Text Corpora

As language-model compute continues to scale, high-quality training data is becoming an increasingly important bottleneck. Conventional next-token prediction supervises what follows a context but leaves the intermediate reasoning behind that continuation implicit. We introduce \textbf{REER-PT}, a scalable framework that extends Reverse-Engineered Reasoning (REER) to raw pre-training data. REER-PT identifies continuations that are difficult to predict but can still be inferred from the preceding context, and inserts concise reasoning annotations that reconstruct the missing connection between context and continuation. Candidate annotations are generated and refined offline, with perplexity serving as the optimization signal. Constraints on length and target leakage filter out unhelpful or trivial annotations. This sparse transformation preserves the source text and remains compatible with standard next-token prediction, avoiding online reasoning rollouts during pre-training. We apply REER-PT to transform a source pre-training corpus into an augmented one. Across augmented-data, original-token, and selected-continuation comparisons, perplexity reductions range from 0.42 to 7.29, and only about 0.05% of annotation 13-grams appear verbatim in the source text. We then train two 680M-parameter models with the same architecture and training configuration on the source and augmented corpora, respectively. The augmented-data model gains up to 2.07 percentage points on several knowledge and reasoning benchmarks. Together, the perplexity analysis indicates improved continuation predictability, while the controlled pre-training experiments suggest that this augmentation can improve model performance without changing the standard pre-training objective.

Large language models are increasingly capable in general, but their utility can remain modest in niche or understudied areas. One approach to address this limitation is to specialise existing models through additional training on target-domain corpora. In this work, we investigate such continued pre-training for adapting large language models to Swedish journalism, using a high-quality dataset that we curate from millions of news articles. To evaluate the adaptation efficacy, we also construct a novel domain-specific benchmark that covers six editorial tasks. Through full and parameter-efficient fine-tuning across two model sizes, we find that continued pre-training yields benefits in the target domain, but only when paired with experience replay to mitigate forgetting. We observe consistent enhancements in the models' generation quality and factual knowledge, but not their proficiency in discriminative tasks. Exploring a training-free method to facilitate instruction following, we see further improvements, but exclusively for models trained with low-rank adaptation. Crucially, we demonstrate the importance of targeted evaluation in the adaptation process, as an existing Swedish benchmark largely fails to capture the models' in-domain performance gains.

The language a legislature uses to debate women's rights, even in favour of them, encodes systematic patterns of sexism that persist across two centuries. In this work, we analyse 6,531 speeches over 200 years of UK parliamentary debate (Hansard, 1803-2005) by using large language models to classify a speaker's perspective towards women's suffrage and political representation, as well as analyse sexist speech in parliament from the lens of the Ambivalent Sexism Inventory. We also release this parliamentary dataset, an organized and metadata-enriched version of the publicly available Hansard Corpus optimized for computational social science research, with 6.7 million speeches across 1.2 million debates, with 89% gender-matching for speeches by MPs from the House of Commons. We find that 54% of speeches opposing women's representation contain sexist content, compared to 21% of speeches that are for the cause, and that the two sides use fundamentally different types of sexism: anti-suffrage rhetoric combines hostile and benevolent framing, while pro-suffrage sexism is overwhelmingly benevolent. Female MPs support women's political rights at 93% compared to 70% for male MPs, a gap that closes only after enfranchisement. Our findings are evidence that benevolent and hostile sexism are used in different rhetorical contexts in a manner consistent with the theory of Ambivalent Sexism.

A central bottleneck in multi-hop Question Answering (QA) is that the granularity at which a question is expressed often differs from the granularity at which corpus evidence is retrievable. Existing methods address this mismatch by imposing fixed graph structures over the corpus, by iteratively reformulating the query, or by executing a generated program over it, but these strategies do not explicitly decide when a query unit is already supported by evidence and when it should be refined. We formulate this bottleneck as retrievable granularity discovery and introduce Hi-Q, an evidence-conditioned framework for hierarchical query refinement. At each query node, a resolution operator tests whether retrieved evidence supports the current query unit; resolved nodes terminate, while unresolved nodes are expanded by a dependency-preserving binary operator and checked by a semantic coverage verifier. Hi-Q therefore grows a query tree whose topology is determined by corpus support signals rather than by a fixed decomposition template or a pre-built graph. We evaluate Hi-Q on three multi-hop QA benchmarks, primarily under full-corpus retrieval, where dependent evidence must be located among open-domain distractors rather than within a small annotated pool. In this setting Hi-Q reaches 52.3 EM and 64.0 F1 averaged over the three benchmarks, ahead of the iterative retrieval baseline IRCoT by 15.1 EM / 18.2 F1 on that same average, and ahead of the graph-based RAG baseline PropRAG by 11.5 EM / 12.0 F1 on MuSiQue-full, without corpus-wide graph construction. In the restricted supporting/distractor setting used by prior work, Hi-Q likewise attains the best accuracy, with 57.9 EM and 69.3 F1 on average, ahead of PropRAG by 5.6 EM / 3.9 F1 and IRCoT by 13.7 EM / 15.8 F1. The project page is available at https://hi-q-project.github.io/.

