Papers

In-context learning (ICL) applies large language models to new tasks by prepending a few labeled demonstrations to a query. Prior work has studied how demonstration ordering affects ICL, but the interaction between label arrangement patterns and model biases has not been systematically examined. We define label collapse as the failure mode in which the model generates query-independent labels, and show that it is architecturally universal at the tested 8-9B scale: five models spanning autoregressive (Llama-3, Qwen3.5) and discrete diffusion (LLaDA, iLLaDA, LLaDA-Instruct) families all exhibit label collapse even under the class-balanced alternating arrangement, with some label word pairs collapsing in virtually every tested condition and others flipping their collapse direction with arrangement. Critically, accuracy alone cannot detect this collapse: models can achieve near-chance accuracy while generating query-independent labels, a score indistinguishable from balanced performance, so the failure can persist undetected in deployment. We measure separation, the log-odds of the two label words at the query position (the quantity that label-bias calibration operates on), as a complementary diagnostic that exposes collapse hidden beneath surface accuracy, together with a measurement protocol that guards label extraction against tokenizer and formatting artifacts. For autoregressive models, collapse persists where the majority-label-bias account does not apply, and arrangement patterns can reverse the collapse direction; for diffusion models, we present a systematic evaluation of ICL failure modes, complementing contemporaneous studies of dLLM context comprehension and demonstration-order sensitivity, and identify open questions about how iterative denoising interacts with label biases.

Syntactic language models (SLMs) that generate sentences along with their syntactic tree structures have shown promise in both syntactic generalization and downstream tasks. Prior studies, however, have mostly evaluated smaller models trained on less pretraining data and tested on narrower task suites. In addition, the design space of tree linearization and attention masks remains underexplored. We conduct a large-scale empirical study of 12 controlled configurations—one terminal baseline and 11 SLM variants spanning different linearizations and masks—at scales up to 1B parameters and 100B training tokens, evaluated across 11 downstream tasks and two syntactic benchmarks. The experimental results show that SLMs with standard causal attention match or exceed structural-mask variants across downstream and syntactic benchmarks. They require no architectural modifications and remain compatible with FlashAttention. We also observe that when scaling up to the 1B-parameter scale, the top-performing SLM variant no longer shows a clear advantage over a compute-matched non-structural baseline, suggesting that SLMs may become less competitive with scale. Our code is available at https://github.com/qlwpc/TG-Interpolation.

When does a shared representation become insufficient for jointly supporting multiple tasks? This question lies at the heart of multi-task learning models. Although task interference in shared representations has been widely observed, its onset, measurable form, and capacity boundary remain poorly understood. In this paper, we revisit this problem through the lens of Bell's theorem. We show that if four binary task outcomes can be jointly represented by a common shared representation, their correlations must satisfy the classical CHSH bound $|S|\leq2$. We introduce NCnet, a controlled multi-task model for studying how representational capacity affects task compatibility. Experiments reveal a clear capacity-dependent transition. Once nontrivial task fitting emerges, constrained capacity can yield $|S|>2$. As capacity further increases, these exceedances progressively disappear and $S$ returns to the bounded region. This systematic pattern provides evidence that limited representational capacity is an important source of the observed incompatibility. We observe similar transitions across CNNs, PLMs, and LLMs, suggesting that the phenomenon extends beyond the controlled NCnet setting. We refer to this capacity-dependent transition as the Bell Boundary, providing a measurable diagnostic for representation bottlenecks, task incompatibility, and capacity allocation in multi-task models.

