Papers

AI is rapidly advancing neuroscience, yet many laboratories fail to fully unleash its potential due to significant interdisciplinary barriers. While pre-trained neural models for physiological data are progressing quickly, their heterogeneous architectures and modality-specific constraints hinder systematic integration, selection, and evaluation. Despite recent advances in large language model (LLM)-based agent systems for intelligent scientific applications, existing approaches often still lack the domain expertise required to effectively select and coordinate diverse neuroscience pre-trained models and handle unique data types in this domain. We present NS-Copilot, an LLM-driven multi-agent system for neuroscience analysis that autonomously supports end-to-end workflows for diverse professional tasks. It unifies domain-specific pre-trained models and supports key neuroscience modalities, including EEG and extracellular spike data, through a natural-language interface. Given raw data and a task description, NS-Copilot orchestrates agents with specialized roles for planning, adaptive control, code generation, and result synthesis, enabling analysis without dataset-specific heuristics. We evaluate NS-Copilot on neuroscience benchmarks spanning Alzheimer's disease, Parkinson's disease, and working memory spike decoding. Across 8 trials per task, the system consistently outperforms strong baselines on the primary metric, demonstrating the ability of NS-Copilot for effective and scalable neuroscience analysis. Our code is publicly available at https://github.com/FrankLiu1102/ns-copilot.

Across three vision-language model architectures (LLaVA-1.5-7B, Qwen2.5-VL-7B, InternVL3-8B), we report a universal negative finding for mid-layer interpretability. On POPE – the benchmark common to all three – the mid layers encode the ground-truth answer in 68-91% of errors, yet this signal is not causally active for the final prediction: residual-stream patching yields 0% non-trivial flip at the layer level on all three architectures, and on two of three at the per-head level (Qwen: 0/12,600 patched forwards). The lone exception, InternVL3 layer-20 head-2, is a non-vocab, self-attending head whose effect is localized to that specific head (p < 1e-4). Despite the null, the errors separate operationally into three failure modes – Perception Failure, Encoded-but-Disconnected, Prior-Override – learnable above 60% on all three architectures, and the architecture's prior direction predicts which of two interventions elicits a category-specific response. We report these mitigation effects under oracle labels as evidence the categories are mechanistically real, not as a deployable method.

SALSA (Semi-Autonomous Literature Summarization Assistant) is an open- source, human-in-the-loop platform for extracting structured scientific datasets from multimodal literature sources. The software combines document parsing, large language models, optical character recognition, computer vision, figure digitization, and user-guided correction tools to recover structured information from text, tables, figures, and captions. Users can configure extraction stages, define dataset schemas, perform interventions on digitized figures, and export verified data for downstream analysis and machine learning. SALSA is designed to automate repetitive literature curation tasks while preserving oversight where expert judgment is required. By supporting customizable extraction workflows across diverse input types, the software provides a flexible framework for scalable, reliable data curation for materials research, and potentially across several disciplines. Users are responsible for ensuring that all inputs, extraction workflows, and downstream uses comply with applicable publisher agreements, copyright and licensing terms, institutional policies, and data-use requirements.

Empirical legal research often relies on turning research questions into structured data extracted from large collections of rulings and judgments. Designing the extraction schema and then extracting the data remain a manual, expertise-heavy bottleneck. We present schematize, an open-source multi-agent system that interactively turns a researcher's problem statement into a validated extraction schema that can later be used for autonomous extraction. Schematize couples (i) a clarification dialogue that elicits implicit expert intent, (ii) iterative schema generation, (iii) data-grounded refinement that tests the schema against documents, and (iv) chat-based post-editing. We evaluated the system with human legal professional, introducing our novel methodology, and schematize achieves top performance in most of tested configurations. While the system is designed to be domain-agnostic and applicable to any document collection, we tailor and evaluate it on legal research problems. We release schematize as a pip-installable Python package with full documentation.

