Named after the hundred-eyed watchman of Greek myth, Argus watches the education landscape: spotting new opportunities, pressure-testing the ventures we're building, and tracing every read back to the real-world signals behind it.
The evidence library: the raw signals the pipeline is watching across the education ecosystem. Every idea is built from these.
arXiv:2605.31483v3 Announce Type: replace Abstract: Despite Bengali being the sixth most spoken language in the world, no prior work has systematically evaluated hallucination in large language models (LLMs) for Bengali. We introduce BenHalluEval, a fine-grained hallucination evaluation framework for Bengali covering four tasks: Generative Question Answering (GQA), Bangla-English Code-Mixed QA, Summarization, and Reasoning. We construct 12,000 hallucinated candidates using GPT-5.4 across twelve task-specific hallucination types, drawn from three existing Bengali datasets, and evaluate seven LLMs spanning reasoning-oriented, multilingual, and Bengali-centric categories under a dual-track protocol that independently measures false-positive rate on ground-truth instances (Track A) and hallucination detection rate on hallucinated candidates (Track B). To jointly penalise both failure modes and prevent inflated scores from uniform response bias, we propose BenHalluScore, a dual-track calibr
arXiv:2605.30880v3 Announce Type: replace Abstract: Text-agent environments are typically modeled as partially observable Markov decision processes (POMDPs), assuming that the simulator's latent state and transition dynamics are hidden from the agent. Yet little work has examined whether executable code can be induced to serve as a world model for prediction and planning under partial observability. We introduce PatchWorld, a gradient-free framework that turns offline trajectories into executable Python world models through counterexample-guided code repair. Instead of predicting the next observation with a black-box model, PatchWorld induces symbolic belief-state programs whose action updates can be inspected, replayed, and locally patched. Across seven AgentGym environments, PatchWorld-Simple achieves the highest code-based planning score among evaluated methods, reaching 76.4% macro success in live one-step lookahead while invoking no LLM calls inside the world-model prediction modu
arXiv:2605.30295v2 Announce Type: replace Abstract: Large language models (LLMs) show promise for clinical reasoning and decision support, but evaluation in realistic, electronic health record-congruent settings remains limited. Existing benchmarks often rely on static datasets or unstructured inputs that do not reflect the structured, interoperable data formats used in clinical systems. We introduce a pipeline for generating clinically realistic HL7 FHIR R4 bundles from unstructured text, enabling controllable evaluation of clinical decision support systems. The pipeline combines staged LLM generation with terminology-grounded validation and repair to reduce hallucinated codes and enforce structural and semantic consistency. Applying this approach to MedCaseReasoning, we construct MedCase-Structured, a synthetic dataset aligned with clinician-authored diagnostic cases, achieving valid FHIR generation for 82.5% of cases. Evaluation on MedCase-Structured reveals consistently lower diagn
arXiv:2605.28512v2 Announce Type: replace Abstract: Self-evolving scientific agents capable of conquering the hard tail of formal mathematics require Compositional Learning Behaviours (CLBs) -- the capacity to ground and recombine novel symbolic structures in context, beyond mere recombination of prelearned atoms. We propose S2B-LM, an adaptation of the CLB-evaluating Symbolic Behaviour Benchmark that removes numerical processing as a confound and adds chain-of-thought scaffolding to elicit rather than merely probe latent CLB competency. Cross-evaluating ten Lean~4 theorem provers on CLB competency in S2B-LM and miniF2F whole-proof performance, we find correlational and causal evidence of our claim: First, a necessary-condition analysis via quadrant test yields $p=0.004$, with model scale being ruled out as a confound. Second, extracting a CLB-encoding activation direction from DeepSeek-Prover-V2-7B using S2B-LM traces via Contrastive Activation Addition and applying it during miniF2F
arXiv:2605.26755v3 Announce Type: replace Abstract: Multilingual fact verification requires evidence that is both relevant and sufficiently complete for reliable factuality prediction. However, existing systems often rely on search snippets, sentence-level evidence, or locally segmented passages, which can miss decisive context and produce fragmented evidence. To overcome these limitations, we propose SEEK, a Semantic Evidence Extraction with an adaptive chunKing framework that constructs coherent evidence chunks from full fact-checking articles by identifying semantic topic transitions and preserving local verification context. The constructed chunks are encoded using a multilingual encoder and then multilingual LLMs are finetuned using LoRA adapter for veracity prediction. Experiments on X-FACT and RU22Fact show that SEEK improves macro-f1 by up to 10% over semantic chunking, 19% over sentence chunking, and 20% over search-snippet baselines. Evidence completeness and significance ana
arXiv:2605.23071v2 Announce Type: replace Abstract: Large language models (LLMs) increasingly rely on long-context processing, but expanding context windows introduces substantial computational and financial costs. Existing context reduction approaches, including retrieval and memory compression methods, are typically evaluated using performance and efficiency metrics independently, limiting systematic comparison and deployment-aware decision-making. This paper introduces The Efficiency Frontier, a unified framework for cost--performance optimization in LLM context management. The framework models context strategy selection as a deployment-aware optimization problem that jointly accounts for task performance, token cost, and preprocessing reuse through amortized cost modeling. Unlike existing evaluations that compare methods in isolation, the proposed framework enables decision-oriented analysis. It identifies when different context management strategies become preferable under varying
arXiv:2605.20712v2 Announce Type: replace Abstract: Automatic speech recognition replaces typing only when correction costs less than manual entry - a threshold determined by error types, not counts: fixing a misrecognized domain term costs far more than inserting a comma. Word error rate (WER) fails on two fronts: it collapses distinct error categories into a single scalar, and it structurally penalizes agglutinative languages where valid sandhi merges inflate scores. We introduce SCRIBE, a diagnostic framework offering categorical error decomposition into lexical, punctuation, numeral, and domain-entity rates via sandhi-tolerant alignment with domain vocabulary injection. Human validation confirms SCRIBE aligns with expert judgment where WER does not. We release SCRIBE, an LLM curation pipeline, benchmarks, and open-weight rich transcription models for Hindi, Malayalam, and Kannada.
