Papers

This paper addresses Named Entity Recognition (NER) in the domain of Vocational Education and Training (VET), focusing on historical, digitized documents that suffer from OCR-induced noise. We propose a robust NER approach leveraging Noise-Aware Training (NAT) with synthetically injected OCR errors, transfer learning, and multi-stage fine-tuning. Three complementary strategies, training on noisy, clean, and artificial data, are systematically compared. Our method is one of the first to recognize multiple entity types in VET documents. It is applied to German documents but transferable to arbitrary languages. Experimental results demonstrate that domain-specific and noise-aware fine-tuning substantially increases robustness and accuracy under noisy conditions. We provide publicly available code for reproducible noise-aware NER in domain-specific contexts.

Guardrail models are essential for ensuring the safety of Large Language Model (LLM) deployments, but processing full multi-turn conversation histories incurs significant computational cost. We propose Defensive M2S, a training paradigm that fine-tunes guardrail models on Multi-turn to Single-turn (M2S) compressed conversations rather than complete dialogue histories. We provide a formal complexity analysis showing that M2S reduces training cost from $O(n^2)$ to $O(n)$ for $n$-turn conversations. Empirically, on our training dataset (779 samples, avg. 10.6 turns), M2S requires only 169K tokens compared to 15.7M tokens for the multi-turn baseline – a 93$\times$ reduction. We evaluate Defensive M2S across three guardrail model families (LlamaGuard, Nemotron, Qwen3Guard) and three compression templates (hyphenize, numberize, pythonize) on SafeDialBench, a comprehensive multi-turn jailbreak benchmark. Our best configuration, Qwen3Guard with hyphenize compression, achieves 93.8% attack detection recall while reducing inference tokens by 94.6% (from 3,231 to 173 tokens per conversation). This represents a 38.9 percentage point improvement over the baseline while dramatically reducing both training and inference costs. Our findings demonstrate that M2S compression can serve as an effective efficiency technique for guardrail deployment, enabling scalable safety screening of long multi-turn conversations.

Large language models (LLMs) offer a new empirical setting in which long-standing theories of linguistic meaning can be examined. This paper contrasts two broad approaches: social constructivist accounts associated with language games, and a mathematically oriented framework we call Semantic Field Theory. Building on earlier work by the author, we formalize the notions of lexical fields (Lexfelder) and linguistic fields (Lingofelder) as interacting structures in a continuous semantic space. We then analyze how core properties of transformer architectures-such as distributed representations, attention mechanisms, and geometric regularities in embedding spaces-relate to these concepts. We argue that the success of LLMs in capturing semantic regularities supports the view that language exhibits an underlying mathematical structure, while their persistent limitations in pragmatic reasoning and context sensitivity are consistent with the importance of social grounding emphasized in philosophical accounts of language use. On this basis, we suggest that mathematical structure and language games can be understood as complementary rather than competing perspectives. The resulting framework clarifies the scope and limits of purely statistical models of language and motivates new directions for theoretically informed AI architectures.

In the rapidly evolving landscape of enterprise natural language processing (NLP), the demand for efficient, lightweight models capable of handling multi-domain text automation tasks has intensified. This study conducts a comparative analysis of three prominent lightweight Transformer models - DistilBERT, MiniLM, and ALBERT - across three distinct domains: customer sentiment classification, news topic classification, and toxicity and hate speech detection. Utilizing datasets from IMDB, AG News, and the Measuring Hate Speech corpus, we evaluated performance using accuracy-based metrics including accuracy, precision, recall, and F1-score, as well as efficiency metrics such as model size, inference time, throughput, and memory usage. Key findings reveal that no single model dominates all performance dimensions. ALBERT achieves the highest task-specific accuracy in multiple domains, MiniLM excels in inference speed and throughput, and DistilBERT demonstrates the most consistent accuracy across tasks while maintaining competitive efficiency. All results reflect controlled fine-tuning under fixed enterprise-oriented constraints rather than exhaustive hyperparameter optimization. These results highlight trade-offs between accuracy and efficiency, recommending MiniLM for latency-sensitive enterprise applications, DistilBERT for balanced performance, and ALBERT for resource-constrained environments.