Sofroniew et al. (2026) report that emotion concepts in Claude Sonnet 4.5 are represented as vectors whose geometry mirrors human affect psychology. We replicate the representational core of that study on the base pretrained model google/gemma-2-27b, inheriting every disclosed parameter, resolving unspecified steps by disclosed rules, and changing only the subject model. From 205,200 newly generated Claude Sonnet 4.5 stories matching the original corpus design, we extract 171 emotion vectors and recover the core results. The leading principal components form an affective circumplex (PC1 carries 26.7% of variance against the original study's ~27%, PC2 13.4% against ~14%), emotions cluster into similar intuitive families, and the geometry holds across a broad late-middle band. The valence axis aligns with human norms (r = 0.72, against 0.81) and is stable across scales and depth. Arousal aligns at r = 0.67 (against 0.66) but only at the full 171-emotion scale and late depth, so we do not classify it as replicated. Extending the original analysis, a 46-layer sweep locates a sharp seam at L22-26, where the geometry consolidates and the vocabulary readout becomes legible. An embedding-layer baseline finds much of the geometry already present in the static token embeddings, with arousal as the exception. On top-activating held-out text, the geometry predicts token-level co-activation at r = 0.907. At least 52% of vectors peak on structurally non-conceptual tokens, a measured floor for max-activation confounds. Even when a document contains the vector's emotion word, the peak lands on that word only 6.1% of the time. Because the subject is a base model and the stimuli are Claude-generated fiction, the recovered structure is a property of the pretrained representation of Claude-rendered emotion. The original's causal and assistant-facing analyses are out of scope. Code and data are released.

Multimodal Large Language Models (MLLMs) are increasingly deployed in safety-critical settings, yet estimating when their answers can be trusted remains underexplored. We systematically compare training-free verbalized, token-level and semantic uncertainty estimation methods on twelve MLLMs and five VQA benchmarks. We find that answer length decides which family wins: token-level entropy on short answers, verbalized methods on sentence-length responses, and semantic methods on long-form generation. Token-level entropy becomes far more informative when estimated at a higher temperature than the low ones used to generate answers, and asking a model to judge its own answer becomes more reliable with model size. Building on this, a length-aware recipe that additionally flags responses on which the model abstains outperforms every individual method on average at about a quarter of the cost of semantic methods, although no method attains the coverage required for safe deployment. These results provide practitioners with principled guidance for selecting estimators by answer length and model scale.

This exploratory pilot study evaluates the scope and perceived accuracy of personal information output from ongoing conversational interactions in generative AI systems using GPT-5.2 Instant and GPT-5.2 Thinking, categorized into three output types: Fact, Inference, and Confidence. Based on the evaluation results obtained from 15 Japanese participants, differences in model design have limited impact on personal information output tendencies. Compared with the Inference type, the Fact type shows a more conservative output pattern. Regarding attribute categories, the findings indicate that Core Personal attributes associated with identification are treated relatively conservatively, whereas Behavioral and Linguistic attributes show higher accuracy across both Fact and Inference outputs. Furthermore, Holistic Profile, Psychological and Cognitive, and Residual attributes are more readily inferred, even when not supported by explicit factual outputs. Notably, the lack of null outputs for these attributes in the Inference type suggests that such inferred profiles may be constructed from indirectly available contextual information. The findings may contribute to future discussions regarding privacy awareness and personal information inference in generative AI systems.

Significant challenges remain in AI-driven educational systems in balancing privacy preservation with accurate cognitive diagnosis. To overcome this, we propose a federated inference framework in which several commercial LLM APIs collaborate without requiring access to raw student data or proprietary model internals. Using multiple federated entities, such as LLaMA-3.3-70B, GPT-4o-mini, and Claude-3-Haiku, our framework builds upon a heterogeneous multi-LLM architecture. The predictions generated by these entities are combined with epsilon-local differential privacy by adding Laplace noise locally to each entity's prediction output before aggregation, while residual-based aggregation mitigates model heterogeneity. Our approach is predicated on an honest-but-curious trust paradigm in which API providers are presumed not to abuse submitted queries, and our differential privacy mechanism shields the published diagnostic results from external inference. We conduct rigorous privacy-utility analysis showing strong privacy guarantees with minimal accuracy loss, and extensive real-world evaluations across three educational benchmarks confirm the framework's practical usability and cross-domain generalizability.