arXiv:2605.13087v2 Announce Type: replace Abstract: Fine-tuning multilingual ASR models like Whisper for low-resource languages often improves read speech but degrades spontaneous audio performance. To diagnose this mismatch, we introduce Vividh-ASR, a complexity-stratified benchmark for Hindi and Malayalam across four tiers: studio, broadcast, spontaneous, and synthetic noise. Through a controlled study of learning-rate timing and curriculum ordering, we find that early large parameter updates improve global WER by 12 absolute points, while a hard-to-easy curriculum adds gains for spontaneous speech. These findings motivate reverse multi-stage fine-tuning (R-MFT), a training recipe that enables a parameter-efficient 244M Whisper model to match or exceed conventionally fine-tuned 769M counterparts. Representational analysis via CKA and SVD reveals effective schedules concentrate adaptation in the decoder, preserving the pre-trained encoder's acoustic geometry. We release the benchmark
arXiv:2605.09253v3 Announce Type: replace Abstract: While recent work in Reinforcement Learning with Verifiable Rewards (RLVR) has shown that a small subset of critical tokens disproportionately drives reasoning gains, an analogous token-level understanding of On-Policy Distillation (OPD) remains largely unexplored. In this work, we investigate high-loss tokens, a token type that--as the most direct signal of student-teacher mismatch under OPD's per-token KL objective--should progressively diminish as training converges according to existing studies; however, our empirical analysis shows otherwise. Even after OPD training reaches apparent saturation, a substantial subset of tokens continues to exhibit persistently high loss; these tokens, which we term Rock Tokens, can account for up to 18\% of the tokens in generated outputs. Our investigation reveals two startling paradoxes. First, despite their high occurrence frequency providing a disproportionately large share of total gradient no
arXiv:2605.00994v2 Announce Type: replace Abstract: Finetuning can significantly modify the behavior of large language models, including introducing harmful or unsafe behaviors. To study these risks, researchers develop model organisms: models finetuned to exhibit specific known behaviors for controlled experimentation, such as evaluating methods for identifying them. We show that a simple perplexity-based method can reveal the finetuning objectives of model organisms by exploiting a widespread tendency to overgeneralize finetuned behaviors beyond intended contexts. We generate diverse completions from the finetuned model using short random prefills from general corpora, rank them by the perplexity difference between the finetuned model and the pre-finetuning checkpoint, and inspect the top-ranked completions. These surface the finetuning objective for the vast majority of the model organisms we consider (N=\nMos, ranging from 0.5 to 70B parameters), including backdoored models, models
arXiv:2604.22503v2 Announce Type: replace Abstract: Hundreds of millions of people use artificial intelligence (AI) for writing assistance. Here, we evaluated how AI writing assistance distorts writer personas - their perceived beliefs, personality, and identity. In three large-scale experiments, writers (N=2,939) wrote political opinion paragraphs with and without AI assistance. Separate groups of readers (N=11,091) blindly evaluated these paragraphs across 29 socially salient dimensions of reader perception, spanning political opinion, writing quality, writer personality, emotions, and demographics. AI writing assistance produced persona distortions across all dimensions: with AI, writers seemed more opinionated, competent, and positive, and their perceived demographic profile shifted towards more privileged groups. Writers objected to many of the observed distortions, yet continued to prefer AI-assisted text even when made aware of them. We successfully mitigated objectionable perso
arXiv:2604.05318v2 Announce Type: replace Abstract: Harmful content detectors, particularly disinformation classifiers, are predominantly developed and evaluated on Standard American English (SAE), leaving their robustness to dialectal variation unexplored. We present DIA-HARM, the first benchmark for evaluating disinformation detection robustness across 50 English dialects spanning U.S., British, African, Caribbean, and Asia-Pacific varieties. Using Multi-VALUE's linguistically grounded transformations, we introduce D-CUBE (Dialectal Disinformation Detection Corpus), a core corpus component of DIA-HARM comprising 195K samples derived from established disinformation benchmarks. Our evaluation of 16 detection models reveals systematic vulnerabilities: human-written dialectal content degrades detection by 1.4-3.6% F1, while AI-generated content remains stable. Fine-tuned transformers substantially outperform zero-shot LLMs (96.6% vs. 78.3% best-case F1), with some models exhibiting catas
arXiv:2604.04385v5 Announce Type: replace Abstract: We localize the policy routing mechanism in alignment-trained language models. An intermediate-layer attention gate reads detected content and triggers deeper amplifier heads that boost the signal toward refusal. In smaller models the gate and amplifier are single heads; at larger scale they become bands of heads across adjacent layers. The gate contributes under 1% of output DLA, yet interchange testing (p < 0.001) and knockout cascade confirm it is causally necessary. Interchange screening at n >= 120 detects the same motif in twelve models from six labs (2B to 72B), though specific heads differ by lab. Per-head ablation weakens up to 58x at 72B and misses gates that interchange identifies; at scale, interchange is the only reliable audit. Modulating the detection-layer signal continuously controls policy from hard refusal through evasion to factual answering. On safety prompts the same intervention turns refusal into harmful guidan