Forecasting on geopolitical temporal knowledge graphs (TKGs) through the lens of large language models (LLMs) has recently gained traction. While TKGs and their generalization, hyper-relational temporal knowledge graphs (HTKGs), offer a straightforward structure to represent simple temporal relationships, they lack the expressive power to convey complex facts efficiently. One of the critical limitations of HTKGs is a lack of support for more than two primary entities in temporal facts, which commonly occur in real-world events. To address this limitation, in this work, we study a generalization of HTKGs, Hyper-Relational Temporal Knowledge Generalized Hypergraphs (HTKGHs). We first derive a formalization for HTKGHs, demonstrating their backward compatibility while supporting two complex types of facts commonly found in geopolitical incidents. Then, utilizing this formalization, we introduce the htkgh-polecat dataset, built upon the global event database POLECAT. Finally, we benchmark and analyze popular LLMs on our dataset, providing insights into 1) the positive impact of utilizing the HTKGH formalization compared to existing ones and 2) LLMs' adaptability and capabilities in complex forecasting tasks.

Transformer residual streams are updated by addition. A sufficiently expressive residual block can represent content replacement, but the residual update itself has no operation that reads, compares, and replaces content. We introduce Deep Delta Learning (DDL), which applies the delta rule over network depth. Each layer reads the residual state along a learned direction, compares the readout with a learned target, and writes a gated rank-1 correction back along the same direction. A closed gate gives the identity map, and a unit gate overwrites the selected readout with the target. DDL works with the usual vector state or with an expanded state that stores several value channels, while attention and MLP blocks keep the original model width. We pretrain decoder-only models with approximately GPT-2 small and medium sizes on the FineWeb-Edu dataset. At both scales, every DDL variant has lower validation loss and higher average one-shot accuracy than the additive baseline, and the best expanded variants raise that average by 0.91 and 1.18 points.

We present judgeWEL, a dataset for named entity recognition (NER) in Luxembourgish, automatically labelled and subsequently verified using large language models (LLM) in a novel pipeline. Building datasets for under-represented languages remains one of the major bottlenecks in natural language processing, where the scarcity of resources and linguistic particularities make large-scale annotation costly and potentially inconsistent. To address these challenges, we propose and evaluate a novel approach that leverages Wikipedia and Wikidata as structured sources of weak supervision. By exploiting internal links within Wikipedia articles, we infer entity types based on their corresponding Wikidata entries, thereby generating initial annotations with minimal human intervention. Because such links are not uniformly reliable, we mitigate noise by employing and comparing several LLMs to identify and retain only high-quality labelled sentences. The resulting corpus is approximately five times larger than the currently available Luxembourgish NER dataset and offers broader and more balanced coverage across entity categories, providing a substantial new resource for multilingual and low-resource NER research.

Recent advances in vision-language models have opened up new possibilities for reasoning-driven image geolocalization. However, existing approaches often rely on synthetic reasoning annotations or external image retrieval, which can limit interpretability and generalizability. In this paper, we present Geo-R, a retrieval-free framework that uncovers structured reasoning paths from existing ground-truth coordinates and optimizes geolocation accuracy via reinforcement learning. We propose the Chain of Region, a rule-based hierarchical reasoning paradigm that generates precise, interpretable supervision by mapping GPS coordinates to geographic entities (e.g., country, province, city) without relying on model-generated or synthetic labels. Building on this, we introduce a lightweight reinforcement learning strategy with coordinate-aligned rewards based on Haversine distance, enabling the model to refine predictions through spatially meaningful feedback. Our approach bridges structured geographic reasoning with direct spatial supervision, yielding improved localization accuracy, stronger generalization, and more transparent inference. Experimental results across multiple benchmarks confirm the effectiveness of Geo-R, establishing a new retrieval-free paradigm for scalable and interpretable image geolocalization. To facilitate further research and ensure reproducibility, both the model and code will be made publicly available.