A language model's prediction of its next token develops across layers, and lens methods track this process by decoding intermediate hidden states into tokens. But a lens reading reflects both the hidden state and the readout (the unembedding matrix) used to decode it. Many lenses are fit on a corpus, and we show that two lenses differing only in their fitting corpus can report different tokens for the same hidden states. We call this dependence corpus conditionality. To examine readout structure independently of the fitting corpus, we introduce Sparse Readout Prism (SRP), which decomposes the readout using only its weights and expresses any token logit or logit difference as a sum of contributions from sparse readout features. This reveals readout features as a new unit of analysis for lens readings, exposing structure that token identities can obscure and enabling comparisons across tokens, contexts, layers, and lenses. Replacing the original readout with SRP's sparse approximation reconstructs 8.9-17.3 percentage points more of the tested logit differences than the strongest of six baselines built on geometric relations among readout rows. Ablating features shifts logit differences in proportion to their SRP contributions. Although token readings vary with the fitting corpus, the dominant readout feature remains stable. Because SRP uses no corpus in its construction, it provides a control independent of the fitting corpus for lens analyses.

The attention prefilling phase of long-context LLM inference scales quadratically, making self-attention a severe computational bottleneck. Traditional sparse attention methods mitigate this through fixed patterns or offline profiling, but lack the flexibility to adapt to input-dependent attention structure. Recent dynamic methods address this by routing heads to sparse patterns in real-time, but rely on indirect routing proxies with overhead and budget allocation mechanisms that overlook the post-softmax mass hierarchy. We present CRISP (Cliff-awaRe Input-adaptive Sparse Prefilling), which identifies and addresses two structural challenges in this dynamic routing paradigm. First, we show that the routing decision can be read directly off the structure of the proxy attention map. We replace the Jensen-Shannon Divergence (JSD) routing with C_struct, a structural proxy that measures mass at Vertical-Slash compatible positions and reproduces JSD's routing decisions while eliminating both the pooled matmul and subsequent KL divergence overhead. Second, we formalize the post-softmax mass cliff and demonstrate theoretically that strictly cumulative coverage thresholds accumulate O(n) background noise at long contexts. CRISP navigates this via a sink-aware threshold grounded in the noise floor. Empirically, across InfiniteBench, RULER and LongBench on two model families, CRISP is the strongest sparse method overall and matches or exceeds exact dense attention on retrieval-heavy benchmarks, recovering up to +28.0 pp on retrieval tasks over baselines and achieving up to a 5.30x attention speedup at 512k tokens, driven primarily by our O(n) noise elimination during selection while preserving structural integrity.

Energy poverty is nearly absent from NLP-for-social-good, and the little existing work is either static retrieval/QA or relies on carbon-intensive cloud LLMs, a self-defeating "computational irony" for a humanitarian setting. We present EqGrid, a closed-loop simulation in which a low-frequency, open-weight LLM policy agent sets price and carbon bounds and targeted subsidies over a community of empirically-grounded household personas, while high-frequency multi-agent RL traders clear a continuous double auction constrained by a physical distribution grid (IEEE-33-bus with Dynamic Operating Envelopes). Our contribution is threefold and directly addresses how to measure the social impact of AI: (i) grounded personas (region-matched socio-demographics) whose load curves are checked for shape and level realism against real smart-meter data; (ii) formal energy-poverty equity metrics (Energy Burden, Gini of EB, LIHC) showing the intervention reduces burden inequality without raising net grid cost; and (iii) a compute-efficiency frontier that measures how much equity performance survives compressing the policy agent from a 235B teacher down to a sub-1B model deployable on a laptop, in estimated energy/carbon per decision. A decoupled-safety design (the LLM sets bounds; a validate-and-project grid gate executes) yields zero grid-constraint violations versus 55 under direct LLM control. On energy-poverty equity, the LLM policy lowers the Gini of energy burden to 0.305 (from 0.351) and mean burden by 28% while cutting cost (outperforming a tuned rule baseline), and a 3B-active model retains 95% of the benefit at roughly 9x lower inference energy than the teacher, with even a 0.8B on-device model retaining 92% at roughly 24x lower energy. We will release code and configs.