Proximal Policy Optimization (PPO) is widely used in reinforcement learning with verifiable rewards for language-model reasoning. While value evolution along trajectory characterizes relative token-level credit, yet standard critics often fail to track it under sparse terminal rewards. To model this evolution, we analyze the relation between successive prefix values and show that they admit a Bayesian decomposition with additive updates in log-odds space. We use this additive structure to construct Tracking-Value PPO (\method), a cumulative value parameterization that sums position-wise scalar outputs into value logits. These outputs are learned through terminal-outcome supervision on the reconstructed values. \method improves aggregate mathematical reasoning scores over PPO by 7.0 and 8.4 points on Qwen3-4B-Base and Qwen3-8B-Base, respectively, while improving out-of-distribution generalization. Gains extend to Qwen3-4B, Qwen2.5-Math-1.5B, Qwen3.5-9B and to countdown, and scientific coding tasks. Further analysis shows more accurate tracking of value evolution and more reliable advantage estimates. Our code is publicly available in https://github.com/LINs-lab/TPPO.

Parallel token proposals need a joint law, not merely accurate token marginals. We construct coordinate-Bethe-ansatz laws from the periodic XXZ chain and compare them with their maximum-entropy pairwise projections at identical one- and two-site moments. In an eight-site, three-excitation sector, symmetry makes the full probability-law difference one-dimensional, determined by a contiguous three-site moment $\delta$. Along the Bethe branch, this gap changes sign and vanishes exactly at $\Delta=(3+\sqrt3)/2$, where the eigenstate law becomes pairwise. A nonlinear two-token readout preserves separate token marginals yet exposes a distance $2|\delta|$; a direct three-token readout preserves all one- and two-token marginals yet exposes $4|\delta|$. At $\Delta=1.5$ these deterministic binary distances are $0.02654573$ and $0.05309145$. We give an exact target-alignment formula and prove a strict speculative accepted-prefix advantage throughout an unrestricted open neighborhood of a constructed target. These are finite-law existence results, not measured language-model or wall-clock gains.

Comparing machine-labelled sentiment across social media platforms is now routine: differences in classifier scores between platforms are read as differences in public opinion. These comparisons rest on an assumption that the classifier measures the same construct, on the same scale, on every platform. Some studies check a classifier's validity platform by platform, but to our knowledge none has modelled the platform as a facet of the classifier's measurement. We adapt the psychometric equivalence toolkit (generalizability theory, measurement invariance and differential item functioning, and score linking) to machine-labelled policy sentiment, treating the platform as a survey mode, and specify a six-step protocol with anchor events fixed in advance, an anchor-density audit, and a two-classifier convergence control. A simulation with known ground truth illustrates what the assumption can cost in one generating configuration fixed in advance: across 500 replications, because the platforms measure through different functions, a pooled index that takes platform scores at face value reports the wrong sign of aggregate sentiment on nearly a quarter of days, whereas an index adjusted with each platform's known measurement function does so on under one day in sixty on average. A case study of the instrument from the first author's MSc dissertation, a fine-tuned BERT classifier over 255,446 UK economic-policy comments from Twitter/X, Reddit, YouTube, and Quora, motivates the design and illustrates the instrument-dependence the protocol addresses. The full four-platform application, under fresh collection with the pre-specified design, is the designated follow-up study. The protocol lets a researcher test whether pooling platforms is defensible in a given corpus, or whether platform scores must be linked before any average is taken.

Text generated by Large Language Models (LLMs) often remains limited in diversity compared to that produced by humans. Recent advances demonstrate the presence of inter- and intra-model collapse in open-ended generation of English text, raising concerns about the long-term homogenization of LLM, and more importantly, human thought. However, our understanding of whether this homogenization extends to languages beyond English is limited. In this work, we investigate the role of language on output diversity. Specifically, we aim to understand whether multilingual LLM-generated text is also homogeneous, and whether this homogeneity matches what we observe in humans. Across 23 open-weight models and 13 languages, we find that the artificial hivemind effect is pervasive but uneven in its extent. The intra-model effect holds in three different embedding spaces and under a lexical metric that uses no embedding model, so it is not an artifact of how diversity is measured. The size of the homogenization gap is associated with a model's competence in a language, measured on a held-out translation benchmark, robustly across embedding spaces. Thus, we demonstrate a performance-diversity tradeoff in multilingual LLMs.