arXiv:2604.02176v3 Announce Type: replace Abstract: While textual frequency has been validated as relevant to human cognition in reading speed, its relatedness to Large Language Models (LLMs) is seldom studied. We propose a novel research direction in terms of textual data frequency, which is an understudied topic, to the best of our knowledge. Our framework is composed of three units. First, this paper proposes Textual Frequency Law (TFL), which indicates that frequent textual data should be preferred for LLMs for both prompting and fine-tuning. Since many LLMs are closed-source in their training data, we propose using online resources to estimate the sentence-level frequency. We then utilize an input paraphraser to paraphrase the input into a more frequent textual expression. Next, we propose Textual Frequency Distillation (TFD) by querying LLMs to conduct story completion by further extending the sentences in the datasets, and the resulting corpora are used to adjust the initial est
arXiv:2604.00835v2 Announce Type: replace Abstract: Large language models are increasingly being deployed as autonomous agents yet their real world effectiveness depends on reliable tools for information retrieval, computation and external action. Existing studies remain fragmented across tasks, tool types, and training settings, lacking a unified view of how tool-use methods differ and evolve. This paper organizes the literature into three paradigms: prompting as plug-and-play, supervised tool learning and reward-driven tool policy learning, analyzes their methods, strengths and failure modes, reviews the evaluation landscape and highlights key challenges, aiming to address this fragmentation and provide a more structured evolutionary view of agentic tool use.
arXiv:2603.26815v3 Announce Type: replace Abstract: Retrieval-Augmented Generation (RAG) systems for financial document QA typically follow a chunk-based paradigm: documents are split into fragments, embedded, and retrieved by similarity. In structurally homogeneous corpora such as regulatory filings, this suffers from cross-document chunk confusion. Semantic File Routing (SFR), which uses LLM structured output to route queries to whole documents, reduces catastrophic failures but sacrifices targeted-chunk precision. We identify this robustness-precision trade-off on the FinDER benchmark (1,500 queries across five groups): SFR achieves higher average scores (6.45 vs. 6.02) and fewer failures (10.3% vs. 22.5%), while chunk-based retrieval (CBR) yields more perfect answers (13.8% vs. 8.5%). To resolve it, we propose Hybrid Document-Routed Retrieval (HDRR), a two-stage architecture that uses SFR as a document filter followed by chunk retrieval scoped to the identified document(s), elimina
arXiv:2603.16856v2 Announce Type: replace Abstract: The prevailing paradigm for improving large language models relies on offline training with human annotations or simulated environments, leaving the rich experience accumulated during real-world deployment entirely unexploited. We propose Online Experiential Learning (OEL), a framework that enables language models to continuously improve from their own deployment experience. OEL operates in two stages: first, transferable experiential knowledge is extracted and accumulated from interaction trajectories collected on the user side; second, this knowledge is consolidated into model parameters via on-policy context distillation, requiring no access to the user-side environment. The two stages are iterated to form an online learning loop, where the improved model collects higher-quality trajectories that yield richer experiential knowledge for subsequent rounds. We evaluate OEL on text-based game environments across multiple model scales a
arXiv:2603.15389v2 Announce Type: replace Abstract: Recent work has demonstrated the curse of depth in large language models (LLMs), where later layers contribute less to learning and representation than earlier layers. Such under-utilization is linked to the accumulated growth of variance in Pre-Layer Normalization, which can push deep blocks toward near-identity behavior. In this paper, we provide evidence that sparsity-like mechanisms can dampen variance propagation and are associated with improved depth utilization Our investigation covers two sources of sparsity: (i) implicit sparsity, which emerges from training and data conditions, including weight sparsity induced by weight decay and attention sparsity induced by long-context inputs; and (ii) explicit sparsity, which is enforced by architectural design, including key/value-sharing in Grouped-Query Attention and expert-activation sparsity in Mixtureof-Experts. Our claim is thoroughly supported by controlled depth-scaling experim
