism misattribution, invention), yielding gold spans by construction; a complementary natural subset documents that real hallucinations are dominated by diffuse conclusion-flips that resist span localization (a human expert accepted 1/18 candidate spans). Evaluating four fine-grained paradigms, we find that NLI-per-clause, consistency-per-sentence, and the dedicated span detector FAVA all localize well above chance, whereas an elaborate KG-triple pipeline localizes no better than chance (+3.3pp, n.s.), bottlenecked by ~59% entity-extraction coverage – despite competitive detection F1 (0.609). Detection competence does not imply faithful localization; architectural explainability must be validated, not presumed.

Large language models have strong potential for use in intelligent tutoring systems, but they often fail to follow effective pedagogical strategies, such as guiding students without revealing final answers. We study the application of a two-stage alignment pipeline for math mistake remediation, combining supervised fine-tuning on tutoring dialogs with Direct Preference Optimization on synthetic preference pairs. We construct a dataset that integrates existing tutoring corpora with synthetic data generated along pedagogical dimensions, such as scaffolding and factuality, and study different input configurations that incorporate solution correctness and gold answers. Experiments show that this approach improves both factual accuracy and pedagogical quality over base models and existing tutoring models. Human evaluation further indicates that our best model is competitive with a strong proprietary baseline, while providing additional benefits in terms of openness, transparency, and reproducibility. Our results highlight the effectiveness of preference-based pedagogical alignment, while also revealing challenges in reliably evaluating tutoring quality.

How do large-scale economic transformations shape cultural production? We address this question by combining computational linguistics, econometrics, and formal modelling, using French drama as a well-documented empirical laboratory. Applying latent Dirichlet allocation to a corpus of 1,215 theatrical texts published between 1700 and 1900, we show that aristocratic discourse centred on sovereignty and political authority was gradually displaced by bourgeois and household economic themes as French capitalism developed. Bayesian vector autoregressive models with max-share shock identification suggest a temporal shift in the literary response to economic shocks: bourgeois everyday-life themes reacted to GDP shocks in the eighteenth century, whereas household-economic concerns became responsive only after 1820, amid accelerating industrialisation. A discrete-choice model shows that peer effects among authors and sensitivity to prevailing economic conditions can jointly account for these dynamics. Monte Carlo simulations reproduce the observed historical trajectory with reasonable fidelity. These findings offer a quantitative framework for understanding how economic transformations propagate into cultural production through identifiable social mechanisms, contributing to the study of cultural evolution and the long-run relationship between institutions and literary discourse.

This paper evaluates the performance of twelve Small Language Models (SLMs) on Arabic natural language processing tasks. The study introduces a benchmark of 240 Arabic test items distributed across eight domains and ten language skills, covering both comprehension-oriented and generation-oriented tasks. All models were evaluated under a controlled zero-shot setting using a standardized Arabic-only prompt template. Model responses were assessed through a multi-model LLM-as-a-judge framework involving GPT-4.1 Mini, Claude Haiku 4.5, and DeepSeek-Chat, with scores aggregated across judges and analyzed by task, skill, and model family. The results show that Gemma 3 (12B) achieved the highest overall score (4.548/5), followed by Aya and C4AI Command Arabic. The observed results suggest that model size alone does not explain Arabic SLM performance. Models with stronger Arabic alignment and more reliable instruction-following behavior tended to perform better across tasks. Common failure patterns among lower-performing models include prompt leakage, hallucination, language drift, incomplete generation, and weak task adherence. Overall, the benchmark provides a structured reference for evaluating compact Arabic language models and supports future work on efficient, reliable, and culturally appropriate Arabic AI systems.

Automated radiology report generation (RRG) has gained increasing attention because it can reduce the heavy workload of clinical report writing. However, most existing methods mainly optimize for natural language generation (NLG) metrics that focus on language fluency, while providing little control over clinically important factors such as precision and recall. As consequence, generated reports may be fluent but not well aligned with different clinical needs. To address this challenge, we propose a reinforcement learning framework for precision recall controllable RRG, where a control parameter explicitly adjusts the trade-off between clinical precision and recall during inference. This design allows the model to flexibly generate reports according to different clinical requirements. To ensure clinical correctness, we introduce a clinical reward into the training objective, which helps improve clinical efficacy (CE) beyond standard language-based optimization. In addition, we apply a group-relative training strategy that normalizes rewards within each training group, reducing reward variance and improving training stability. Extensive experiments on the MIMIC-CXR dataset show that our method consistently outperforms state-of-the-art approaches in both NLG and CE evaluation metrics, while providing reliable control over the CE precision recall trade-off.