Assessments of cultural alignment have become an important part of the development and improvement of large language models (LLMs). However, the majority of the evaluations treat culture as a single snapshot, investigating only whether a model represents a society accurately at the current time. Research in cultural psychology shows that cultural values change at different rates and directions over time. Therefore, a "culturally aware" model should capture not only where a culture is today but also how it has changed over time. We examine this missing dimension of cultural awareness using more than two decades of the World Values Survey data. We compare the cultural trajectories of 40 countries with the trajectories produced by four state-of-the-art (SOTA) LLMs on the Inglehart-Welzel cultural map. Our findings show that while models generally place countries close to their most recent surveyed positions, these representations tend to lag several years behind that position. They also capture only part of the magnitude of the observed change, introduce movement where little occurred, and rarely reproduce reversals in countries' trajectories. These findings point to temporal flattening and suggest that snapshot accuracy can give an incomplete picture of cultural awareness in LLMs and have implications for model evaluation, representational harms, and the governance of culturally aware AI systems.

Activation steering suppresses undesired behaviors in language models by adding a steering vector to the hidden state during generation. Recent conditional methods such as CAST and DSAS improve the behavior-capability trade-off by deciding when to intervene, but once active, they apply the full dense vector to all hidden dimensions, regardless of whether a neuron carries concept information or already lies in the desired regime. We introduce dimension-level conditioning as a complementary axis of selectivity that also decides which neurons to intervene on. Our method, GAPS (Gated Activation steering via Posterior and Separability), combines two training-free gates: a static separability gate that restricts steering to neurons with statistically reliable concept information (via AUROC), and a dynamic posterior gate that steers a neuron only when its current activation is better explained by the undesired concept under a Gaussian model. The gates add O(D) overhead per token, and they plug into existing conditional methods. On toxicity mitigation (RealToxicityPrompts) and concept removal (OneSeC) with Gemma-3 (4B) and Qwen-3 (1.7B), GAPS consistently matches or improves the Pareto front of its token-level counterparts; under a fixed capability budget, DSAS+GAPS reduces Gemma-3's toxicity rate from 6.52% to 0.48%, versus 3.52% for DSAS alone. Ablations attribute most of the gain to the posterior gate.

A major question in cognitive modeling concerns the behavioral alignment between large language models and humans across linguistic and non-linguistic tasks. Unlike standard LLMs, large reasoning models (LRMs) are optimized with reinforcement learning from verifiable rewards, encouraging correct solutions to reasoning tasks rather than preference-aligned responses. Recent work (de Varda et al., 2025) investigates the cost of thinking in humans and LRMs by comparing human reaction times with model reasoning traces across a range of reasoning tasks. We isolate this alignment by turning to abductive reasoning: unlike deductive tasks, its difficulty cannot be inferred from formal structure and offers no shortcuts a model could exploit to mimic effort without genuine search, providing firmer ground for empirical claims of shared effort. We find further evidence of alignment between LRM and human reasoning effort, as well as evidence that models and humans tend to make similar errors. Finally, we show that decoding methods that let models explore multiple reasoning paths increase alignment in reasoning cost between humans and LRMs across the three models tested.

Embedding-based code retrieval is a core component of coding agents and retrieval-augmented code generation, where retrieving correct code matters more than retrieving lexically similar code. Existing code-retrieval benchmarks do not plant controlled, execution-verified single-edit variants of each query's canonical implementation in the search pool, leaving the question of whether embeddings can functionally discriminate correct from near-clone-but-incorrect code unanswered in a retrieval setting. Resolving this requires a benchmark whose search pool itself contains the relevant counterfactuals – execution-verified buggy variants near-identical to each canonical – so that a retriever's rank ordering can be directly tested for functional discrimination rather than topical or identity overlap. We introduce ExecRetrieval, 939 Python tasks each paired with one execution-verified canonical implementation and up to four execution-verified buggy distractors, each generated by a mechanical mutation making a single targeted edit, and evaluate 23 dense embedding configurations plus BM25 under provider-native invocation with paired McNemar tests and query-level bootstrap intervals. With near-clone counterfactuals in the pool, the top hosted system reaches exec@10 = 1.00 but only exec@1 = 0.331; rank-1 misses are paired buggy variants 91.5-99.4% of the time across the four leading systems, and the canonical scores below at least one of its four paired distractors in 67-78% of queries on the leading systems. The full dataset, execution oracle, embedding matrices, environment snapshot, and pairwise statistical tests are released at the URL in Appendix D.