arXiv:2603.15130v3 Announce Type: replace Abstract: Indirectness is a common feature of daily communication, yet is underexplored in NLP research for both low-resource as well as high-resource languages. Indirect Question Answering (IQA) aims at classifying the polarity of indirect answers. In this paper, we present two multilingual corpora for IQA of varying quality that both cover English, Standard German and Bavarian, a German dialect without standard orthography: InQA+, a small high-quality evaluation dataset with hand-annotated labels, and GenIQA, a larger training dataset, that contains artificial data generated by GPT-4o-mini. We find that IQA is a pragmatically hard task that comes with various challenges, based on several experiment variations with multilingual transformer models (mBERT, XLM-R and mDeBERTa). We suggest and employ recommendations to tackle these challenges. Our results reveal low performance, even for English, and severe overfitting. We analyse various factors
arXiv:2603.12050v2 Announce Type: replace Abstract: Translations systematically diverge from texts originally produced in the target language, a phenomenon widely referred to as translationese. Translationese has been attributed to production tendencies (e.g. interference, simplification), socio-cultural variables, and language-pair effects, yet a unified explanatory account is still lacking. We propose that translationese reflects cognitive load inherent in the translation task itself. We test whether observable translationese can be predicted from quantifiable measures of translation task difficulty. Translationese is operationalised as a segment-level translatedness score produced by an automatic classifier. Translation task difficulty is conceptualised as comprising source-text and cross-lingual transfer components, operationalised mainly through information-theoretic metrics based on LLM surprisal, complemented by established syntactic and semantic alternatives. We use a bidirecti
arXiv:2603.09785v3 Announce Type: replace Abstract: This paper introduces an updated and combined version of the bidirectional English-German EPIC-UdS (spoken) and EuroParl-UdS (written) corpora containing original European Parliament speeches as well as their translations and interpretations. The new version corrects metadata and text errors identified through previous use, refines the content, updates linguistic annotations, and adds new layers, including word alignment and word-level surprisal indices. The combined resource is designed to support research using information-theoretic approaches to language variation, particularly studies comparing written and spoken modes, and examining disfluencies in speech, as well as traditional translationese studies, including parallel (source vs. target) and comparable (original vs. translated) analyses. The paper outlines the updates introduced in this release, summarises previous results based on the corpus, and presents a new illustrative s
arXiv:2603.08195v2 Announce Type: replace Abstract: Motivation: The rapid growth of biological data has intensified the need for transparent, reproducible, and well-documented computational workflows. The ability to clearly connect the steps of a workflow in the code with their description in a paper would improve workflow comprehension, support reproducibility, and facilitate reuse. This task requires the linking of bioinformatics tools in workflow code with their mentions in a published workflow description. Results: We present CoPaLink, an automated approach that integrates three components: named entity recognition (NER) for identifying tool mentions in scientific text, NER for tool mentions in workflow code, and entity resolution based on word embedding similarity. We propose approaches for all three steps, achieving a high individual F1-measure (77 - 90) and a joint accuracy of 66 when evaluated on Nextflow workflows using Sentence-BERT. CoPaLink leverages corpora of scientific a
arXiv:2603.03915v2 Announce Type: replace Abstract: Large Language Models (LLMs) have shown remarkable potential in developing role-playing agents (RPAs). However, current evaluation frameworks rely heavily on well-known fictional characters, raising a critical concern: models may be leveraging their internal training memory of these characters rather than demonstrating role-playing capabilities. This reliance often leads to significant performance degradation when RPAs encounter unseen or out-of-distribution personas. To address this, we propose a more rigorous evaluation protocol designed to decouple role-playing proficiency from character recognition. Our experiments across multiple benchmarks demonstrate that anonymizing characters degrades performance, confirming that name exposure provides implicit cues that mask a model's true capability. To mitigate this, we investigate diverse personality augmentation as a method to enhance role fidelity in anonymous settings. We systematicall
arXiv:2603.03335v2 Announce Type: replace Abstract: Large language models (LLMs) exhibit a wide range of capabilities, including mathematical reasoning, code generation, and linguistic behaviors. We show that Transformer architectures contain small subsets of attention heads that are necessary for certain capabilities. Zeroing out as few as five task-specific heads can degrade performance by up to $60\%$ on standard benchmarks measuring the capability of interest, while largely preserving performance on unrelated tasks. We introduce a compressed sensing-based method that exploits the sparsity of these heads to identify them via strategic knockouts and a small number of model evaluations. We validate these findings across Llama and Qwen models ranging from 1B to 14B parameters and a diverse set of capabilities including mathematical abilities and code generation, revealing a modular organization in which specialized capabilities are dependent on sparse, functionally distinct components.