Nowadays, large multilingual translation models demonstrate impressive translation capabilities in the machine translation benchmarks. This raises a practical question to the developers: is it worth developing translation models specialized for a particular language pair if you only need to support that language pair? To give an anecdotal answer to this question, we develop a family of small language models (0.8B, 1.4B, 3.3B, and 7B parameters) specialized for Japanese-English bidirectional translation. We employ a two-stage supervised fine-tuning approach followed by Multi-Objective GRPO (Ichihara et al. 2025) to train models on synthetically generated parallel corpora. We evaluate our models on WMT and real-world translation benchmarks across business, legal, medical, financial, and patent domains. While multilingual models achieve strong performance on WMT benchmarks, our compact models outperform them on real-world benchmarks, suggesting the practical utility of developing specialized translation models even in the era of large multilingual models.

This work examines sexualised AI-generated English-speaking voices offered by a popular commercial platform. New technologies may enable sexual empowerment and greater diversity in gender expression, yet toxic masculinity, heteronormativity, and the abuse of women and LGBTQ+ people remain pervasive online. Drawing on a Feminist HCI perspective, we examine how commercial voice AI systems reproduce and circulate particular performances of gender. We conducted a listening experiment with a diverse group of listeners, combining quantitative adjective selection, qualitative free-text responses, and acoustic analysis. Participants evaluated male- and female-coded voices presented with either sexualised scripts or neutral text. Results reveal a narrow range of gender expression, largely binary and heteronormative. Female-coded voices are more frequently described using sexualised and submissive terms, while male-coded voices are more often associated with dominance and positive traits.

Large language models (LLMs) are increasingly used to communicate and explain scientific concepts, yet their tendency to hallucinate poses significant risks in this high stakes use-case. Prior scientific hallucination evaluation work remains largely restricted to the biomedical domain, treats hallucination as a binary task, and has not examined the growing family of scientifically fine-tuned LLMs. We address these gaps with SciFactCheck, a benchmark of 2,500 prompts across five scientific domains, paired with a modular evaluation framework targeting three factuality hallucination types: unverifiability, overclaim, and attribution. Using a controlled minimal-pairing design, we evaluate 18 LLMs by comparing each scientifically fine-tuned model against its general-purpose base. Our results indicate that 1. Scientifically fine-tuned models exhibit degraded factual reliability across all hallucination types and scientific domains, and 2. Fine-tuned models are internally less confident yet linguistically more assertive. A human pilot study further reveals that current fact-checking tools show only modest agreement with expert judgments on scientific content, and that defining scientifically check-worthy claims remains contested even among human annotators. Our findings fundamentally challenge current methods of domain-specific fine-tuning for factuality and call for developing improved verification infrastructure for scientific content.

Large language models (LLMs) store and recall factual knowledge, yet the precise mechanism of how entity representations are transformed to enable specific attribute retrieval remains underexplored. In this work, we investigate this mechanism through the lens of an "attribute-computation path"-a sequence of computational steps over the entity representation required to elicit a target attribute. We then propose an iterative patching protocol to identify a minimal subset of layers necessary for this computation. Applying our method to LLaMA 3.1 8B and Qwen3 8B, we find that these paths are non-contiguous, often skipping layers, and that models possess multiple, functionally-equivalent paths for the same entity and fact, highlighting a high degree of redundancy in attribute computation. This implies that knowledge computation is highly distributed, potentially explaining the localization-editing mismatch and suggesting that knowledge storage and retrieval in LLMs is far from being well understood.

Voice reconstruction using Text-to-Speech (TTS) offers a communication method for people with speech disorders, which aims to retain their speaker identity while improving intelligibility. Previous work generally relies on Mean Opinion Score (MOS) to evaluate naturalness and speaker similarity, but this has limited sensitivity and reliability. We propose an evaluation framework with subjective and objective components. Subjectively, we evaluate perceived intelligibility and speaker identity using Best Worst Scaling (BWS) with situational framing. Objectively, we demonstrate that standard measures fail to predict reconstruction success for highly unintelligible speakers, so we introduce a novel dual-reference distributional measure to assess the trade-off between intelligibility and speaker identity. By evaluating the output of 17 zero-shot TTS systems for 193 speakers, we show that our framework provides a reliable and task-aligned approach for assessing voice reconstruction.