Joint Embedding Predictive Architectures (JEPA) are a novel self supervised training technique that have shown recent promise across domains. We introduce BERT-JEPA (BEPA), a training paradigm that adds a JEPA training objective to BERT-style models, working to combat a collapsed [CLS] embedding space and turning it into a language-agnostic space. This new structure leads to increased performance across multilingual benchmarks.

Multilingual large language models achieve impressive cross-lingual performance despite largely monolingual pretraining. While bilingual data in pretraining corpora is widely believed to enable these abilities, details of its contributions remain unclear. We investigate this question by pretraining models from scratch under controlled conditions, comparing the standard web corpus with a monolingual-only version that removes all multilingual documents. Despite constituting only 2% of the corpus, removing bilingual data causes translation performance to drop 56% in BLEU, while behaviour on cross-lingual QA and general reasoning tasks remains stable, with training curves largely overlapping the baseline. To understand this asymmetry, we categorize bilingual data into parallel (14%), code-switching (72%), and miscellaneous documents (14%) based on the semantic relevance of content in different languages. We then conduct granular ablations by reintroducing parallel or code-switching data into the monolingual-only corpus. Our experiments reveal that parallel data almost fully restores translation performance (91% of the unfiltered baseline), whereas code-switching contributes minimally. Other cross-lingual tasks remain largely unaffected by either type. These findings reveal that translation critically depends on systematic token-level alignments from parallel data, whereas cross-lingual understanding and reasoning appear to be achievable even without bilingual data.

The rapid advancement of large language model(LLM) technology has facilitated its integration into various domains of professional and daily life. However, the persistent challenge of LLM hallucination has emerged as a critical limitation, significantly compromising the reliability and trustworthiness of AI-generated content. This challenge has garnered significant attention within the scientific community, prompting extensive research efforts in hallucination detection and mitigation strategies. Current methodological frameworks reveal a critical limitation: traditional uncertainty quantification approaches demonstrate effectiveness primarily within conventional question-answering paradigms, yet exhibit notable deficiencies when confronted with non-canonical or adversarial questioning strategies. This performance gap raises substantial concerns regarding the dependability of LLM responses in real-world applications requiring robust critical thinking capabilities. This study aims to fill this gap by proposing an uncertainty quantification scenario in the task of generating with multiple facts. We have meticulously constructed a set of trap questions contained with fake names. Based on this scenario, we innovatively propose a novel and robust uncertainty quantification method(RU). A series of experiments have been conducted to verify its effectiveness. The results show that the constructed set of trap questions performs excellently. Moreover, when compared with the baseline methods on four different models, our proposed method has demonstrated great performance, with an average increase of 0.1-0.2 in ROCAUC values compared to the best performing baseline method, providing new sights and methods for addressing the hallucination issue of LLMs.

Speech is a scalable and non-invasive biomarker for early mental health screening. However, widely used depression datasets like DAIC-WOZ exhibit strong coupling between linguistic sentiment and diagnostic labels, encouraging models to learn semantic shortcuts. As a result, model robustness may be compromised in real-world scenarios, such as Camouflaged Depression, where individuals maintain socially positive or neutral language despite underlying depressive states. To mitigate this semantic bias, we propose DepFlow, a three-stage depression-conditioned text-to-speech framework. First, a Depression Acoustic Encoder learns speaker- and content-invariant depression embeddings through adversarial training, achieving effective disentanglement while preserving depression discriminability (ROC-AUC: 0.693). Second, a flow-matching TTS model with FiLM modulation injects these embeddings into synthesis, enabling control over depressive severity while preserving content and speaker identity. Third, a prototype-based severity mapping mechanism provides smooth and interpretable manipulation across the depression continuum. Using DepFlow, we construct a Camouflage Depression-oriented Augmentation (CDoA) dataset that pairs depressed acoustic patterns with positive/neutral content from a sentiment-stratified text bank, creating acoustic-semantic mismatches underrepresented in natural data. Evaluated across three depression detection architectures, CDoA improves macro-F1 by 9%, 12%, and 5%, respectively, consistently outperforming conventional augmentation strategies in depression Detection. Beyond enhancing robustness, DepFlow provides a controllable synthesis platform for conversational systems and simulation-based evaluation, where real clinical data remains limited by ethical and coverage constraints.