arXiv:2603.03305v2 Announce Type: replace Abstract: Large language models (LLMs) are increasingly used to generate executable outputs, JSON objects, and API calls, where a single syntax error can make the output unusable. Constrained decoding enforces validity token-by-token via masking and renormalization, but it can distort generation when the model assigns low probability mass to valid continuations, pushing decoding toward locally valid yet semantically incorrect trajectories. We propose \emph{Draft-Conditioned Constrained Decoding (DCCD)}, a simple two-step, training-free inference procedure that decouples semantic planning from structural enforcement: an unconstrained draft is generated first, and constrained decoding is then applied, conditioned on this draft, to guarantee validity. We analyze DCCD through a KL-projection view, showing that draft conditioning increases feasible mass and reduces the cumulative "projection tax" induced by hard constraints, with an optional best-of
arXiv:2602.21608v2 Announce Type: replace Abstract: Bangla-English code-mixing is widespread across South Asian social media, yet resources for implicit meaning identification in this setting remain scarce. Existing sentiment and sarcasm models largely focus on monolingual English or high-resource languages and struggle with transliteration variation, cultural references, and intra-sentential language switching. To address this gap, we introduce MixSarc, the first publicly available Bangla-English code-mixed corpus for implicit meaning identification. The dataset contains 9,087 manually annotated sentences labeled for humor, sarcasm, offensiveness, and vulgarity. We construct the corpus through targeted social media collection, systematic filtering, and multi-annotator validation. We benchmark transformer-based models and evaluate zero-shot large language models under structured prompting. Results show strong performance on humor detection but substantial degradation on sarcasm, offens
arXiv:2602.02498v2 Announce Type: replace Abstract: Large language models can produce toxic or inappropriate text even for benign inputs, creating risks when deployed at scale. Detoxification is therefore important for safety and user trust, particularly when we want to reduce harmful content without sacrificing the model's generation quality. Many existing approaches rely on model retraining, gradients, or learned auxiliary components, which can be costly and may not transfer across model families or to truly black-box settings. We introduce a test-time procedure that approximates the gradient of completion toxicity with respect to the input embeddings and uses a small number of descent steps to steer generation toward less toxic continuations. This is achieved with zeroth-order optimization that requires only access to input embeddings, a toxicity scoring function, and forward evaluations of the model. Empirically, the approach delivers robust toxicity reductions across models and pr
arXiv:2602.02320v4 Announce Type: replace Abstract: Molecular function is largely determined by structure. Accurately aligning molecular structure with natural language is therefore essential for enabling large language models (LLMs) to reason about downstream chemical tasks. However, the substantial cost of human annotation makes it infeasible to construct large-scale, high-quality datasets of structure-grounded descriptions. In this work, we propose a fully automated annotation framework for generating precise molecular descriptions that preserve complete structural details at scale. Our approach builds upon and extends a rule-based chemical nomenclature parser to interpret IUPAC names and construct enriched, structural XML metadata that explicitly encodes molecular structure. This metadata is then used to guide LLMs in producing accurate natural-language descriptions. Using this framework, we curate a large-scale dataset of approximately $163$k molecule--description pairs. A rigorou
arXiv:2601.15251v2 Announce Type: replace Abstract: Large language models (LLMs) have achieved impressive proficiency in basic arithmetic, rivaling human-level performance on standard numerical tasks. However, little attention has been given to how these models perform when numerical expressions deviate from the prevailing conventions present in their training corpora. In this work, we investigate numerical reasoning across a wide range of numeral scripts and formats. We show that LLM accuracy drops substantially when numerical inputs are rendered in underrepresented scripts or formats, despite the underlying mathematical reasoning being identical. We further demonstrate that targeted prompting strategies, such as few-shot prompting and explicit numeral mapping, can greatly narrow this gap. Our findings highlight an overlooked challenge in multilingual numerical reasoning and provide actionable insights for working with LLMs to reliably interpret, manipulate, and generate numbers acros
arXiv:2601.12033v2 Announce Type: replace Abstract: Quantization is widely adopted to reduce the computational cost of large language models (LLMs); however, its implications for fairness and safety, particularly in dynamic quantization and multilingual contexts, remain underexplored. In this work, we conduct a systematic study of how static and dynamic quantization methods impact fairness and safety across benchmarks measuring intrinsic and extrinsic bias and safety alignment. For fairness, we evaluate English, French, Dutch, Spanish, and Turkish; for safety, we focus on English, Korean, and Arabic. Our findings reveal that quantization consistently degrades fairness and safety, with dynamic methods demonstrating greater stability than static ones. Moreover, fairness degradation varies across languages, while safety deterioration is especially pronounced in non-English settings. To address these risks, we introduce Critical Weight Protection, a novel technique that identifies and pres