Automatic Speech Recognition (ASR) systems, despite achieving remarkable accuracy in general-purpose domains with native speech (L1), struggle in domains like Air Traffic Control (ATC) due to strong channel noise, a presence of non-native (L2) English accents, and data scarcity. We propose a synthetic data generation pipeline with acoustical properties simulations specifically designed to address this lack of real data to improve recognition accuracy in the ATC domain. Our approach leverages a combination of neural generation techniques, including Text-to-Speech, Voice Conversion, L2-to-L1 accent conversion, and a novel controllable L1-to-L2 accent conversion framework built to simulate accented speech. Our experiments with the Whisper model on the ATCO2 corpus demonstrate that fine-tuning with either synthetic data alone, or a mix of real and synthetic data, significantly improves the word error rate over out-of-the-box and real data only baselines respectively.

Deep neural network-based automatic speaker verification (ASV) systems achieve impressive performance but their embedding representations remain opaque, lacking a structured and perceptually verifiable explanation of the vocal characteristics they encode. Existing approaches either require annotation of speaker attributes or introduce alternative representations whose interpretability is unvalidated with listeners. We propose Listenable Interpretable Speaker Embeddings (LISE), a label-free framework that decomposes pretrained speaker embeddings into a small set of components. This decomposition yields a structured representation that supports the analysis of what information has been encoded by speaker embeddings. LISE preserves ASV performance with negligible EER degradation on x-vector and ECAPA-TDNN. Crucially, the interpretability of these components for human listeners is demonstrated through listening experiments, where participants distinguished speakers with 83.9% accuracy.

Reinforcement learning for multi-step LLM agents often relies on scalar rewards that indicate success but cannot explain why a trajectory is good or bad. Rubric-based rewards improve interpretability through natural-language criteria, but existing methods share two limitations: they score at the trajectory level, offering no guidance for individual steps; and their scorer is closed-source and static, so it cannot adapt as the agent evolves during training. We propose ARCO (Adaptive Rubric CO-evolution), which generates a per-step rubric and predicts a rubric-conditioned step-level reward for each action, and continually updates this rubric model on on-policy rollouts so that its criteria and scores co-evolve with the agent's improving behavior. Across HotpotQA, 2WikiMultiHopQA, and MuSiQue with two open-source backbones, ARCO achieves the highest EM in all settings over outcome-, rubric-, and process-reward baselines, and analyses show its rubrics are step-specific, robust to design choices, and useful for diagnosing agent behavior. Code and data are available at https://github.com/zihangtian/ARCO.

Rejecting inputs outside the defined in-distribution (IND) service scope is critical for large language model (LLM) services, where unsupported requests should be filtered before full generation. Existing out-of-distribution (OOD) detectors often rely on final outputs or final-layer representations, leaving unclear where service-boundary signals are most clearly encoded inside the model; they also lack a theoretical guarantee for held-out inputs. In this paper, we introduce SCOPE (Sequential Conformal OOD Probing and Evaluation), a framework that selects a readable hidden layer, constructs a conformal gate with IND calibration, and uses a supermartingale e-process to certify persistent service-boundary evidence. Experiments across multiple LLM backbones and six carefully designed boundary conditions show that SCOPE improves gate-level rejection over standard final-layer detectors, while revealing how different OOD boundaries take different geometric forms in hidden space.