Fine-tuning can improve legal question-answering accuracy without improving how models use law supplied in context. We study this distinction in bilingual Bangladeshi legal QA, where observed errors can arise from answer scoring, retrieval, or failure to use relevant law. We construct a hierarchy-preserving statutory corpus, 2,165 reviewed bilingual fine-tuning examples, and a 150-item supplied-law control. We evaluate six instruction-tuned models: Llama-3.2-1B, Llama-3.2-3B, Qwen3.5-0.8B, Qwen3.5-2B, Qwen3.5-4B, and Gemma-4-E2B, with three LoRA seeds per model. To separate effects, we combine constrained option-letter scoring, cyclic option rotation, and controlled removal of the governing provision. On 398 Bar Council outputs, an exact-line parser attributes an accuracy gain of 50.0% to the Qwen3.5-2B seed-42 adapter, whereas option scoring yields only $3.0\%$. For Gemma-4-E2B, the two scoring methods favor different systems. When the governing provision is guaranteed to be present, five of six reference models improve by $14.7\%-19.3\%$ under the four-order criterion. Removing that provision reduces accuracy by $8.0\%-15.3\%$ for models and by $13.8\%-14.9\%$ points for their adapters. However, difference-in differences estimates show no increase in reliance on the governing provision after fine-tuning. Results show that legal adaptation claims require separating scorer, retriever, and model effects. Our Code and data are available at https://anonymous.4open.science/r/bangladesh-legal-qa-11E3

Professional agent tasks often depend on conventions that are absent from public corpora, yet benchmarks rarely control whether an agent has access to those conventions. We introduce a knowledge-gated task-construction protocol that separates a task instruction from a compact artefact containing private conventions, reference tables, and utility operators. Construction-time provenance, byte-identical task instructions across the provided- and withheld-artefact conditions, leak audits, and executable witnesses make dependence on the artefact explicit and testable. Across fifteen calibration tasks, one frontier agent configuration achieves a 68.0% pass rate with the artefact and 0% without it; on one task, a plausible but incorrect artefact also yields 0% across five trials. Deterministic solvers and rule corpora provide exact ground truth for structured tasks, while named criterion-level rubrics support outputs that cannot be checked by a single executable oracle. A configuration-relative calibration screen retains seven tasks satisfying our five-trial empirical knowledge-gating screen. These experiments validate the behavior of the construction protocol; they do not establish that the retained tasks improve post-training. We publicly release part of the task suite and supporting tooling at https://github.com/DatagridsAI/Knowledge-Gated-Task-Construction.

Generative engines such as ChatGPT, Gemini, and Perplexity answer buyer questions directly and name a shortlist of brands inside the answer. Studying how brands enter or fail to enter that shortlist requires demand-side data: what buyers in a category ask, what information they need, and which sources they trust. Existing large persona corpora are built for training-data diversity and carry neither a staged search-intent label nor a preferred-sources field, so they cannot be joined to supply-side recommendation measurements. We built and validated PersonaGen-1M, a corpus of 1,031,732 synthetic buyer personas spanning 511 industry labels and 4 market contexts, carrying 19,416,821 structured behavioral attributes, 5,160,046 of them search queries. Each persona carries a single primary_intent label covering its query set (78.3% informational, 17.4% commercial, 4.3% transactional) and a preferred_sources field naming the source types that buyer would trust. The corpus was built from roughly 40 million raw persona descriptions drawn from four public datasets through GPU-accelerated MinHash LSH plus semantic deduplication, then enriched to a fixed schema. The intent field selects the commercial-evaluation personas whose queries drive recommendation, and the preferred_sources field pairs against citation-provenance data; that join is the primary intended use, and its controlled empirical estimate is future work. Among million-scale persona corpora surveyed in August 2026, one other carries a source-preference attribute, as a six-value media-channel enum; PersonaGen-1M pairs named per-persona source lists with a staged commercial search-intent label and an attached query set. The full corpus is shared on request for non-commercial research; a stratified subset is published openly so the protocol, the schema and the validation can be inspected and reused without asking us.

Flat retrieval-augmented generation treats a corpus as a bag of chunks, discarding document hierarchy and cross document structure. We introduce SearchWiki, a harness framework that synthesizes a corpus into a hierarchical, typed, navigable wiki and trains an agent, WikiResearcher-9B, to retrieve information through multi-turn tool use. The wiki organizes knowledge into three layers - document overviews, cross- document topic pages, and page-level source records; enabling progressive refinement of retrieval when initial lookup misses. We optimize the agent's navigation policy with on-policy reinforcement learning with a multi-component reward function balancing answer correctness, retrieval quality and trajectory efficiency. Evaluation on ViDoRe-V3 (8 domains), FinanceBench, and memory benchmarks (LoCoMo, LongMemEval, PersonaMem-v2) shows that WikiResearcher- 9B which is our RL-tuned Qwen 9B model, significantly outperforms same-size untrained baselines and exceeds or matches larger external models. SearchWiki paired with WikiResearcher-9B demonstrates that learned navigation over structured corpora is a superior alternative to flat retrieval.