Visual contrastive decoding corrects answers using the prediction difference between the original image and a reference condition, yet the same contrast also turns supported abstentions into incorrect answers. We find that abstentions are lost at two stages: contrastive decoding with candidate filtering can exclude the abstention token before any comparison takes place, and the reference branch's response to the abstention instruction is removed when the reference prediction is subtracted. We therefore propose instruction-preserving decoding, which keeps the effect of the abstention instruction on the original-image prediction intact while leaving the contrast strength freely adjustable. Using a perturbed image and an image-free prediction as references, we obtain IP-VCD and IP-M3ID. On decision utility (DU), which credits both correct answers and supported abstentions, both methods outperform all compared non-IP baselines on every one of nine models on VizWiz, and IP-VCD also leads on eight of nine models on Food-101. Candidate measurements and paired per-question comparisons show that computing the instruction effect and the visual contrast term separately lets contrastive decoding support both correct answers and supported abstentions.

Existing benchmarks for GUI agents primarily focus on increasing task difficulty by extending task horizons. Besides, enabling agents to explore unfamiliar environments and successfully complete tasks within them is also a critical yet challenging problem. We introduce LatentGUIWorld, a benchmark of six tasks in three matched pairs organized by the evidence needed for control: $I_0$ uses prior knowledge and available observations, $I_1$ requires interaction that directly reveals hidden semantics, and $I_2$ identifies dynamics from action-response relationships. Frontier models struggle especially on $I_2$ tasks. We introduce MOVE, which separates movement from button events and exposes intermediate visual feedback, together with demonstrations of iterative exploration and control. Ablations show that both action granularity and multi-step demonstrations improve performance. After SFT, LatentLearner-9B achieves 50.0% overall success on LatentGUIWorld, comparable to several frontier models. Joint RL further raises egocentric Drag and Ten-Choice success to 92.0% and 75.3%. Behavioral analysis reveals multi-step exploration. In egocentric Drag, RL increases Pearson correlation between error-normalized commands and ideal gain compensation from 0.610 to 0.803. Together, these behaviors show that models learn to explore through interaction and adapt their control to environment dynamics. The same format improves GUI grounding after RL, reaching 67.37% mean accuracy across five benchmarks versus 66.72% for conventional RL.

Does a large language model have a self, and if so, of what kind? We separate two functional levels familiar from biology: a self-index, which distinguishes one's own material from external material and shapes processing, as corollary discharge does in nearly all animals, and self-reference, the higher ability to recognize what is one's own, as in mirror self-recognition. We probe both in Qwen3.8-27B by placing the same text in an Assistant, a User, or a System message rendered by the model's own chat template, and report three levels of evidence. Persistence: as predicted in advance, the representation of the text separates the model's own message from both external sources at all 64 layers and for every test text, along a shared direction, even when the answer does not depend on who supplied the text. Coupling: asked who wrote the preceding message, the model claims its own messages (0.77) and assigns User and System messages alike to the user, and how strongly it claims each message is predicted by the size of its self-index (Spearman correlation up to 0.68). Intervention: exchanging the internal states that follow a message, rather than those of its text, reverses the self-attribution between layers 15 and 31; removing the self-index direction from them at layer 39 lowers the claim more than a random direction; and in arithmetic the model adopts a wrong step more often when it wrote the step itself (0.92 versus 0.75), an effect carried by the step's representation at layers 31 to 39. Reasoning first in thinking mode raises the probability of the correct answer (0.96 instead of 0.77), but recognition is present without it. We propose that self-reference is a higher-level ability built on a self-index, a content-independent displacement superimposed on the whole representation of the model's own messages.