Persistent memory supports personalized agents, but a stale stored fact can override current authoritative evidence without warning. We study when this harm begins as model capability changes. We evaluate a frozen, closed-set, action-scored benchmark with 2 suites that represent 2 different meanings of "no memory" (a Benefit suite, unsolvable without the stored fact, and a Safety suite, in which an authoritative tool always holds the correct value), on a same-family model-size series (Qwen3 0.6/1.7/4/8B). The Memory Trust Gap reflects over-trust rather than confusion. In the Benefit suite, models answer with the stale value 0.92-1.00 of the time at every scale. In the Safety suite, harm below the no-memory baseline under the trap conditions ($\Delta_{\mathrm{mem}}$) is capability-gated, with the larger models collapsing most once a stale note is made to look current. In a $2\times2\times2\times2$ factorial, which feature triggers over-trust depends on both the feature and model scale. Removing a label amplifies over-trust at every size, and a recency feature (stale dated newer) fools the larger models harder. Source authority is weak and scale-flat, and position changes from positive to negative across the Qwen3 model-size series. We confirm these scale interactions with direct cross-size contrast tests rather than overlapping per-model intervals. Mitigation is likewise capability-dependent: exposing metadata improves accuracy for the capable models, but only pre-resolving the conflict restores accuracy for the 2 smaller checkpoints. The same pattern appears on the capable models in an independent Llama-Instruct model-size series and on 2 external datasets (RGB, MisBench). A framing control finds no consistent advantage for the memory label: at the 3 smaller scales, models trust a stale document more than a stale memory; at 8B, the difference is not significant.

Recorded lecture videos, often enhanced with search and summarization features, are a standard study resource. However, students cannot easily ask course specific questions or verify answers against an instructor's lecture. We report a semester-long deployment of VideoPoints platform with a retrieval-augmented chatbot that answers from course lecture materials and returns timestamped citations. The chatbot retrieves only from the active course, uses chapter summaries to guide transcript ranking, and returns clickable timestamped citations. Students used it for quick lookups and exam review. Across 833 messages, 70.5% included citations, none crossed a course boundary, and when no lecture evidence matched, the chatbot usually declined rather than answering. Among the users, citations were the most consistently useful feature, while practice-question generation was the strongest unmet request. We also evaluated the design on the real-world test split of EduVidQA, a public multimodal benchmark for lecture-video question answering. Our design improved correct-lecture retrieval by 6.3 percentage points over dense-only retrieval. Together, the results show that effective deployment depends on course isolation, supported citations, and alignment with students' study practices.

Sentence-level recognition of depression symptoms is challenging because similar expressions can differ in symptom relevance, and language-model inference is insufficiently grounded in diagnostic definitions. This study proposes a two-stage framework separating symptom-candidate generation from definition-grounded verification. A contrastively fine-tuned sentence encoder generates a symptom candidate per sentence, and a fine-tuned language model verifies whether the candidate is present or absent using the sentence, its context, and a candidate-specific diagnostic definition, checking its judgment against that definition before answering. Evaluated against encoder, inference-based, medical, and general LLM baselines and a matched single-stage supervised classifier, the proposed pipeline attains the best accuracy and F1 scores of all methods, with rationales matching expert-authored annotations. A preliminary clinical audit indicates moderate alignment with diagnostic definitions, with explanation quality strongly dependent on prediction correctness. The results support decomposing symptom recognition into candidate generation and definition-grounded verification, though performance remains limited for rare categories.

Patients with depression present with diverse symptom profiles, yet clinical practice routinely reduces this variation to a single severity score. Large language models (LLMs) can potentially capture various symptoms and their severity from patient speech. However, how depressive symptoms are represented inside LLMs remains poorly understood, limiting clinical trust. To examine whether internal model activations match clinician judgment, we analyzed the residual stream of Gemma-3-27B-PT using mechanistic interpretability techniques. Recording activations across symptom descriptions drawn from validated clinical instruments, we found that symptom groups geometrically separated the most at layer 21 across multiple distance metrics. Using Semantic Projection, we then projected held-out naturalistic text onto Symptom Vectors constructed from these instruments. The resulting per-symptom coefficients preserved clinician-annotated rank ordering across mood, somatic, and suicidality axes. Furthermore, a single depression vector in Layer 21 separates held-out depressive from non-depressive text (AUC = 0.789), which can be used as an emotional valence gate that restricts symptom projection to depressive speech. These results reveal a decorrelated, clinician-aligned symptom signal readable directly from internal activations, offering a mechanistic foundation for interpretable depression-assessment tools.