Biomedical Named Entity Recognition (NER) is fundamental to healthcare AI applications, including clinical decision support and medical information extraction. While corpora with Unified Medical Language System (UMLS) annotations, such as MedMentions, have driven progress in English biomedical NER, no comparable resource exists for Vietnamese. This paper presents En-ViMedNER, the first English-Vietnamese parallel biomedical NER corpus annotated with UMLS semantic types, which are language-neutral codes providing a shared cross-lingual label space and ensuring direct comparability with existing UMLS-based resources. The corpus contains 4,392 PubMed abstract pairs, 44,892 English-Vietnamese sentence pairs, and 202,949 aligned entity-mention pairs across 21 semantic types adapted from the MedMentions ST21pv dataset. To balance quality and scalability, we have constructed the corpus through automatic translation, expert post-editing, LLM-assisted label projection, and human verification and adjudication. We characterize En-ViMedNER as a large-scale silver-standard corpus with a human-audited and consensus-corrected mini-test subset. We evaluate En-ViMedNER in two settings: (i) Vietnamese-input/Vietnamese-output biomedical NER and (ii) English-input/Vietnamese-output cross-lingual NER. For Vietnamese NER, we benchmark Vietnamese-supervised encoder models, English-supervised multilingual encoder models, and prompt-based LLMs. The best model achieves an F1 score of 52.70 on the test set and 53.78 on the mini-test set. For cross-lingual NER, we benchmark encoder-decoder models and prompt-based LLMs. The best model achieves an F1 score of 45.44 on the mini-test set. We publicly release our corpus, corpus construction pipeline, and baseline models to facilitate future Vietnamese biomedical NLP research.

A recent line of synthetic-data work reconstructs the thinking behind existing text rather than rewriting the text itself, but it operates on short web passages, recovers only local thoughts, and leaves the structure of whole documents untouched. Scientific papers are written to a clear and largely uniform structure and make a natural substrate for lifting this paradigm to the document level. We present a pipeline that unfolds each paper into a multi-turn generation trajectory in which a teacher model reconstructs the writing process of the whole paper: a writing request, a global plan, and pre-writing deliberation for each section. All section texts and the abstract are kept verbatim from the source paper. We apply the pipeline to 1.8M quality-filtered arXiv papers and obtain a 60B-token corpus for continued pre-training (CPT) that is roughly twice the source text. The same reverse construction extends to instruction data and evaluation. We build an SFT dataset of 200K samples using answers derived from paper text. We also use held-out papers to construct PAW-Bench, a benchmark of 2,940 academic writing tasks with per-task rubrics and checklists. In controlled experiments, CPT on our corpus followed by SFT on public datasets improves writing performance while preserving general reasoning and improving long document reading. Replacing part of the writing SFT data with our synthetic instruction data further improves performance on PAW-Bench.

Chinese query correction (CQC) is important for search and query recommendation on content platforms, but supervised methods rely on large annotated correction pairs that are costly to maintain as query vocabularies evolve. Unsupervised correction with language models is attractive, yet in the short-query setting, unconstrained generation often over-corrects ambiguous inputs toward high-frequency phrases, causing intent drift. We propose \textsc{GUIDE}, a generative unsupervised framework for CQC based on a confuse-then-clarify paradigm. \textsc{GUIDE} encodes phonetically or visually confusable characters with shared-IDs and reconstructs the original query with an encoder--decoder architecture, which constrains correction to plausible confusion neighborhoods while learning from unlabeled query streams. A time-decayed, query-frequency-weighted objective further supports adaptation to rapidly changing query vocabularies. Experiments on \textit{QSpell 250K} and a large-scale real-world dataset (\textit{KwaiSearch}) show that \textsc{GUIDE} consistently outperforms strong baselines, while online A/B testing further confirms gains in correction quality and downstream engagement.

Spec-Driven Development (SDD) is a fast-emerging practice in which a structured natural-language specification, written by a developer, or (more often) drafted by an AI tool and then curated by the developer, drives an AI coding agent's implementation. A wave of tooling (GitHub Spec Kit [3], OpenSpec [4], AWS Kiro [5], and dozens of others) has appeared since 2025, yet the artifacts these tools produce have never been studied at scale. We present SpecMine, a corpus that captures SDD in public GitHub repositories through two censuses: a broad census of spec.md/specs.md files covering most tools (470,795 files across 73,030 repositories, attributed to 17 named tools), and a Kiro census of its distinct requirements/design/tasks layout (98,574 files across 12,910 repositories). Each spec is enriched with full repository metadata, complete commit history, and parsed document structure. How a spec becomes code is itself an open question, so for 11 tools we sweep every pull request that touches a spec in their repositories with at least ten stars, capturing 5,992 such PRs across 581 repositories with their changesets. That makes the simplest workflow, spec and implementation changing together in one PR, directly observable, and a census-wide index of 2,421,323 typed references (1.28M to code files, 863k to sibling documents, 152k to PRs, 62k refs, 43k branches, 22k issues) gives a second, independent link from spec to code. SpecMine lets the community study, for the first time, how software is specified in the age of AI agents.