Teacher rationales can help a smaller model answer health professions questions, but they can also reinforce errors. We examined when they help and how this evidence can guide training using Brazilian Portuguese questions from medical and multiprofessional residency examinations. First, we measured how a teacher rationale changed the student's probability of selecting the correct answer and compared this effect with rationales drawn from other questions. The complete rationale increased this probability by 21~p.p. when the teacher was correct and decreased it by 14~p.p. when the teacher was wrong. By intervening on the reasoning content, we showed that shuffled rationales did not produce the same pattern. Guided by this result, we trained only with rationales attached to correct teacher answers and compared answer-only training, step-by-step distillation, and our proposed Distill-SFT method. Distill-SFT showed a modest advantage over step-by-step distillation on previously missed questions under reasoning, while using fewer training instances by combining the answer and rationale into a single target. We also investigated catastrophic forgetting using the out-of-domain ENEM dataset. Answer-only training reduced reasoning accuracy by 12~p.p. relative to baseline, compared with approximately 4~p.p. for either rationale-based training method. Together, these findings support a workflow that audits rationale content before distillation, verifies teacher answers before supervision, and trains compact models efficiently for health professions question answering.

Language models have recently solved open Erdős problems and achieved gold-medal level performance on International Mathematical Olympiad problems. But their solutions, written in natural language, cannot yet be verified by proof assistants. An erroneous reasoning step can invalidate every step that comes after, so it is advantageous to detect errors early. Several methods have been proposed to detect erroneous reasoning quickly based on the features of the forward pass of the language model that performs reasoning. In this work, we study whether the directed Forman curvature of that model's attention graph, a local quantity computed at an edge from the attention weights around it, carries information about a step's correctness. To do so, we evaluate four per-step verifiers on ProcessBench with and without the curvature appended to their input. Appending curvature values improves the overall mean performance of all four verifiers. ReProbe, the verifier that takes every token of a step as input, shows the largest improvement. Adding the curvature increases the time of the language model's forward pass by 18% on average.

Recruitment applications span diverse operations, from drafting job descriptions and interpreting queries to organizing profiles, assessing fit, and recommending candidates and jobs. Systematic LLM evaluation therefore requires task coverage that reflects this range of business needs. We introduce EHR-Bench, an open Chinese benchmark comprising twelve tasks across six capability families and five recruitment scenarios. The tasks cover generation, extraction, classification, retrieval, semantic similarity, and reciprocal ranking within one evaluation framework. An internal recruitment dataset guides task definitions, the occupation taxonomy and sampling targets, output requirements, and scoring protocols. Following these specifications, we construct 8,336 task instances from public materials and controlled augmentation, with source-aware expert review and documented provenance. We evaluate twelve LLMs through task-level, capability-level, and aggregate comparisons. Closely ranked models exhibit distinct task strengths, and performance varies across input sources, reference evidence, and candidate pools. Five shared models retain the same observed Overall ordering on a private operational reference, with variation in capability and task correspondence. EHR-Bench provides a reproducible foundation for systematic model assessment, task-specific model selection, and further research on LLMs for recruitment. The project repository is available at https://github.com/EhrBench/EHR-Bench.

Decision APIs that return a probability for each option are routinely used as if those probabilities were properties of the options themselves. That reading is valid only if the model satisfies independence of irrelevant alternatives (IIA): adding an option that is neither the right answer nor its strongest rival should leave the odds between those two unchanged. We test this in a pre-registered study of jev-1.13.0, a hosted decision model, using 299,463 calls on intent classification (BANKING77, CLINC150) and topic classification (AG News). Every added option moves the log-odds between the gold label and the strongest wrong label: the root-mean-square systematic shift is 0.20-0.52 nats, 1.2 to 3 times the run-to-run variation, even when the added option takes at most 0.1% of the probability mass. The shifts are directional. An option similar to the wrong answer draws probability from it (+0.10 nats toward the gold label). A paraphrase of the wrong answer produces an intermediate pattern: the gold label's probability is almost conserved, but the paraphrase receives only about a third of the wrong answer's share rather than the half that strict substitution implies. The distortion grows steadily with the logarithm of the number of options, reaching about 1 nat at 255 options, and off-domain padding pushes the odds against the gold label. In contrast, questions packed into one call do not interact (bias < 0.001). Scoring each option with an independent yes/no question removes the option-set dependence but picks the right answer 8.7 points less often. Jev sits between two local baselines, an option-independent scorer and a prompt readout. Its probabilities are best read as relative to the option list they were computed over.