Retrieval heads, attention heads that copy information from earlier context to the current position, have been proposed as the mechanistic substrate for long-context recall. Rotary position embeddings (RoPE) rotate queries and keys by frequencies decaying with a base hyperparameter theta, and a natural hypothesis is that this rotation either prevents retrieval heads from forming or degrades their function. We test both across four open-weight 7-8B models spanning multi-head and grouped-query attention and a 100x range of theta, using paired-seed needle-in-a-haystack tests, layer-clustered permutation, and causal head-masking. (i) Retrieval heads are causally necessary: masking the 87 detected heads in OLMo-2 collapses recall from 1.00 to 0.00, while masking matched random heads has no effect; this replicates in Qwen. (ii) Higher theta does not reduce retrieval-head count (LLaMA-3.1 at theta=500K has 47 heads vs LLaMA-2 at theta=10K with 42), refuting the prevention hypothesis. (iii) The norm-utility relation is family-specific and significant in opposite directions (Qwen d=-0.49, OLMo d=+0.50, both significant; LLaMA null); since OLMo and LLaMA-3.1 share theta=500K yet differ, the effect is not theta-driven. (iv) Building on Chiang and Yogatama (2025), a controlled patch shows that zeroing the lowest-frequency RoPE dimensions of retrieval heads degrades recall dose-dependently (1.00 to 0.18 when 32 of 128 dimensions are zeroed, vs 0.98 for random dimensions); the effect is head-specific and task-specific. The causal variable is RoPE frequency, not norm-utility. The direction holds in all five models patched (OLMo-2, Qwen2.5-7B/14B, Gemma-2, Mistral) across four lineages and two scales. We do not claim cross-model magnitude. Code and a paired-seed harness are released.

Advances in deep learning and end-to-end Automatic Speech Recognition (ASR) have enabled robust multilingual models, but evaluation metrics remain limited in assessing accuracy. Efforts to improve or replace the common metric Word Error Rate (WER) often focus on English, leaving evaluations for low-resource languages under-explored and hindering fair cross-lingual comparisons. We present OpenWER, an open-source implementation that improves WER robustness through language-specific normalisation and compound word detection. A token-based Levenshtein alignment preserves complementary metrics and allows metadata embedding for granular accuracy scores. Our analysis of 52 languages shows absolute WER reductions of up to 25% compared to common libraries. OpenWER contributes to fairness in ASR research by increasing the reliability of WER across diverse languages and enabling more comprehensive accuracy evaluations.

Psycholinguistics studies show that human readers fall for coherence illusions: an incoherent discourse can seem coherent simply because a distractor matches what comes next. We investigate whether Dutch language models (6 monolingual and 4 multilingual) show the same behavior on texts that link back to earlier context with words such as 'again' and 'too'. First, we find that surprisal at the critical word tracks human acceptability judgments and eye-tracking data. Models are more surprised by incoherent continuations, but a matching distractor in the prior context reduces this surprisal. Second, attention entropy identifies heads that behave differently under coherence vs. incoherence. We find that ablating these heads shows transfer effects across experiments, suggesting a shared mechanism. Third, we introduce energy from the associative-memory literature as a metric to quantify discourse coherence. Taken together, our results show that coherence illusions arise in Dutch LLMs, with entropy and energy exposing mechanisms that operate across settings.

This paper revisits the grounding problem for large language models in light of recent vector-grounding accounts. I accept the shift from classical symbol grounding to vector grounding, but argue that the current debate remains incomplete in two respects. First, reference is often treated too thinly, as if it were a fixed link between an isolated expression and an object. I argue instead that reference is profile-based, context-sensitive, discourse-level, affectively shaped, and norm-governed. Even in the human case, reference is publicly stabilized through patterns of use, correction, distinction, inference, and continuation rather than through identical private representations. Second, vector grounding requires an account of numerical realization. LLMs do not acquire reference through human perception, memory, intention, embodiment, or understanding. Rather, through optimization, they parameterize linguistic traces of human world-directed practice. In a finite vector system, referential profiles must be distributed, may be superposed, and are recovered through context-sensitive computation. Weights, activations, attention-mediated hidden states, softmax-trained contrasts, and inner-product alignments are the mathematical sites at which inherited linguistic relations become stable and causally active. Mechanistic interpretability findings, including entity-like features, knowledge neurons, and emotion-related activation directions, provide indirect support for this view. They do not show that LLMs possess human reference. They support a more limited thesis: LLMs may possess derivative, language-mediated, profile-based, and numerically structured forms of reference.