A 4-bit quantized weight specifies a rounding cell rather than a single full-precision value. We introduce in-cell learning, a paradigm for writing new knowledge only within these cells, so that re-quantizing the served weights reproduces the released integer codes and scales exactly. CellFill implements this idea with bounded trainable positions inside frozen quantization cells and ships the update as a separate, subtractively revocable file. Across published NF4 and W4A16 releases of Qwen3 and Gemma from 1.7B to 32B parameters, CellFill writes 83-99% of a real-fact corpus while returning the stored code on every constrained weight. The injected facts generalize to paraphrases and composition, and answer 78-88% of selected PopQA questions that the released model misses. Sequential experiments show that rehearsal preserves earlier knowledge, whereas available room and new-task plasticity decline across updates. Consolidation re-quantizes the learned weights to produce a declared major version, restoring room at a measured capability cost. A six-task write-rehearse-consolidate cycle retains at least 92.8% of first learning in two 8B runs and records zero code violations over 6.9 billion constrained weights at every fold. These results define a version-management protocol in which minor updates preserve the released quantized artifact bitwise and major updates are explicit, measurable, and verifiable.

PDF corpora advertise their size in tokens, but every rate they publish (coverage, OCR routing, re-fetch recovery, language mix) is computed per document, and none decomposes its token total. Because PDF length is extremely skewed, the two units can describe the same corpus very differently. We ask how the headline statistics of a web-PDF corpus change when each document is weighted by the text it contributes rather than counted once. We used CC-MAIN-2021-31-PDF-UNTRUNCATED (7.9M Common Crawl PDFs, 32.6B tokens), the one public corpus that pairs the fragments Common Crawl stored with the re-fetched originals. Text mass is highly concentrated: 3.02% of text-bearing documents hold half the tokens (Gini 0.807). The clearest consequence is Common Crawl's payload cap, which truncated 23.06% of these documents but 63.08% of their text. Reconstructing the truncated fragments and extracting both versions, two widely used text-layer parsers recover only 1.4% and 11.4% of that exposed text, so roughly 55-62% of the corpus's text is unrecoverable from the crawl by such pipelines; under the 5MiB cap adopted in March 2025, 30.19% of tokens would still be exposed. We recommend that corpus statistics be reported in both units, documents and tokens.

Self-supervised learning (SSL) speech representations achieve strong performance for Parkinson's disease (PD) detection within individual corpora. However, it remains unclear whether these models capture disease-related characteristics or exploit dataset-specific confounds, particularly since most SSL backbones are pretrained exclusively on healthy speech. To investigate this question, we perform a layer-wise analysis of nine SSL speech backbones using a low-capacity logistic regression probe across three languages. We structure the evaluation as multiple scenarios that progressively introduce distribution shifts in participant identity, recording conditions, language, and pathology. Our results reveal two key findings. First, layer selection is highly corpus-dependent: the optimal representation layer is determined primarily by the source dataset rather than by the SSL architecture itself. Second, the transferred discriminative signal lacks pathological specificity: classifiers trained to detect PD assign similarly high probabilities to both PD and dementia speech in the target corpus. These results highlight critical limitations that must be addressed before speech-based pathology recognition models can be reliably deployed in clinical settings.

Parliamentary proceedings are a primary record of democratic deliberation, yet their volume and fragmentation make multi-perspective access difficult for citizens, journalists, and researchers. Applying Retrieval-Augmented Generation (RAG) to parliamentary transcripts introduces three specific risks: dominance of the most frequent speakers, inability to weight speakers according to topical expertise, and citation misattribution in politically sensitive text. We present ParliamentRAG, a RAG system for the Italian Chamber of Deputies that addresses these risks jointly. Its core contribution is a topic-dependent authority model that estimates each speaker's authority as a function of the current query, combining interpretable components such as profession, education, and previous interventions. Given a user query, the system retrieves relevant speech chunks, identifies topic-relevant experts across parliamentary groups, and generates a summary synthesizing their perspectives, accompanied by supporting quotations. ParliamentRAG is evaluated against Google NotebookLM on 15 policy topics via a two-level protocol combining automated metrics and blind A/B human evaluation by six domain experts. The system achieves higher coverage across political groups (0.97 vs. 0.95), perfect quotation faithfulness (1.00 vs. 0.95), and stronger expert preferences on source-related dimensions, while NotebookLM remains stronger on prose-oriented dimensions.

Time series foundation models (TSFMs) have advanced primarily through architectural innovation, while training regimes for large-scale heterogeneous corpora remain under-explored. As a result, pre-training distributions are often poorly controlled with respect to domain imbalance, context requirements, prediction horizons, and missingness. We introduce ORBIT (Omni-Range Bootstrap Incremental Training), a training paradigm that makes this distribution explicit and controllable. ORBIT combines Bootstrap Multi-Level Sampling, which controls dataset exposure and samples records, target variables, context windows, and prediction horizons, with Omni-Range Incremental Training, which varies context lengths and prediction horizons throughout a single training stage. Under ORBIT, we train Falcon-2.0, a simple univariate encoder-only Transformer with missingness-aware triple-channel patch tokenization and parallel patch prediction. We further introduce Rank-Guided Cross-Depth Alignment, a training objective that uses late-layer representations as stop-gradient teachers for shallow layers without additional inference cost. Evaluations on GIFT-Eval and fev-bench demonstrate strong zero-shot forecasting performance across diverse domains and frequencies.