Machine translation (MT) is increasingly utilized in multilingual crisis communication, yet conventional evaluation does not directly assess whether translated emergency messages preserve their actionability. We introduce SIREN (Situation-aware Index for Response and Emergency ActioNability), a human evaluation framework assessing whether translation choices compromise protective action, directive force, and urgency of the source text. We evaluate SIREN in two complementary settings. At the error level, we apply it to HOPE-annotated Italian–English translations produced by ChatGPT-4o and Google Translate: 31.63% and 48.73% of errors, respectively, compromise actionability, predominantly by affecting the protective action. At the message level, we evaluate English–French translations produced by TranslateGemma-4B and Google Translate, with actionability preserved in 85.34% and 92.67% of translations, respectively. TranslateGemma-4B failures predominantly weaken directive force, whereas Google Translate failures more often compromise protective actions and receive Major impact ratings. These findings show that translation systems can differ not only in how often actionability is lost, but also in the nature and functional impact of these failures. Alongside SIREN, we contribute an English–French crisis-translation resource with human actionability annotations and post-edited reference translations, supporting further research on function-oriented evaluation for crisis communication. Our experimental data is available at https://github.com/mcstaiano/SIREN.

State space models provide efficient linear-complexity processing for long sequences, but their performance frequently degrades as context length grows. We propose SuPER-Mamba, a training-free framework that uses Surprisal as a model-intrinsic signal for long-range state routing in pretrained Mamba models. SuPER-Mamba computes position-wise Surprisal from next-token predictions, aggregates neighboring responses through Local Pooling, and uses the resulting scores to prioritize Global-Channel updates without query-dependent scoring or additional training. We evaluate SuPER-Mamba on three state-space models across the LongBench-E benchmark. The results show that SuPER-Mamba improves the average score from 8.06 to 15.69, corresponding to a 94.7% relative gain over the Base Models and a 13.1% improvement over the state-of-the-art prior baseline. Ablation and length-wise analyses further support the effectiveness of the proposed routing framework.

Processing multilingual text requires both identifying a broad range of languages and distinguishing the language of each token in inputs containing multiple languages. We introduce MixLID, a token-level language identification model supporting 111 language-script combinations. MixLID follows a two-stage training process: it first learns to distinguish individual languages using the language labels of monolingual examples, then continues training on mixed-language data. The second stage uses concatenated examples from existing code-switching data alongside augmented examples created through word substitution based on correspondences between English source words and loanwords. For monolingual identification, MixLID outperforms all baselines in both accuracy and macro F1 on four benchmarks within the language coverage shared with each baseline. For mixed-language identification, it outperforms the baselines in identifying the language of individual tokens on every evaluated language pair. MixLID achieves high performance across a broad range of languages in both monolingual and mixed-language settings. The model is publicly available at https://huggingface.co/nlpai-lab/MixLID.

Sentence representation learning is still an open challenge in Natural Language Processing. In this work, we propose a new self-supervised framework for learning sentence representations, using a special type of neural network called a recursive hypernetwork. Our proposed model composes the representation of a sentence from representations of words by applying a recursive composition through the parse tree. We maintain a separate syntactic and semantic representation, and the semantic composition is guided by the information from the syntactic representation. To train this model, we introduce a novel set of six self-supervised tasks. By analysing the performance on 7 probing tasks, we validate that the generated sentence representation encodes richer linguistic information than both averaging baselines and state-of-the-art alternatives. Furthermore, we assess the impact of the six proposed self-supervised training tasks through ablation studies. We also demonstrate that the representations generated by our model are stable for sentences of varying length and that the semantic composition operators adapt to different syntactic categories\footnote{Code available: https://github.com/vasudev2020/codi/tree/main}.