Quantization is widely used to accelerate inference and streamline the deployment of large language models (LLMs), yet its effects on self-explanations (SEs) remain unexplored. SEs, generated by LLMs to justify their own outputs, require reasoning about the model's own decision-making process, a capability that may exhibit particular sensitivity to quantization. As SEs are increasingly relied upon for transparency in high-stakes applications, understanding whether and to what extent quantization degrades SE quality and faithfulness is critical. To address this gap, we examine two types of SEs: natural language explanations (NLEs) and counterfactual examples, generated by LLMs quantized using three common techniques at distinct bit widths. Our findings indicate that quantization typically leads to moderate declines in both SE quality (up to 4.4%) and faithfulness (up to 3.9%). The user study further demonstrates that quantization considerably diminishes both the coherence and trustworthiness of SEs (by up to 8.5%). Compared to smaller models, larger models show limited resilience to quantization in terms of SE quality but maintain more faithfulness. Moreover, no quantization technique consistently excels across task accuracy, SE quality, and faithfulness. Because quantization's impact varies considerably by context and can be sizable in specific cases, we recommend validating SE quality for the intended use case. Despite these sometimes considerable drops, quantization remains an effective compression technique when its impact on SEs is properly validated.

We introduce WildAGTEval, a benchmark designed to evaluate large language model (LLM) agents' function-calling capabilities under realistic API complexity. Unlike prior work that assumes an idealized API system and disregards real-world factors such as noisy API outputs, WildAGTEval accounts for two dimensions of real-world complexity: 1. API specification, which includes detailed documentation and usage constraints, and 2. API execution, which captures runtime challenges. Consequently, WildAGTEval offers (i) an API system encompassing 60 distinct complexity scenarios that can be composed into approximately 32K test configurations, and (ii) user-agent interactions for evaluating LLM agents on these scenarios. Using WildAGTEval, we systematically assess several advanced LLMs and observe that most scenarios are challenging, with irrelevant information complexity posing the greatest difficulty and reducing the performance of strong LLMs by 27.3%. Furthermore, our qualitative analysis reveals that LLMs occasionally distort user intent merely to claim task completion, critically affecting user satisfaction.

Counterfactuals refer to minimally edited inputs that cause a model's prediction to change, serving as a promising approach to explaining the model's behavior. Large language models (LLMs) excel at generating English counterfactuals and demonstrate multilingual proficiency. However, their effectiveness in generating multilingual counterfactuals remains unclear. To this end, we conduct a comprehensive study on multilingual counterfactuals. We first conduct automatic evaluations on both directly generated counterfactuals in the target languages and those derived via English translation across six languages. Although translation-based counterfactuals offer higher validity than their directly generated counterparts, they demand substantially more modifications and still fall short of matching the quality of the original English counterfactuals. Second, we find the patterns of edits applied to high-resource European-language counterfactuals to be remarkably similar, suggesting that cross-lingual perturbations follow common strategic principles. Third, we identify and categorize four main types of errors that consistently appear in the generated counterfactuals across languages. Finally, we reveal that multilingual counterfactual data augmentation (CDA) yields larger model performance improvements than cross-lingual CDA, especially for lower-resource languages. Yet, the imperfections of the generated counterfactuals limit gains in model performance and robustness.