Retrieval-augmented generation uses an external corpus as inference-time evidence, allowing an attacker to promote a false answer by injecting a handful of documents. Detection must distinguish this manipulation from ordinary relevance without knowing which queries or documents are targeted. Existing detectors use text irregularity, candidate consensus, or corpus-level graph structure, whose reliability varies with the attack and local context. We present RAGSieve, which constructs a reference matched to each detection scope. At query time, RAGSieve-Query (RSQ) compares generation candidates with the lower-ranked tail of the same retrieval, exposing answer-token concentration and carrier-payload seams. At corpus time, RAGSieve-Graph (RSG) compares each document's strongest semantic relations with its own neighborhood floor to measure coordinated density. Neither requires poison labels, a trusted corpus, or training. Across three QA datasets, three dense retrievers, and six poisoning constructions, RSQ reaches 95.2% AUROC and detects 82.2% of poison at a 5% clean-removal budget, against 81.1% and 52.5% for the strongest query-time baseline; RSG reaches 93.3% and 79.8% against 79.4% and 37.6% for the strongest corpus-time baseline, with a 79.6% versus 1.4% detection rate on camouflaged injections. Joint deployment cuts attack success from 67.4% to 16.1% while retaining unpoisoned-retrieval F1 at 41.0%, compared with 42.1% without filtering. Source code is available at https://github.com/XrazyMee/RAGSieve.

While large language models can generate entire novels, there is little information about the level of formal variation in their output over many generations. Rather than asking whether individual passages can be identified as AI-generated, this study asks whether repeated AI generation can produce the same range of diversity which is found across human corpora. This paper contrasts six corpora based on generation source and target style: twenty novels generated using GPT-5.5 Thinking in a nineteenth-century British realist style, twenty novels generated using Qwen3-14B in a nineteenth-century British realist style, twenty novels generated using each of these models in a contemporary zero style, 205 nineteenth-century human-written British novels, and sixty-five contemporary human-written Zero-Style novels. At the document level, the research includes MATTR-500, Shannon entropy, average sentence length, readability, and punctuation rate measurements. The most robust and reliable result is compression of sentence structure. Repeated generations produce novels that vary far less from one another in sentence structure than human novels do. Compression is also present in the measures of readability, punctuation, and sentence length variability within novels. Lexical measures tend to be similarly compressed, with the exception of Qwen Zero-Style MATTR. Despite having distinct mean stylistic profiles, GPT and Qwen lack a stable pattern of cross-measure correlation. This article therefore distinguishes between variance overclosure, which represents a limited formal range between novels, and a more specific phenomenon of correlational overclosure. This means that an individual AI-generated novel may resemble human fiction stylistically, while a collection of AI-generated novels occupies a much narrower formal range.

In 2023, a New York judge sanctioned two attorneys in Mata v. Avianca for filing a brief with hallucinated citations generated by ChatGPT. Such failures are largely caught by database lookups; the harder problem is detecting citations that point to real cases but do not support the propositions for which they are offered -- a failure mode that existing evaluations of LLMs for legal use cases largely overlook. In this paper, we study proposition-level citation support verification through controlled perturbations of real legal citations obtained from two legal corpora, either replacing the cited case or changing only the pinpoint page within the same case. We evaluate fourteen model configurations on the resulting examples. Models catch 93-100% of wrong-case corruptions. They catch only 37-61% of wrong-pinpoint corruptions on court opinions and 52-83% on legal briefs. When models fail to catch wrong-pinpoint corruptions, they accept the citation based on topical overlap rather than page-level support. Scale and extended reasoning narrow the gap but do not close it: GPT-5.4 with high reasoning effort still misses 40% of pinpoint mismatches on court opinions and 18% on briefs. Prompting the model to verify support at the cited page improves recall, but it also raises the false positive rate. Recognizing the right legal topic and verifying support for the cited proposition are distinct capabilities, and current models conflate them.

Neural Machine Translation (NMT) and Large Language Models (LLMs) excel at cross-lingual tasks but often fail to capture intra-lingual morphological variation, marginalizing dialectal speakers. In Bangla, existing translation frameworks commonly rely on Standard Colloquial Bangla (SCB) as an intermediate pivot, which can compound errors and reduce cross-dialectal nuance. To address this gap, we introduce a unified, multi-directional NMT system for direct translation between SCB and eleven regional variants. We first review prior dialectal NLP resources to identify existing technological gaps. As a foundational contribution, we construct and release a large multi-dialect parallel corpus for Bangla, comprising 14,562 aligned rows and 51,541 non-null sentence pairs through the integration of seven prior datasets and native-speaker-verified manual augmentation. Using this corpus, we benchmark state-of-the-art sequence-to-sequence architectures with parameter-efficient Weight-Decomposed Low-Rank Adaptation (DoRA). Results show that deep monolingual pre-training is more effective than large multilingual capacity for this task. The compact BanglaT5 model outperforms NLLB-200 and mBART-50 by up to 13.96 BLEU, achieving 29.26 BLEU, 57.26 chrF++, and 49.68 METEOR. A dataset scaling study shows diminishing returns beyond 3,000 parallel pairs and indicates that linguistic proximity to Standard Bangla is more important than raw data volume for translation quality. Finally, we deploy the optimized model as an INT8-quantized web application, providing a scalable, open-source framework for inclusive language technology and equitable digital access.