We present PatTransBench, a 12,000-item Chinese-English/Japanese/Korean patent machine translation (PatMT) benchmark with a four-dimensional evaluation protocol. The benchmark comprises 6,000 short-text (sentence/claim) and 6,000 long-text (paragraph) pairs, evenly split across three language pairs and both translation directions, built via IPC-section-stratified sampling and dualthreshold quality gating from upstream patent-family parallel corpora. The evaluation protocol spans literal n-gram overlap (BLEU, ROUGE-L, chrF), cross-lingual semantic textual similarity (CLSTS), 38-probe-language consistency testing, and serving performance under concurrency. A case study on PatMT2-7B and its fp8/AWQ4/AutoRound-4bit variants reveals direction-dependent quantization effects masked by aggregate scores and non-monotonic language-failure patterns across quantization schemes that are invisible to single-metric evaluation. We plan to release the benchmark, protocol, and case-study results upon publication to support reproducible PatMT evaluation.

Reinforcement learning with verifiable rewards (RLVR) is reshaping how large language models (LLMs) acquire reasoning capabilities, yet the optimizers used remain largely inherited from teacher-forcing training. This raises a fundamental question: does RLVR induce a distinct optimization geometry that calls for a different optimizer? We investigate this question through the spectral structure of LLM gradients and find that gradient energy progressively concentrates into a shrinking effective subspace, which is characteristic of RLVR, as it is absent under supervised fine-tuning (SFT) and is largely removed when the verifiable reward is replaced by a random one. Moreover, the rank required to capture a fixed fraction of gradient energy steadily decreases over training, revealing that the effective dimensionality of useful updates itself evolves with training. These findings motivate a new optimization principle: rather than prescribing the update rank externally, an optimizer should adapt its rank to the evolving spectral structure of the gradient. We introduce Ares (Adaptive Rank-Effective Spectral Optimization), which estimates the effective update rank from the gradient's spectral energy distribution, tracks its evolution, and confines the gradient signal driving each update to the resulting low-rank subspace. Across LLMs of different scales, Ares improves accuracy over widely used optimizers and generalizes across RLVR algorithms and tasks. Results suggest that effective optimization may require optimizers to be designed around the geometry and dynamics of the training process. We release the code and data at https://github.com/RUCAIBox/Ares.

Model-internal signals, such as neuron activations and gradients, have been used for instruction data selection, yet how neuron relevance should be characterized remains underexplored. In particular, should selection prioritize the magnitude of relevance or how broadly it is distributed across neurons? To investigate this question, we construct sample-level relevance profiles using AttnLRP-guided attribution and introduce two complementary criteria: Effective Relevance Support (ERS), which measures the effective support of positive relevance, and Positive Relevance Magnitude (PRM), which captures its average strength. Our analyses reveal that these characterizations produce distinct sample rankings and selection preferences despite sharing the same underlying relevance profiles. On Qwen2 models, PRM exhibits stronger associations with activation-based selection, whereas ERS produces substantially different rankings. These differences extend to task composition, within-task sample preferences, and semantic selection patterns, with varying degrees of agreement across backbones. Experiments across three language models and five downstream benchmarks demonstrate the practical value of relevance-guided selection, with ERS achieving the best overall average performance rank among the evaluated methods. Our findings highlight that how neuron relevance is distributed and characterized, rather than its magnitude alone, matters for instruction data selection.

Training LLM agents to interact with users over multiple turns is challenging because the learning signal depends not only on the agent, but also on the user simulator and environment, which jointly shape the agent's experience. Simulated users can be stochastic or inconsistent, while environments may contain unreliable tools, incomplete state coverage, and incorrect verifiers. Prior work mainly addresses the resulting instability through reward shaping or denser supervision, while treating these components as fixed. We instead introduce EAU, a meta-agent framework for environment–agent–user co-evolution. During training, EAU monitors trajectories, tool traces, rewards, and training statistics; diagnoses failures and uninformative interactions; and proposes targeted modifications to the user simulator, environment, reward design, and RL configuration. Validated changes are hot-swapped into the ongoing run, allowing the training setup to evolve together with the policy. These adaptations also persist across runs, so later training can build on previously discovered improvements rather than restarting from a fixed setup. We enable this process through an editable symbolic representation of user and environment dynamics, while the agent remains a neural LLM optimized with RL. Across three interactive-agent benchmarks and two backbone models, EAU consistently outperforms prior RL methods, improving the overall score over the strongest baseline by 14.7–15.0% and increasing successful task completion on ColBench by up to 36.7%, showing that co-evolving the training system with the policy can substantially improve multi-turn RL.