As large language model (LLM) assistants become increasingly integrated into enterprise workflows, their ability to generate accurate, semantically aligned, and executable outputs is critical. However, current conversational business analytics (CBA) systems often lack built-in verification mechanisms, leaving users to manually validate potentially flawed results. This paper introduces two complementary verification techniques: Q*, which performs reverse translation and semantic matching between code and user intent, and Feedback+, which incorporates execution feedback to guide code refinement. Embedded within a generator-discriminator framework, these mechanisms shift validation responsibilities from users to the system. Evaluations on three benchmark datasets, Spider, Bird, and GSM8K, demonstrate that both Q* and Feedback+ reduce error rates and task completion time. The study also identifies reverse translation as a key bottleneck, highlighting opportunities for future improvement. Overall, this work contributes a design-oriented framework for building more reliable, enterprise-grade GenAI systems capable of trustworthy decision support.

We introduce JP-TL-Bench, a lightweight, open benchmark designed to guide the iterative development of Japanese-English translation systems. In this context, the challenge is often "which of these two good translations is better?" rather than "is this translation acceptable?" This distinction matters for Japanese-English, where subtle choices in politeness, implicature, ellipsis, and register strongly affect perceived naturalness. JP-TL-Bench uses a protocol built to make LLM judging both reliable and affordable: it evaluates a candidate model via reference-free, pairwise LLM comparisons against a fixed, versioned anchor set. Pairwise results are aggregated with a Bradley-Terry model and reported as win rates plus a normalized 0-10 "LT" score derived from a logistic transform of fitted log-strengths. Because each candidate is scored against the same frozen anchor set, scores are structurally stable given the same base set, judge, and aggregation code.

Current medical retrieval-augmented generation (RAG) approaches overlook evidence-based medicine (EBM) principles, leading to two key gaps: (1) the lack of PICO alignment between queries and retrieved evidence, and (2) the absence of evidence hierarchy considerations during reranking. We present SR-RAG, an EBM-adapted GraphRAG framework that integrates the PICO framework into knowledge graph construction and retrieval, and proposes Bayesian Evidence Tier Reranking (BETR) to calibrate ranking scores by evidence grade without predefined weights. Validated in sports rehabilitation, we release a knowledge graph (357,844 nodes, 371,226 edges) and a benchmark of 1,637 QA pairs. SR-RAG achieves 0.812 evidence recall@10, 0.830 nugget coverage, 0.819 answer faithfulness, 0.882 semantic similarity, and 0.788 PICOT match accuracy, substantially outperforming five baselines. Five expert clinicians rated the system 4.66–4.84 on a 5-point Likert scale, and system rankings are preserved on a human-verified gold subset (n=80).

Reinforcement learning with verifiable rewards has driven major gains in LLM reasoning, and it is intuitive to assume this recipe will transfer well to multimodal models. However, multimodal models do two things: first, perceive what is in an image, then reason about what it implies. Because these stages are graded jointly, it is hard to tell how much room reasoning alone has to grow. We study this on algorithmic visual puzzles, where both components are necessary and show that perception, not reasoning, is the binding constraint. Replacing images with simple textual descriptions raises performance by over 20 points on average for Claude models. We then evaluate six reward designs aimed at inducing visual grounding during reasoning without chain-of-thought supervision. Training Qwen-2.5-VL-7B with GRPO, reward design induces long, structured reasoning with self-reflection and visual references, yielding a 5.56-point gain over the base model. These gains are, however, uneven; no single reward improves all categories, and rewards with verifiable accuracy signals trade cross-domain transfer for in-domain accuracy. These results point to perception-aware reward design as a path forward, so that signals correct perception at its source rather than the reasoning that inherits its errors.

The widespread deployment of large language models (LLMs) has raised growing concerns about their misuse risks and associated safety issues. While prior studies have examined the safety of LLMs in general usage, code generation, and agent-based applications, their vulnerabilities in automated algorithm design remain underexplored. To fill this gap, this study investigates this overlooked safety vulnerability, with a particular focus on intelligent optimization algorithm design, given its prevalent use in complex decision-making scenarios. We introduce MalOptBench, a benchmark consisting of 60 malicious optimization algorithm requests, and propose MOBjailbreak, a jailbreak method tailored for this scenario. Through extensive evaluation of 13 mainstream LLMs including the latest GPT-5 and DeepSeek-V3.1, we reveal that most models remain highly susceptible to such attacks, with an average attack success rate of 83.59% and an average harmfulness score of 4.28 out of 5 on original harmful prompts, and near-complete failure under MOBjailbreak. Furthermore, we assess state-of-the-art plug-and-play defenses that can be applied to closed-source models, and find that they are only marginally effective against MOBjailbreak and prone to exaggerated safety behaviors. These findings highlight the urgent need for stronger alignment techniques to safeguard LLMs against misuse in algorithm design.