arXiv:2601.07988v2 Announce Type: replace Abstract: While NLP typically treats documents as independent and unordered samples, in longitudinal studies, this assumption rarely holds: documents are nested within authors and ordered in time, forming person-indexed, time-ordered $\textit{behavioral sequences}$. Here, we demonstrate the need for and propose a longitudinal modeling and evaluation paradigm that consequently updates four parts of the NLP pipeline: (1) evaluation splits aligned to generalization over people ($\textit{cross-sectional}$) and/or time ($\textit{prospective}$); (2) accuracy metrics separating between-person differences from within-person dynamics; (3) sequence inputs to incorporate history by default; and (4) model internals that support different $\textit{coarseness}$ of latent state over histories (pooled summaries, explicit dynamics, or interaction-based models). We demonstrate the issues ensued by traditional pipeline and our proposed improvements on a dataset o
arXiv:2601.06972v2 Announce Type: replace Abstract: In speech language modeling, two architectures dominate the frontier: the Transformer and the Conformer. However, it remains unknown whether their comparable performance stems from convergent processing strategies or distinct architectural inductive biases. We introduce Architectural Fingerprinting, a probing framework that isolates the effect of architecture on representation, and apply it to a controlled suite of 24 pre-trained encoders (39M-3.3B parameters). Our analysis reveals divergent hierarchies: Conformers implement a "Categorize Early" strategy, resolving phoneme categories 29% earlier in depth and speaker gender by 16% depth. In contrast, Transformers "Integrate Late," deferring phoneme, accent, and duration encoding to deep layers (49-57%). These fingerprints suggest design heuristics: Conformers' front-loaded categorization may benefit low-latency streaming, while Transformers' deep integration may favor tasks requiring r
arXiv:2601.05366v2 Announce Type: replace Abstract: Large Language Models (LLMs) are increasingly deployed as agents that invoke external tools through structured function calls. While recent work reports strong tool-calling performance under standard English-centric evaluations, the robustness of tool calling under multilingual user interactions remains underexplored. In this work, we introduce MLCL, a diagnostic benchmark, and conduct a systematic evaluation of multilingual tool calling across Chinese, Hindi, and the low-resource language Igbo. Through fine-grained error analysis, we show that many failures occur despite correct intent understanding and tool selection. We identify parameter value language mismatch as a dominant failure mode, where models generate semantically appropriate parameter values in the user's language, violating language-invariant execution conventions. We further evaluate several inference-time system strategies and find that while these strategies substant
arXiv:2601.04693v2 Announce Type: replace Abstract: Although negation is known to challenge large language models (LLMs), benchmarks for evaluating negation understanding-especially in Korean-are scarce. We conduct a corpus-based analysis of Korean negation and show that LLM performance degrades under negation. We then introduce Thunder-KoNUBench, a sentence-level negation understanding benchmark that reflects the empirical distribution of Korean negation phenomena. Evaluating 47 LLMs on Thunder-KoNUBench, we analyze the effects of model size and instruction tuning, and perform error analysis to better understand model behavior. We further show that fine-tuning on Thunder-KoNUBench improves negation understanding and broader contextual comprehension in Korean.
arXiv:2512.19612v2 Announce Type: replace Abstract: This paper introduces MauBERT, a multilingual extension of HuBERT that leverages articulatory features for robust cross-lingual phonetic representation learning. We continue HuBERT pre-training with supervision based on a phonetic-to-articulatory feature mapping in 55 languages. Our models learn from multilingual data to predict articulatory features or phones, resulting in language-independent representations that capture multilingual phonetic properties. Through comprehensive ABX discriminability testing, we show MauBERT models produce more context-invariant representations than state-of-the-art multilingual self-supervised learning models. Additionally, the models effectively adapt to unseen languages and casual speech with minimal self-supervised fine-tuning (10 hours of speech). This establishes an effective approach for instilling linguistic inductive biases in self-supervised speech models.
arXiv:2511.06516v4 Announce Type: replace Abstract: Many LLM applications require only narrow capabilities, yet standard post-training quantization (PTQ) methods allocate precision without considering the target task. This can waste bits on layers that are less relevant to the task signal while over-compressing layers that are critical for downstream behavior. We propose Task-Aware Quantization (TAQ), a training-free, weight-only mixed-precision PTQ framework that uses a small set of unlabeled task calibration prompts to allocate higher precision to task-relevant transformer layers under a fixed bit budget. TAQ estimates layer importance from hidden representations and output sensitivity, and we instantiate it with three scoring rules: TAQ-IS, based on activation information and stability; TAQ-KL, based on output-distribution sensitivity under a quantization-noise proxy; and TAQ-O, a label-informed oracle diagnostic for analyzing layer sensitivity. Across several benchmarks, TAQ outper