Spoken dialogue state tracking recovers slot-value pairs from speech, where ASR errors concentrate in entity values and persist across turns, making it both a generation and an editing problem. A strong per-turn text editor corrects much of this but, operating on the transcript alone, leaves three recoverable errors: a value predicted inconsistently across turns, an omitted slot, and a value the audio does not support. We present AVERT, which scores each candidate value by combining cross-turn agreement with a trained audio-conditioned verifier and resolves the three error types with three operators, vote, add, and swap, each restricted to the slots where its error is common. On SpokenWOZ, a base speech-LLM reaches 33.04 JGA, a text editor 38.34, and AVERT 40.13, without retraining either. This is in the range of a 1B end-to-end system that consumes the full spoken history (39.32), though AVERT uses two 1B decoders rather than one. The audio verifier contributes a statistically significant gain, and restricting each operator to a selected slot subset matters: removing it lets unrestricted voting overwrite correct categorical values and fall below the editor.

Farsi, spoken by more than 120 million people, lacks a comprehensive benchmark for dialogue generation and understanding. We introduce TALKFA, a unified benchmark comprising three complementary datasets: (1) WIKI-FADIAL, 4.2K Wikipedia-grounded dialogues for knowledge-grounded generation; (2) DAILYDIALOG-FA, 6.6K dialogues annotated for dialogue acts and emotions; and (3) PLAYDIAL-FA, 2.1K theatrical dialogues with sentiment labels. While LLMs assist data construction, every dialogue undergoes multi-stage review and revision by native Farsi speakers, and only the final human-approved dialogues are released. Experiments with six LLAMA and MISTRAL models show that LoRA substantially improves dialogue generation while requiring only 25-50% of the training data to recover over 90% of the final performance gains. Across classification tasks, FABERT achieves the best dialogue-act performance, LORA-MISTRAL-7B performs best on emotion recognition, and MISTRAL-24B achieves the highest sentiment score. Human evaluation and independent external validation demonstrate the reliability of the benchmark, while comparisons with GPT-4.1 as an LLM judge reveal that automatic metrics substantially overestimate dialogue quality. Zero-shot evaluation with frontier LLMs further shows that TalkFa remains a challenging benchmark. We will release all datasets, annotation guidelines, code, and checkpoints.

Large language models (LLMs) are increasingly deployed on mobile devices, making energy efficiency a key deployment constraint, yet the energy impact of prompt design remains underexplored. This paper aims to understand how two prompt properties, cognitive load and phrasing pattern, shape the energy behavior of on-device LLM inference. We conduct a broad empirical study covering prompt properties, datasets, models, and devices, with phase-level profiling that separates prefill and decode energy. We find that cognitive load primarily affects the energy cost per token, while phrasing pattern affects energy largely through token usage. Our energy-quality analysis further shows that prompt design reshapes the attainable frontier differently across models, highlighting the need for model-aware prompt design in energy-efficient on-device LLM inference. Code, datasets, and scripts are available at https://amai-gsu.github.io/PromptProperty/.

How do learners avoid overgeneralizations such as Tom laughed me without explicit negative evidence? Constructionists have posited two proposals that describe indirect negative evidence against overgeneralizations: preemption (which privileges exposure to near-synonymous construction—e.g., she made him laugh) vs. entrenchment (all exposures to a verb's grammatical usages, including cases like He laughed). We disentangle these hypotheses by running controlled rearing experiments on LMs trained on child-caregiver conversations, where we systematically remove preemptive vs. non-preemptive evidence. We find that while LMs avoid overgeneralizations, they do not show preemption at a verb-specific level, instead showing weak but non-zero evidence of abstract preemption. Combined with results from analyzing the LMs' training dynamics, we find that LMs treat competing structures as indirect positive—as opposed to negative—evidence in the verb-specific condition. Insofar as preemption is the more plausible route to avoiding overgeneralizations in humans, our results point the need for there to be sensitivities to indirect negative evidence in neural network learners, and suggest new human experiments to test abstract preemption.