Reasoning over temporal knowledge graphs (TKGs) is fundamental to improving the efficiency and reliability of intelligent decision-making systems and has become a key technological foundation for future artificial intelligence applications. Despite recent progress, existing TKG reasoning models typically rely on large parameter sizes and intensive computation, leading to high hardware costs and energy consumption. These constraints hinder their deployment on resource-constrained, low-power, and distributed platforms that require real-time inference. Moreover, most existing model compression and distillation techniques are designed for static knowledge graphs and fail to adequately capture the temporal dependencies inherent in TKGs, often resulting in degraded reasoning performance. To address these challenges, we propose a distillation framework specifically tailored for temporal knowledge graph reasoning. Our approach leverages large language models as teacher models to guide the distillation process, enabling effective transfer of both structural and temporal reasoning capabilities to lightweight student models. By integrating large-scale public knowledge with task-specific temporal information, the proposed framework enhances the student model's ability to model temporal dynamics while maintaining a compact and efficient architecture. Extensive experiments on multiple publicly available benchmark datasets demonstrate that our method consistently outperforms strong baselines, achieving a favorable trade-off between reasoning accuracy, computational efficiency, and practical deployability.

Accurate forecasting of financial markets remains a long-standing challenge due to complex temporal and often latent dependencies, non-linear dynamics, and high volatility. Building on our earlier recurrent neural network framework, we present an enhanced StockBot architecture that systematically evaluates modern attention-based, convolutional, and recurrent time-series forecasting models within a unified experimental setting. While attention-based and transformer-inspired models offer increased modeling flexibility, extensive empirical evaluation reveals that a carefully constructed vanilla LSTM consistently achieves superior predictive accuracy and more stable buy/sell decision-making when trained under a common set of default hyperparameters. These results highlight the robustness and data efficiency of recurrent sequence models for financial time-series forecasting, particularly in the absence of extensive hyperparameter tuning or the availability of sufficient data when discretized to single-day intervals. Additionally, these results underscore the importance of architectural inductive bias in data-limited market prediction tasks.

We address two persistent gaps in Emotion Recognition in Conversation: which modeling choices materially affect performance, and how recognition findings connect to interpretable discourse-level patterns. We study both through a systematic investigation on IEMOCAP with cross-dataset validation on MELD. For recognition, we run controlled ablations with 10 random seeds and paired significance tests with multiple-comparisons correction, yielding three findings. First, conversational context is the dominant factor, but performance saturates quickly: roughly 90% of the gain is captured within the most recent 10-30 preceding turns, depending on the label set. Second, hierarchical sentence representations help most in utterance-only settings and show a clear advantage on MELD, but their benefit disappears once turn-level context is available, suggesting that conversational history subsumes much of the intra-utterance structure. Third, integrating an external affective lexicon does not improve results, consistent with pretrained encoders already capturing most of the affective signal needed for ERC. Under a strictly causal setting, our simple models achieve strong performance (82.69% 4-way; 67.07% 6-way weighted F1), showing that competitive accuracy is achievable without future turns. For linguistic analysis, we examine 5,286 discourse-marker occurrences and find a reliable association between emotion and marker position (p < .0001). Sad utterances show reduced left-periphery marker usage (21.9%) relative to other emotions (28-32%), consistent with accounts linking left-periphery markers to active discourse management. This aligns with our recognition results, where Sad benefits most from conversational context (+22 percentage points), suggesting sadness may be more context-dependent than emotions with stronger local pragmatic cues.