In parallel decoding, diffusion language models commit several tokens per forward pass, so co-committed tokens cannot see each other, breaking multi-token format units that parsers need. On ECHO, an open-weight one-step chest X-ray (CXR) report generator, 32.7-36.7% of evaluation reports lack a parseable Impression heading and score as empty. We propose intent-triggered anchoring, a training-free decoding rule that completes a format unit once the model commits its first token, inserting only units with a unique, content-free completion; fallbacks write the unit when the model stops or loops without starting it. On three CXR evaluation sets, the rule raises readable rates from 63.3-67.3% to 100.0% with no regressions and a 95th-percentile latency ratio of 0.68-1.28 against official decoding. Broken headings state findings absent from the reference; adding the heading at the end leaves them in 15.5-31.0% of reports, our rule in at most 0.6%. For math answer markers, the rule raises accuracy in the most parallel setting of five diffusion language models by 5.8-20.8 points on GSM8K and 5.1-17.1 on GSM-Symbolic, held out from tuning. With lenient answer extraction, gains are 3.3-12.5 and 2.7-6.3 points; the rule adds 1.4-8.6 points over a stop-only fallback for four of the five models.

The contention is that the current batch of Large Language Models don't *understand* language, but they *do* it. Rather than LLMs forming some kind of “distributed representation” of symbols, the claim is that LLMs are capturing “Cultural Intelligence” and grounding symbols in what the sociologists call practices.

Clinical oncology documentation records decisive information in free text, where it remains readable but not retrievable at scale, leaving cohort identification dependent on incomplete diagnostic codes or manual review. The current standard, coding individual terms, captures what a document mentions, not what it asserts: meaning lies in how those terms relate to one another. We present a method for deriving queryable knowledge graphs from clinical narratives, conforming to the Extended Grammar of SNOMED CT across multiple ontologies. The method specifies how entities should be annotated and related. It defines five relation classes, each bound to a construction rule that assembles the entities and relations into post-coordinated expressions. A deterministic implementation, UNIGRAPHE, applies those rules to construct knowledge graphs from annotated text. Applied to expert-annotated real-world oncology documentation, UNIGRAPHE reconstructed 5,299 annotations into 2,196 post-coordinated expressions and shows that relation structure is the missing piece to make clinical narrative formally queryable.

We present PatMT2 (Patent Machine Translation, v2), a two-stage LoRA pipeline that adapts the Hy-MT2-7B dedicated translation model to four-language (Chinese/English/Japanese/Korean) patent translation. On six translation directions, the Stage-2 checkpoint reaches a macro-average BLEU of 43.39 (BF16) to 43.57 (AWQ INT4), improving over the Hy-MT2-7B base by up to +8.8 BLEU and also exceeding Qwen3-32B and several external systems on the same average. We then measure the cost of adaptation: probing 38 languages absent from the fine-tuning data, the model still produces correct, fluent output on 34, while four low-resource languages collapse into distinct, reproducible failure modes. We treat these as diagnostic cases rather than retention rates, and report their mechanisms without attributing them to a single cause. Paired bootstrap tests confirm the translation gains over all baselines (p < 0.001).