Real-world communication often requires pragmatic reasoning: interpreting meanings implied through context and cultural convention rather than stated literally. Existing pragmatic evaluation remains largely limited to English and high-resource languages, leaving Indic languages unexplored despite their linguistic and cultural diversity. We introduce VakyArth, the first pragmatic benchmark for Indic languages, designed as a diagnostic evaluation covering Hindi, Punjabi, Tamil, and Malayalam. VakyArth evaluates models across five phenomena: deixis, speech acts, implicature, social pragmatics, and coherence; through multiple-choice questions, natural language inference, and translation, with all items authored by native speakers. Across multilingual large language models (LLMs) of varying families and sizes, we find consistent failures on pragmatic meanings rooted in Indic linguistic and cultural conventions. Our analysis shows systematic differences across languages and tasks: MCQ accuracy exceeds NLI accuracy in all model-language combinations, translation performance does not reliably track pragmatic understanding, and Indo-Aryan languages show a translation advantage over Dravidian languages. We further show that automatic translation metrics can miss fluent but pragmatically unfaithful outputs, especially for implicature and deixis.

Meme understanding goes beyond recognizing visual content or literal text; it requires implicit cultural knowledge and pragmatic inference that most vision-language models still lack. We introduce MemeCULT-1K, a multilingual benchmark of 1,000 South Asian memes in Bengali, English, and Hindi, where each meme is paired with a cultural context note and three human-written explanations, along with a supplementary set of 54 Bengali regional dialect memes. We evaluate thirteen popular Vision Language Models (VLMs) under two settings: meme-only and context-aware. Providing minimal cultural context yields consistent gains across all models and languages: mean SBERT similarity improves from 44.6 to 56.4 (+11.8), BLEURT from 37.3 to 42.3 (+5.0), and LLM-as-a-Judge scores from 2.57 to 3.43 out of 5 (+0.86). Fine-grained error analysis reveals that closed-source models fail mainly on entity and reference misidentification, while open-source models are bottlenecked by broader cultural knowledge gaps, with linguistic and phonological failures proving the most context-resistant across both. These results highlight the difficulty of culturally grounded meme understanding and motivate future work on explicit cultural knowledge integration. Our dataset and code are publicly available at TawsifDipto17/MemeCULT-1K.

Statutes are increasingly parsed by machines before people read them, and the parsers disagree: on Missouri's statutes, two independently written extractors diverge on numeric-threshold presence at a false-negative rate of 0.43. We ask what formal logic survives such noise. We build a passive survival certificate for the Duquenne-Guigues implication basis of machine-extracted statutory contexts: per-attribute inter-extractor disagreement is measured, replayed against the basis in 1,000 Monte Carlo trials, and an implication is certified only when a one-sided Wilson 95% lower bound on survival reaches 0.95; every certified implication carries premise spans and a minimal counterexample. On 29,365 Missouri sections and 502 Indian central-Act sections, the preregistered held-out gate passes (10 statute families across 7 Titles exact; 16 across 11 with 5% tolerance), yet under one globally deployed error model 93.2% of held-out chapters fall below the informativeness floor, and a 2x2 factorial assigns that to calibration-rate transfer, not selection. The certificate is usable but fragile: deploy it per-chapter-calibrated or error-tolerant. Code, data products, and the audit trail, including one retracted claim, are released.

Mobile payment applications in Nepal are graphically mediated and largely inaccessible to visually impaired users. This paper presents SpeakPay, a voice-first digital wallet, and documents the central technical contribution: a controlled study of domain adaptation for low-resource financial speech recognition. We introduce NepFinSpeech-403, a 403-utterance dataset of Nepali financial voice commands (send, load, and balance operations spanning 237 unique numerals), and fine-tune Whisper large-v2 with LoRA. On the held-out test set, the domain-adapted model reduces Word Error Rate from 129.95% (zero-shot baseline) to 42.58% — a 67.2% relative reduction — and improves Devanagari numeral recognition accuracy from 0.0% to 73.9%. We find that word-level metrics understate the practical task-level impact: domain adaptation improves the Transaction Success Rate from 1.67% to 33.33%, a roughly 20x gain. The improvement is consistent at the individual-utterance level (sign test, $p < 10^{-17}$) and across all command types. A data efficiency analysis shows that as few as 100 domain-specific utterances are sufficient to halve the zero-shot WER, with performance plateauing around 300 examples. Error analysis reveals systematic numeral confusion patterns (zero insertion/deletion, prefix hallucination) that account for the majority of remaining transaction failures. The trained system is deployed as a publicly accessible voice-first web application. All code, dataset, model weights, and this paper are released at https://github.com/subedibiraj/speakpay.