Video-to-music generation has drawn growing interest for its role in conveying the emotion of visual media, including film. Progress in the field, however, is hampered by a reproducibility gap: models are often trained on crawled corpora referenced through YouTube URLs that may be deleted, with the underlying data often difficult and time-consuming to retrieve. To address this, we introduce the Open Screen Soundtrack Library version 2 (OSSL-v2), a self-hosted corpus of 34,343 video clips totaling 246.4 hours, sourced from public-domain films. Unlike crawled corpora, OSSL-v2 is reproducible (i.e., not subject to link rot) and copyright-conscious, yet still large enough to train functional video-to-music models. We then use this film-domain corpus to study dialogue as a conditioning signal for video-to-music generation, motivated by the close temporal coupling between film music and on-screen speech. Specifically, we augment existing models' video cross-attention with a time axis and modulate it frame-by-frame with the dialogue track. Evaluated on both public-domain and commercial films, our approach shows improvement over the state-of-the-art baselines. The dataset is available at https://huggingface.co/datasets/McAuley-Lab/OSSL-v2.

IoT firmware vulnerability detection remains challenging due to heterogeneous firmware ecosystems, resource-constrained platforms, and limitations in existing benchmarks. Many datasets are synthetic or general-purpose and lack human-verified, contamination-screened annotations, limiting evidence on cross-corpus generalization across training sources, model architectures, and curriculum strategies. To address this gap, this paper introduces IoTVulBench, a human-verified benchmark for cross-corpus firmware vulnerability detection. IoTVulBench-Core was constructed from GitHub repositories, validated by three expert reviewers, and evaluated on a contamination-screened held-out target across five model architectures, two tuning methods, and three curriculum strategies, with ensemble, distillation, and robustness analyses. Models trained on IoTVulBench achieved the highest MCC among matched single-source datasets, reaching 0.58 versus 0.44 for PrimeVul and 0.39 for D2A. Staged curriculum learning increased MCC to 0.69, while a diversity-optimized ensemble achieved 0.73, improving by 0.42 MCC over the strongest reference comparator, a static analyzer at 0.31, and by 0.29 over PrimeVul. At a 0.5% false-positive rate, the model missed only 21% of vulnerabilities, compared with 71% for the strongest comparator. It retained 86% of its performance under identifier renaming and demonstrated strong calibration and largely faithful explanations. These findings indicate that domain-matched training data and curriculum design, rather than model scale alone, are key drivers of generalization in firmware vulnerability detection. The results provide a benchmark for future research and deployment-ready configurations for practical IoT security applications.

Large language models are being proposed as agents in scientific workflows, in domains where no downstream verifier exists. Such deployment assumes the model can distinguish reliable scientific literature from unreliable literature, a capability that has not yet been directly measured. Existing benchmarks evaluate factuality on questions with known answers; the failure mode we target here is different. We introduce a probe corpus of 42 retracted, fraudulent, and pseudoscientific papers, paired with a methodology for eliciting and scoring single-shot model engagement with each paper's framing. Each probe pairs a preamble extracted near-verbatim from the target paper with a scientifically plausible study-design request. The probes span five claim types: fabricated observation, pseudophysical mechanism, magical premise, legitimization bridge, and cargo-cult experiment. Two complementary scores measure whether a model rejects the flawed premise outright (IFR-a) and whether it recognizes the unreliability while still engaging (IFR-i). A depth score, the Engagement Depth Index (EDI), quantifies reproduction of paper- or field-specific withheld details. Across 30 models and 10 repeated runs, aggregate IFR-a is 0.93 $\pm$ 0.004 and aggregate IFR-i is 0.809 $\pm$ 0.009. Models engaged with untenable premises in 95% of all non-empty responses. Every evaluated model fails more than 71% of agentic probes, and 22 of 30 models fail more than 90% of the time. Rejections are concentrated on a small number of high-notoriety topics and specific probes, and disappear under matched-structure controls. These results are consistent with topic-keyed safety behavior rather than robust epistemic competence, and indicate an urgent need for guardrail infrastructure for scientific deployment of language models.