Real-time AI captioning now serves audiences at multilingual events, yet computer-assisted interpreting (CAI) tools are still built for interpreters, rarely by them. This article reports an autoethnographic design study in which a professional interpreter with no programming training built a local-first CAI system through expert-led 'vibe coding', co-creating with an LLM architect and a coding agent. Analysis of the complete development dialogues shows how tacit professional knowledge, from segmentation to booth infrastructure, was externalized into design after generic defaults failed in use. In a demanding public scientific event, three mechanisms of in-booth appropriation emerge under a firmly human-led division of labor; telemetry from a first fully instrumented assignment shows a fully local chain sustaining a median translation latency of 0.94 s across 971 utterances. The case is read as lead-user innovation newly enabled by generative AI, with implications for the technological literacy of interpreters and the future of interpreter training.

Preference alignment enables language models to better reflect human values, but the preference data contains sensitive information, especially when derived from surveys about political beliefs, social norms, religion, gender attitudes, and cultural values. This paper studies privacy-preserving alignment on questionnaire-derived data from the World Values Survey, where supervision takes the form of ordinal labels rather than standard pairwise rankings. Existing efforts mainly focus on general privacy protections upon model training, while data-specific methods for survey data remain unexplored. We first show the vulnerability of aligned models, then formulate the task as rating prediction and introduce label differential privacy. Building on this formulation, we construct a differential private pipeline utilizing three privacy mechanisms: randomized response, exponential mechanism, and DP-SGD. Experiments show that simple label-level mechanisms outperform gradient-level methods, and proper aggregation function for preprocessing also make large contribution. The new fine-grained private comparison motivates explorations toward language model alignment with formal privacy guarantees.

The growing use of language models for forecasting motivates a broader evaluation of their world-model capabilities. Such evaluation must go beyond predictive accuracy to assess hidden-state inference and intervention prediction. Real-world forecasting benchmarks offer limited access to hidden-state ground truth and controlled experiments. We introduce CivWorldBench, a benchmark built on a multi-agent strategy game engine with recorded internal states, controlled observations, and paired intervention rollouts. Its 1,900 decision points from 475 worlds support three task levels: association forecasting, belief inference, and intervention prediction. Reference predictors support attributable evaluation of capability gaps, with an expert harness and empirical ceiling estimates contextualizing scores. Stored trajectories make intermediate predictions verifiable, while agent-play experiments test whether world-model capabilities support decision-making. The main panel comprises eleven frontier language models. Composite benchmark scores correlate with real-world forecasting leaderboards on matched model panels, more closely than with a general-capability index, supporting its use as a proxy for real-world forecasting capability. Strong association forecasts do not consistently extend to belief inference or intervention prediction, and effect magnitudes remain poorly calibrated despite informative direction predictions. Process verification reveals errors missed by final-answer evaluation, and structured reasoning improves intervention prediction for the strongest model in the process study. Agent-play experiments identify useful assisted plans and motivate evaluating decision-relevant capabilities beyond association forecasting. These findings identify hidden-state inference, intermediate-state prediction, and effect calibration as priorities for world-model training.

Current image-text evaluation often relies on a global score, which gives limited insight into which semantic details are supported by the image. Explainable metrics provide rationales, but the visual evidence supporting those rationales can still be difficult to verify. We introduce VisExMEM, a reference-free evaluator that decomposes captions into atomic claims, localizes visual evidence, verifies each claim, and averages claim-level support into a global score. Explanations use the same evidence and decisions. We train Qwen3.5-based 4B and 9B models on 100K samples with multi-task SFT, counterfactual refinement, and evidence-aware GDPO. Across eight benchmarks, VisExMEM-9B achieves the highest descriptive composite among open-source evaluators in our comparison, exceeding VQAScore by 11.7 points while remaining competitive with GPT-5.6 Sol. We also introduce AGITA-Bench, containing 600 human-annotated image-text pairs and 8.7K atomic claims for global and claim-level evaluation. For untuned Qwen3.5 and GPT-5.6 Sol controls, decomposed prompting raises the same composite by 3.6-5.1 points relative to holistic prompting. Human evaluation shows that VisExMEM explanations outperform prior explainable metrics and approach GPT-5.6 Sol.