Large language models (LLMs) augmented with external tools have demonstrated remarkable capability in solving complex real-world tasks. However, existing approaches suffer from two key challenges: brittle multi-step and multi-turn reasoning caused by incompatible tool output types and API schemas, and performance degradation under large tool catalogues. To address these, we introduce Tool Primitives, a design that replaces rigid API schema-based invocation with natural language as the interface for tool calling, where each tool is wrapped with an LLM interface that handles schema resolution and execution internally, enabling natural inter-tool communication for nested and multi-turn tool calling. Building on Tool Primitives, we host ToolFace, a centralized repository of 25,519 functions from which LLMs dynamically retrieve only the relevant tools at inference time, eliminating the need to enumerate raw API schemas in context. To orchestrate Tool Primitives and ToolFace reliably in complex settings, we further propose HEART, a Harness Engineering framework via Agent-native, Reusable Tool Primitives, comprising a Planner, Router, and Verifier that jointly support dynamic tool invocation planning, multi-step execution, and feedback-driven recovery. Experiments on five benchmarks demonstrate that HEART outperforms SFT-based models by $10\%$ on average and surpasses GPT-5.4, Claude-4.6-Sonnet, and Gemini-3.1-Pro by $6\%$ on average while reducing API cost by up to $85\%$. On 50 real-world tasks, HEART achieves $84\%$ task completion, $3.8\times$ the average of three frontier commercial models ($22\%$).

Grounded question answering systems should answer only when the supplied evidence supports the answer. In multi-hop QA, this requirement is difficult because partial evidence can make an unsupported answer appear plausible. We study selective answering through evidence sufficiency boundaries: for the same question, a model should abstain under unsupported or partially supported context, answer when the context first becomes sufficient, and keep the answer stable when redundant evidence is added. We introduce Evidence Sufficiency Boundary Training, a generation-native training framework that constructs ordered evidence chains and supervises the abstain-to-answer transition directly. The method combines level supervision, a boundary flip margin, post-boundary stability, and answer recall protection. We build evidence chains from HotpotQA, 2WikiMultiHopQA, and MuSiQue, then evaluate models with chain metrics, raw QA utility, and unsupported-answer rates on external non-answerable sets. With Qwen2.5-3B-Instruct and LoRA adaptation, Evidence Sufficiency Boundary Training gives the strongest boundary localization among the tested systems, with flip accuracy of 0.807 compared with 0.781 for a token-level abstention baseline. It also achieves the lowest overall unsupported-answer rate on external non-answerable evaluation, 0.095 compared with 0.101 for the same baseline, while retaining competitive raw QA F1. The results show that grounded selective answering improves when training marks the evidence level where refusal should give way to answering.

Security teams and researchers choose knowledge-graph extraction tooling for threat reports on the strength of published triple-F1 scores, yet those scores depend on how predicted triples are matched to gold annotations. We could reimplement the stated matching rule for only five of twelve inspected systems. Re-scoring ten system outputs on shared documents under eight protocols reverses eleven of forty-five pairwise orderings; one fixed prediction set spans 0.16–0.70 F1. On an external, human-adjudicated set, no mechanical matcher—lexical, embedding, or entailment—agrees with the reviewers more than seven times in ten; an LLM judge agrees far more often. To separate component effects from matcher rewards, we build CTIForge, whose deterministic validation layer can vary while extraction is held byte-identical. Across seven tested deployment configurations, no hosted backbone loses precision under validation and every offline one does. Backbone, decoding, and prompting covary across those configurations; on the one backbone we could serve both ways, serving alone reproduces the split. It coincides with a roughly 2.8-fold increase in actions explicitly disputing entity type, consistent with hand-written rules encoding the conventions of the extractor against which they were developed. We release the pipeline, protocol suite, and per-triple audit records.