arXiv:2511.05852v4 Announce Type: replace Abstract: Knowledge editing (KE) offers a lightweight alternative to retraining for updating large language models (LLMs). Meanwhile, fine-tuning remains the default operation for adapting LLMs to new domains and tasks. Despite their widespread adoption, these two post-training interventions have been studied in isolation, leaving open a crucial question: if we fine-tune an edited model, do the edits survive? This question is motivated by practical objectives: removing covert or malicious edits, and preserving beneficial edits. If fine-tuning impairs edits (Fig.1), current KE methods become less efficient, as a newly fine-tuned model requires re-editing; if edits persist, fine-tuned models risk propagating hidden malicious edits, raising serious safety concerns. To this end, we systematically quantify edit decay after fine-tuning across 254 experimental configurations. Our results show that in general, edits decay substantially after subsequent
arXiv:2510.25013v2 Announce Type: replace Abstract: Mechanistic interpretability aims to reverse-engineer large language models (LLMs) into human-understandable computational circuits. However, the complexity of pretrained models often obscures the minimal mechanisms required for specific reasoning tasks. In this work, we train small, attention-only transformers from scratch on a symbolic version of the Indirect Object Identification (IOI) task, a benchmark for studying coreference-like reasoning in transformers. Surprisingly, a single-layer model with only two attention heads achieves perfect IOI accuracy, despite lacking MLPs and normalization layers. Through residual stream decomposition, spectral analysis, and embedding interventions, we find that the two heads specialize into additive and contrastive subcircuits that jointly implement IOI resolution. Furthermore, we show that a two-layer, one-head model composes information from the previous layer primarily through query-key inter
arXiv:2510.09278v2 Announce Type: replace Abstract: Training expert LLMs in domains with scarce data is difficult, often relying on multiple-choice questions (MCQs). However, standard outcome-based reinforcement learning (RL) on MCQs is risky. While it may improve accuracy, we observe it often degrades reasoning quality such as logical consistency. Existing solutions to supervise reasoning, such as large-scale Process Reward Models (PRMs), are prohibitively expensive. To address this, we propose CLARity, a cost-effective RL framework that enhances reasoning quality using only a small, general-purpose LLM. CLARity integrates a consistency-aware reward mechanism with a 2-stage refine-then-monitor training pipeline to enhance reasoning consistency, and a dynamic data reformulation strategy to to better exploit limited data. Experiments demonstrate that CLARity improves response consistency by 16.5% and accuracy by 7.5% over baselines. Human evaluations further confirm holistic improvement
arXiv:2510.06732v2 Announce Type: replace Abstract: Large language models (LLMs) are increasingly used as rerankers in information retrieval, yet their ranking behavior can be steered by small, natural-sounding prompts. To expose this vulnerability, we present Rank Anything First (RAF), a two-stage token optimization method that crafts concise textual perturbations to consistently promote a target item in LLM-generated rankings while remaining hard to detect. Stage 1 uses Greedy Coordinate Gradient to shortlist candidate tokens at the current position by combining the gradient of the rank-target with a readability score; Stage 2 evaluates those candidates under exact ranking and readability losses using an entropy-based dynamic weighting scheme, and selects a token via temperature-controlled sampling. RAF generates ranking-promoting prompts token-by-token, guided by dual objectives: maximizing ranking effectiveness and preserving linguistic naturalness. Experiments across multiple LLMs
arXiv:2509.02292v2 Announce Type: replace Abstract: What if large language models could not only infer human mindsets but also expose every blind spot in team dialogue such as discrepancies in the team members' joint understanding? We present a novel, two-step framework that leverages large language models (LLMs) both as human-style annotators of team dialogues to track the team's shared mental models (SMMs) and as automated discrepancy detectors among individuals' mental states. In the first step, an LLM generates annotations by identifying SMM elements within task-oriented dialogues from the Cooperative Remote Search Task (CReST) corpus. Then, a secondary LLM compares these LLM-derived annotations and human annotations against gold-standard labels to detect and characterize divergences. We define an SMM coherence evaluation framework for this use case and apply it to six CReST dialogues, ultimately producing: (1) a dataset of human and LLM annotations; (2) a reproducible evaluation f
arXiv:2508.06482v2 Announce Type: replace Abstract: Humans communicate with increasing efficiency in multi-turn interactions, by adapting their language and forming ad-hoc conventions. In contrast, prior work shows that LLMs do not naturally show this behavior. We develop a post-training process to develop this ability through targeted fine-tuning on heuristically identified demonstrations of convention formation. We evaluate with two new benchmarks focused on this capability. First, we design a focused, cognitively-motivated interaction benchmark that consistently elicits strong convention formation trends in humans. Second, we create a new document-grounded reference completion task that reflects in-the-wild convention formation behavior. Our studies show significantly improved convention formation abilities in post-trained LLMs across the two evaluation methods.