Medical Vision-Language Models (Med-VLMs) achieve strong expert-level performance, yet their ability to generate patient-accessible descriptions remains underexplored. With the 21st Century Cures Act now mandating immediate patient access to diagnostic imaging results, evaluating whether Med-VLMs can bridge this Expert-Lay Gap is both urgent and clinically consequential for patient education and shared decision-making. To this end, we introduce MedLayXPlain, the first large-scale multimodal benchmark and evaluation framework for Medical Lay Language Generation (MLLG). MedLayXPlain-122K provides 122,789 region-grounded samples across 8 imaging modalities from 12 publicly available source datasets, each comprising a medical image with paired expert and lay captions anchored in a three-level Unified Medical Language System (UMLS) ontology hierarchy spanning 7 semantic groups, 43 semantic types, and 2,411 medical concepts. Lay captions are constructed via Hierarchical Ontology-Verified Refinement (HOVER), a three-step pipeline combining patient-centric vocabulary mapping, LLM-based constrained rewriting, and cross-model visual verification to enforce semantic equivalence while preventing hallucination. We further introduce MedLayEval, a lightweight 3B evaluator distilled from a 27B verifier that scores expert-lay alignment across five clinically grounded attributes, addressing the poor correlation between standard NLG metrics and clinical judgment. Benchmarking 33 VLMs on MedLayXPlain-122K reveals a systematic Expert-Lay Gap: medical VLMs achieve strong expert captioning but suffer significant lay-register degradation, while general-purpose VLMs produce more accessible language yet lack clinical precision, confirming that neither current paradigm adequately serves patient-facing communication.

Dementia diagnosis requires integrating multi-modal clinical assessments from diverse informants and clinicians under incomplete and heterogeneous data conditions. Yet most AI-driven approaches remain Alzheimer's disease (AD)-centric, framing the problem as binary AD detection or three-stage AD progression modeling within well-curated research settings. This pathology-driven paradigm overlooks the broader, syndrome-level nature of dementia, which spans multiple stages, phenotypes, and etiologies. In this paper, we propose Dementia-Agents, a clinically aligned multi-agent framework for real-world dementia staging and phenotyping. The framework follows a three-step workflow: (1) a data agent translates structured clinical records into semantically faithful textual representations that preserve missing-data signals and routes them to domain-aligned experts; (2) five fine-tuned expert agents generate domain-level predictions; and (3) a coordinator agent performs probabilistic aggregation to produce final staging and phenotyping decisions. We develop and evaluate Dementia-Agents on a real-world clinical cohort of 1,066 patients from two cognitive neurology services. Compared with monolithic multi-modal large language models (MLLMs) and prior medical multi-agent systems, our approach achieves consistent improvements in diagnostic performance for real-world syndrome-level dementia staging and phenotyping, while preserving domain-level interpretability.

Attorneys, judges, and pro se filers increasingly use AI to draft legal documents, yet these tools frequently fabricate citations. Despite predictions that newer models would hallucinate less or that court sanctions would deter negligent filers, we found over 1,000 filings containing fabricated citations—with this number growing year-over-year. This study evaluates whether AI-based systems can mitigate these errors by automatically detecting hallucinations. We propose a taxonomy of legal citation hallucinations grounded in actual court filings and introduce a dataset of 1,300 brief excerpts containing injected errors. Benchmarking five models in agentic and non-agentic settings as well as Claude Code reveals that while the latest iterations perform better—GPT-5 achieves 84.4% recall and a 55.0% F1 score in an agentic framework—all models struggle with subtle error categories. Agentic verification remains resource-intensive, with GPT-5 averaging 15.3 steps per excerpt. Furthermore, restricted information access limits the efficacy of even the best agents. This gap creates policy concerns, as it disadvantages both AI systems and litigants who lack subscriptions to commercial legal databases. Together, our dataset, tools, and policy recommendations provide a foundation for building and auditing reliable legal citation checking tools.

Long-term memory systems allow LLM agents to preserve information beyond a single context window, but most systems focus on storing and retrieving facts after extraction, leaving the write decision under-specified. What deserves memory can depend on the user's current task, topic, activity, or interaction partner, while uniform extraction applies one notion of importance across these different situations. We formulate this challenge as preference-conditioned write control and introduce AdaMem, which uses adaptive natural-language Memory Policies to personalize what an agent writes to memory. Each policy represents the user's memory preference for a particular interaction context, is updated from periodic feedback, and controls subsequent memory writing. We evaluate this loop in AdaMem-Bench, which assigns different memory preferences to six concurrent interaction personas across five ten-week stories. Across two extraction models and two feedback modes, AdaMem improves average QA accuracy over Mem0 from 80.0% to 84.35% while reducing persistent memory by 9.27%. Our analyses show that explicit feedback helps models learn better memory policies, but current models still struggle to translate those policies into reliably selective writing behavior. AdaMem thus demonstrates the promise of adaptive write control while exposing policy execution as a central limitation of current memory agents. Our code is publicly available: https://github.com/galaxyChen/AdaMem