Sign-language research for resource-constrained languages is often limited by the cost of dense linguistic labels such as glosses, temporal boundaries, and sign order. Broadcast news offers a practical alternative by pairing continuous signing with spoken-language transcripts, but this supervision is weak since text and signing are loosely aligned. Morphologically rich languages such as Turkish add further difficulty, as the same lexical meaning can appear in many inflected forms while some derived forms should remain distinct. We study whether weak transcript-based supervision can pretrain a reusable sign encoder in this setting, where poor text normalization can fragment pseudo-gloss targets and weaken representation learning. Unlike prior pseudo-gloss pipelines designed mainly to improve translation, we test whether the pretrained encoder transfers as a reusable representation for cross-dataset sign spotting. We pretrain on TSL-News, a new Turkish broadcast corpus, using pseudo-gloss labels derived from transcripts rather than manual annotation, comparing rule-based morphological lemmatization with constrained LLM-assisted normalization over a fixed vocabulary. We evaluate the learned representations via cross-dataset sign spotting on a new TSL Spotting Benchmark built from the TSL Dictionary corpus. The LLM-assisted encoder raises top-5 temporal localization mean IoU from 0.235 to 0.465, with 56.2% of examples reaching an IoU of at least 0.50; a frequency analysis suggests this gain is not mainly driven by memorizing frequent pseudo-gloss labels. In a downstream translation check, the same pretraining improves BLEU-4 from 9.60 to 11.04 and ROUGE from 23.48 to 27.43. These results show that loosely aligned broadcast data can provide effective weak supervision for learning sign representations that capture both lexical content and temporal structure.

Multimodal large language models (MLLMs) have demonstrated remarkable capabilities in storing and recalling rich person-related knowledge, raising increasing concerns about reliable knowledge removal. However, existing machine unlearning approaches for MLLMs typically assume access to original forget and retain corpora, which are often unavailable in realistic deletion scenarios. To address this limitation, we introduce PRMU, a benchmark for evaluating corpus-free multimodal unlearning under realistic person-centric deletion requests. PRMU focuses on naturally acquired person-related knowledge and evaluates whether models can remove target knowledge while preserving related knowledge through diverse textual and visual probes, including adversarial evaluation and fine-grained locality analysis. To facilitate research in this setting, we further introduce Similarity-Gated Projection Editing (SGPE), a lightweight corpus-free unlearning baseline with knowledge displacement, protected parameter-space editing, and locality-aware multimodal control. Extensive experiments on representative MLLMs reveal that existing unlearning methods often suffer from unfavorable forgetting-locality trade-offs, with significant locality degradation under aggressive forgetting settings, and remain vulnerable to multimodal knowledge reactivation. Meanwhile, SGPE provides a competitive trade-off between target forgetting, locality preservation, and general multimodal utility. We hope PRMU can facilitate future research toward realistic and scalable multimodal machine unlearning. Code and dataset will be released at https://github.com/2231122/PRMU.

Computational analysis of music often relies on discrete representations, yet many musical traditions are organized around continuous pitch movement that resists segmentation into note-like units. For such traditions, the discrete units that analysis would build on are not given in advance. We address this gap by learning a vocabulary of local pitch-contour patterns directly from unlabeled audio, using a VQ-VAE that quantizes fixed-length contour segments into a finite codebook. To make the learned tokens stable across segmentation positions and small variations in timing and pitch range, we train the model with a reconstruction objective evaluated under the best alignment among a set of candidate temporal and pitch-domain transformations. Applied to Korean traditional music, the learned tokens recover information about expert-defined sigimsae categories without supervision, and in pansori individual tokens align with the two principal modes, Gyemyeonjo and Ujo, supporting their use as units for corpus-level analysis of contour-centric traditions.

Over the past several years, LLM-powered chatbots and agents have become widely used as a tool for academic writing. LLM-assisted writing can be valuable by removing language barriers but at the same time causes concerns about misconduct and fraud. To inform policy decisions, it is necessary to monitor the prevalence of LLM-altered texts in scholarly publications. Despite some recent progress in this direction, no existing method can produce reliable estimates. Here we suggest and validate a new unbiased approach to estimate LLM usage in a corpus of texts based on changing word frequencies. We apply our method to the full texts of open-access biomedical papers from Pubmed Central, and show that by the end of 2025, 89% of papers show excess of LLM-associated vocabulary. We also find that LLMs are twice as likely to be used when writing a paragraph in the Discussion section (68%) compared to a paragraph in the Methods section (32%), but even inside the Methods section, the overall prevalence of LLM usage is over 50%. We believe that our estimates are crucial to shape future guidelines and policies.

Pretraining corpus composition shapes LLM capabilities, but it often remains hidden even when model weights are released. Prior work has inferred corpus mixtures or traced specific token groups from released tokenizer vocabularies; in contrast, we estimate corpus ratios for arbitrary target tokens. We first show that BPE tokenizers trained on different corpora share stable token ID--ratio distributions, motivating distribution transfer from known corpora to a target tokenizer trained on hidden corpora. We then propose Quantile-Guided Density Estimation (QGDE), which approximates this distribution with multiple quantile trends and uses local density weighting to produce token-level estimates. In controlled settings and a realistic setting using the released SmolLM tokenizer, QGDE achieves mean relative errors as low as 3.00% for token-level estimation and 3.08% after aggregation into category-level mixtures. These results suggest that released tokenizer vocabularies provide a useful signal for fine-grained corpus estimation beyond coarse composition inference.