arXiv:2507.23220v2 Announce Type: replace Abstract: Traditional topic models are effective at uncovering latent themes in large text collections. However, due to their reliance on bag-of-words representations, they struggle to capture semantically abstract features. While some neural variants use richer representations, they are similarly constrained by expressing topics as word lists, which limits their ability to articulate complex topics. We introduce Mechanistic Topic Models (MTMs), a class of topic models that operate on interpretable features learned by sparse autoencoders (SAEs). By defining topics over this semantically rich space, MTMs can reveal deeper conceptual themes with expressive feature descriptions. Moreover, uniquely among topic models, MTMs enable controllable text generation using topic steering vectors. To properly evaluate MTM topics against word list approaches, we propose \textit{topic judge}, an LLM-based pairwise comparison evaluation framework. Across eight
arXiv:2507.05257v4 Announce Type: replace Abstract: Recent benchmarks for Large Language Model (LLM) agents primarily focus on evaluating reasoning, planning, and execution capabilities, while another critical component-memory, encompassing how agents memorize, update, and retrieve long-term information-is under-evaluated due to the lack of benchmarks. We term agents with memory mechanisms as memory agents. In this paper, based on classic theories from memory science and cognitive science, we identify four core competencies essential for memory agents: accurate retrieval, test-time learning, long-range understanding, and selective forgetting. Existing benchmarks either rely on limited context lengths or are tailored for static, long-context settings like book-based QA, which do not reflect the interactive, multi-turn nature of memory agents that incrementally accumulate information. Moreover, no existing benchmarks cover all four competencies. We introduce MemoryAgentBench, a new bench
arXiv:2507.02804v2 Announce Type: replace Abstract: Recent progress in large-scale reinforcement learning (RL) has notably enhanced the reasoning capabilities of large language models (LLMs), especially in mathematical domains. However, current multimodal LLMs (MLLMs) for mathematical reasoning often rely on one-to-one image-text pairs and single-solution supervision, overlooking the diversity of valid reasoning perspectives and internal reflections. In this work, we introduce MathV-DP, a novel dataset that captures multiple diverse solution trajectories for each image-question pair, fostering richer reasoning supervision. We further propose Qwen-VL-DP, a model built upon Qwen-VL, fine-tuned with supervised learning and enhanced via group relative policy optimization (GRPO), a rule-based RL approach that integrates correctness discrimination and diversity-aware reward functions. Our method emphasizes learning from varied reasoning perspectives and distinguishing between correct yet dis
arXiv:2506.05121v3 Announce Type: replace Abstract: A recent line of research on spoken language assessment (SLA) employs neural models such as BERT and wav2vec 2.0 (W2V) to evaluate speaking proficiency across linguistic and acoustic modalities. Although both models effectively capture features relevant to oral competence, each exhibits modality-specific limitations. BERT-based methods rely on ASR transcripts, which often fail to capture prosodic and phonetic cues for SLA. In contrast, W2V-based methods excel at modeling acoustic features but lack semantic interpretability. To overcome these limitations, we propose a system that integrates W2V with Phi-4 multimodal large language model (MLLM) through a score fusion strategy. The proposed system achieves a root mean square error (RMSE) of 0.375 on the official test set of the Speak & Improve Challenge 2025, securing second place in the competition. For comparison, the RMSEs of the top-ranked, third-ranked, and official baseline systems
arXiv:2506.00400v4 Announce Type: replace Abstract: LLM-based prompt optimization, which uses LLM-provided ``textual gradients'' (feedback) to refine prompts, has emerged as an effective method for automatic prompt engineering. However, its scalability and stability are unclear when using more data in training. We systematically investigate the potential and challenges of scaling training data in textual gradient descent. We show that naively scaling training examples is infeasible due to both explicit context-length limits and an implicit context wall, where long-context degradation yields diminishing returns. Inspired by prior wisdom in stochastic gradient descent, we propose Textual Stochastic Gradient Descent with Momentum (TSGD-M), which reweights updates through momentum sampling, using bootstrapped minibatch validation accuracy as importance weights over historical prompts. To stabilize TSGD and enable effective scaling within a limited context window, TSGD-M carries prior promp
arXiv:2504.14969v2 Announce Type: replace Abstract: This paper proposes a framework for evaluating large language models (LLMs) on Chinese topic constructions, focusing on their sensitivity to island constraints. Drawing inspiration from Tian et al. (2024), we outline an experimental design for testing LLMs' grammatical knowledge of Mandarin syntax. While no experiments have been conducted yet, this proposal aims to provide a foundation for future studies and invites feedback on the methodology.
arXiv:2503.16550v2 Announce Type: replace Abstract: Neural network language models (LMs) are confronted with significant challenges in generalization and robustness. Currently, many studies focus on improving either generalization or robustness in isolation, without methods addressing both aspects simultaneously, which presents a significant challenge in developing LMs that are both robust and generalized. In this paper, we propose a bi-stage optimization framework to uniformly enhance both the generalization and robustness of LMs, termed UEGR. Specifically, during the forward propagation stage, we enrich the output probability distributions of adversarial samples by adaptive dropout to generate diverse sub models, and incorporate JS divergence and adversarial losses of these output distributions to reinforce output stability. During backward propagation stage, we compute parameter saliency scores and selectively update only the most critical parameters to minimize unnecessary deviatio
arXiv:2502.11491v3 Announce Type: replace Abstract: Large language models (LLMs) have shown remarkable capabilities in natural language processing. However, in knowledge graph question answering tasks (KGQA), there remains the issue of answering questions that require multi-hop reasoning. Existing methods rely on entity vector matching, but the purpose of the question is abstract and difficult to match with specific entities. As a result, it is difficult to establish reasoning paths to the purpose, which leads to information loss and redundancy. To address this issue, inspired by human reverse thinking, we propose Ontology-Guided Reverse Thinking (ORT), a novel framework that constructs reasoning paths from purposes back to conditions. ORT operates in three key phases: (1) using LLM to extract purpose labels and condition labels, (2) constructing label reasoning paths based on the KG ontology, and (3) using the label reasoning paths to guide knowledge retrieval. Experiments on the WebQ