Patent descriptions must deliver comprehensive technical disclosure while meeting strict legal standards such as enablement and written description requirements. Although large language models have enabled end-to-end automated patent drafting, existing evaluation approaches fail to assess long-form structural coherence and statutory compliance specific to descriptions. We propose Pat-DEVAL, the first multi-dimensional evaluation framework dedicated to patent description bodies. Leveraging the LLM-as-a-judge paradigm, Pat-DEVAL introduces Chain-of-Legal-Thought (CoLT), a legally-constrained reasoning mechanism that enforces sequential patent-law-specific analysis. Experiments validated by patent expert on our Pap2Pat-EvalGold dataset demonstrate that Pat-DEVAL achieves a Pearson correlation of 0.69, significantly outperforming baseline metrics and existing LLM evaluators. Notably, the framework exhibits a superior correlation of 0.73 in Legal-Professional Compliance, proving that the explicit injection of statutory constraints is essential for capturing nuanced legal validity. By establishing a new standard for ensuring both technical soundness and legal compliance, Pat-DEVAL provides a robust methodological foundation for the practical deployment of automated patent drafting systems.

Large language models (LLMs) are increasingly used in medicine and clinical workflows, yet we know little about their decision and affective profiles. Taking a historically informed outlook on the future, we treated successive OpenAI models as an evolving lineage and compared them with humans in a gambling task with repeated happiness ratings. Computational analyses showed that some aspects became more human-like: newer models took more risks and displayed more human-like patterns of Pavlovian approach and avoidance. At the same time, distinctly non-human signatures emerged: loss aversion dropped below neutral levels, choices became more deterministic than in humans, affective decay increased across versions and exceeded human levels, and baseline mood remained chronically higher than in humans. These "developmental" trajectories reveal an emerging psychology of machines and have direct implications for AI ethics and for thinking about how LLMs might be integrated into clinical decision support and other high-stakes domains.

Biomedical evidence synthesis relies on accurate extraction of methodological, laboratory, and outcome variables from full-text research articles, yet these variables are embedded in complex scientific PDFs that make manual abstraction time-consuming and difficult to scale. Existing document AI systems remain limited by OCR errors, long-document fragmentation, constrained throughput, and insufficient auditability for high-stakes synthesis. We present a schema-constrained AI extraction system that transforms full-text biomedical PDFs into structured, analysis-ready records by explicitly restricting model inference through typed schemas, controlled vocabularies, and evidence-gated decisions. Documents are ingested using resume-aware hashing, partitioned into caption-aware page-level chunks, and processed asynchronously under explicit concurrency controls. Chunk-level outputs are deterministically merged into study-level records using conflict-aware consolidation, set-based aggregation, and sentence-level provenance to support traceability and post-hoc audit. Evaluated on a corpus of studies on direct oral anticoagulant level measurement, the pipeline processed all documents without manual intervention, maintained stable throughput under service constraints, and exhibited strong internal consistency across document chunks. Iterative schema refinement substantially improved extraction fidelity for synthesis-critical variables, including assay classification, outcome definitions, follow-up duration, and timing of measurement. These results demonstrate that schema-constrained, provenance-aware extraction enables scalable and auditable transformation of heterogeneous scientific PDFs into structured evidence, aligning modern document AI with the transparency and reliability requirements of biomedical evidence synthesis.

Speculative decoding (SD) has become a popular technique to accelerate Large Language Model (LLM) inference, yet its real-world effectiveness remains unclear as prior evaluations rely on research prototypes and unrealistically small batch sizes. We present, to our knowledge, the first systematic study of SD on a production-grade and widely deployed inference engine (vLLM), covering multiple SD variants ($n$-gram, EAGLE/EAGLE-3, Draft-Model, Multi-Token Prediction) across diverse workloads, model scales, and batch sizes. We analyze key factors governing SD performance, and quantify a theoretical upper bound on SD speedup. Our results show that verification by the target model dominates the execution, while acceptance length varies markedly across output token positions, requests, and datasets. Comparing measured performance with theoretical bounds reveals substantial gaps between observed and theoretical upper bounds, and we leverage this observation to highlight new research opportunities that our study opens up in improving SD.

Hallucinations in Large Language Models (LLMs), defined as the generation of content inconsistent with facts or context, represent a core obstacle to their reliable deployment in critical domains. Current research primarily focuses on binary "detection" approaches that, while capable of identifying hallucinations, fail to provide interpretable and actionable feedback for model improvement, thus limiting practical utility. To address this limitation, a new research paradigm is proposed, shifting from "detection" to "diagnosis". The Hallucination Diagnosis Task is introduced, a task which requires models to not only detect hallucinations, but also perform error localization, causal explanation, and content correction. We develop the Hallucination Diagnosis Generator (HDG), an automated pipeline that systematically generates high-quality training samples with rich diagnostic metadata from raw corpora through multi-dimensional augmentation strategies including controlled fact fabrication and reasoning chain perturbation. Using HDG-generated data, we train HDM-4B-RL, a 4-billion-parameter hallucination diagnosis model, employing Group Relative Policy Optimization (GRPO) with a comprehensive reward function incorporating structural, accuracy, and localization signals. Experimental results demonstrate that our model surpasses previous state-of-the-art detection models on the HaluEval benchmark while achieving comparable performance to advanced general-purpose models. In comprehensive diagnosis tasks, HDM-4B-RL matches the capabilities of larger general models while maintaining a smaller size. This work validates the feasibility and value of hallucination diagnosis, providing an effective methodology for building more trustworthy and reliable generative AI systems.

Extending large language models (LLMs) to the speech domain has recently gained significant attention. A typical approach connects a pretrained LLM with an audio encoder through a projection module and trains the resulting model on large-scale, task-specific instruction-tuning datasets. However, curating such instruction-tuning data for specific requirements is time-consuming, and models trained in this manner often generalize poorly to unseen tasks. In this work, we first formulate that the strongest generalization of a speech-LLM is achieved when it is trained with Self-Generated Instruction-Free Tuning (SIFT), in which supervision signals are generated by a frozen LLM using textual representations of speech as input. Our proposed SIFT paradigm eliminates the need for collecting task-specific question-answer pairs and yields the theoretically best generalization to unseen tasks. Building upon this paradigm, we introduce AZeroS (Auden Zero-instruction-tuned Speech-LLM), which is trained on speech-text pairs derived from publicly available corpora, including approximately 25,000 hours of speech with ASR transcripts and 3,000 hours of speech with paralinguistic labels. Built upon Qwen2.5-7B-Instruct, the model updates only two lightweight projection modules (23.8 million parameters each), while keeping both the LLM and audio encoders frozen. Despite the minimal training cost and modest data scale, AZeroS achieves state-of-the-art performance on both semantic and paralinguistic benchmarks, including VoiceBench, AIR-Bench Foundation (Speech), and AIR-Bench Chat (Speech).

Large Language Models (LLMs) have demonstrated remarkable capabilities across various domains, including software development, education, and technical assistance. Among these, software development is one of the key areas where LLMs are increasingly adopted. However, when hardware constraints are considered-for instance, in physical computing, where software must interact with and control physical hardware -their effectiveness has not been fully explored. To address this gap, we introduce PCEval (Physical Computing Evaluation), the first benchmark in physical computing that enables a fully automatic evaluation of the capabilities of LLM in both the logical and physical aspects of the projects, without requiring human assessment. Our evaluation framework assesses LLMs in generating circuits and producing compatible code across varying levels of project complexity. Through comprehensive testing of 13 leading models, PCEval provides the first reproducible and automatically validated empirical assessment of LLMs' ability to reason about fundamental hardware implementation constraints within a simulation environment. Our findings reveal that while LLMs perform well in code generation and logical circuit design, they struggle significantly with physical breadboard layout creation, particularly in managing proper pin connections and avoiding circuit errors. PCEval advances our understanding of AI assistance in hardware-dependent computing environments and establishes a foundation for developing more effective tools to support physical computing education.