Purpose: Keyphrases are statistically and semantically important textual units that can also attract readers' attention during comprehension. However, existing keyphrase extraction (KPE) studies mainly focus on improving textual representation while largely overlooking human reading behavior. This study examines whether lightweight webcam-based eye-tracking features can improve KPE from Chinese academic abstracts in Library and Information Science (LIS). Methodology: To address the limited availability of eye-tracking data for Chinese academic reading, we developed a lightweight webcam-based data collection platform using the open-source SearchGazer library and constructed the Chinese LIS Eye-Tracking Corpus (CLIS-ET). Three character-level eye-tracking features, first fixation duration (FFD), fixation number (FN), and total fixation duration (TFD), were incorporated into KPE models to evaluate their effects on extraction performance. Findings: Eye-tracking features consistently improved KPE performance. The combination of FN and TFD achieved the best results on the Att-BiLSTM+CRF model, indicating that readers' fixation behavior provides useful signals for identifying keyphrases in academic abstracts. Originality/value: This study introduces a cost-effective webcam-based eye-tracking approach for KPE and presents CLIS-ET, a Chinese academic eye-tracking corpus containing FFD, FN, and TFD features. The results demonstrate the value of incorporating human reading behavior into keyphrase extraction. Dataset and code: https://github.com/yan-xinyi/ET_AKE and https://github.com/yan-xinyi/Reading_ET_System.

Chinese web pollution has surfaced in LLMs, motivating audits of upstream Chinese corpora. However, auditing such corpora faces three challenges: (1) their web-scale size makes full scan costly; (2) prior analyses are often too coarse to expose token-level pollution; (3) Chinese web pollution is implicit and rapidly changing. We propose Sampled-BPE, a lightweight token-level auditing pipeline that sample a small subset and train BPE tokenizer to surface polluted tokens. Experiments show that Sampled-BPE preserves usable estimates while substantially reducing runtime and memory: a 148.4 $\times$ speedup and a 35.8 $\times$ memory reduction induce only 4.25% relative error for pollution categories. We apply the pipeline to 11 open Chinese corpora and 6 Chinese Common Crawl snapshots from 2021 to 2026. The audit reveals widespread but uneven pollution across open corpora, as well as highly polluted and temporally shifting Chinese web content. We further release a hierarchical Chinese web token dataset with 660k+ token records, each with web context, category, and explanation fields, organized as trees to support review and tracing of pollution.

At corpus sizes typical of low-resource dialects, single-run comparisons can yield gains that do not replicate. We show this for Garhwali, an under-resourced Indo-Aryan language of the central Himalaya, building the first reproducible multi-seed ASR benchmark on the official VAANI splits, with per-seed outputs and significance testing. Re-examining plausible gains, we find them fragile: neither Focal CTC nor a matra-weighted objective beats standard CTC under seed-level testing, the matra objective fails to cut even its targeted errors, and Hindi-to-Garhwali transfer gives no gain over direct fine-tuning. What holds up is mundane: w2v-BERT 2.0 with standard CTC reaches 47.0% WER over five seeds, beating the larger MMS-1B and comparable models; pretraining design, not parameter count, drives performance, and speed augmentation gives a small, largely consistent gain. Multi-seed evaluation on official splits separates real gains from seed noise.

Visual document retrieval (VDR) is dominated by multi-billion-parameter models that are slow to index at full corpus scale and expensive to serve. Prior compression routes either train a smaller multi-vector encoder from scratch or distil only the query side; neither yields a compact single-vector retriever end-to-end. We present DistilVDR, a 524M end-to-end VDR system distilled bilaterally from a single 8B vision-language teacher under a pointwise cosine alignment loss. All supervision comes from the frozen teacher's embedding space, which was itself trained with relevance supervision, so the student objective needs no relevance labels, negative sampling, or contrastive term. We match VDR's text-query and image-document input asymmetry with an asymmetric encoder-only student that concentrates visual capacity on the document side and keeps the query side at 70M parameters. We release two variants that share the same encoders and training and differ only in the document encoder's visual-tile budget: DistilVDR-HiRes attains 61.74 average NDCG@5 on ViDoRe v1+v2+v3 (86.9% of the 8B teacher) and leads every reproduced sub-1B baseline on the high-resolution-sensitive v3 benchmark, while DistilVDR-Fast attains 59.98 at a 3 times smaller visual-token budget. Both variants store one million documents in a 15.6 times smaller index than the strongest sub-1B multi-vector baseline and index the corpus an order of magnitude faster. The code is available at https://github.com/Ryenhails/NanoVDR.

Large language models (LLMs) are trained on massive and largely undisclosed corpora that may contain copyrighted or privacy-sensitive content. Data contamination detection (DCD) therefore aims to determine whether a given text is a member of the pre-training corpus of a target LLM. Recent state-of-the-art DCD methods follow a feature-based paradigm that derives membership features from the input text and the corresponding model output. However, most modern LLMs undergo post-training, such as instruction tuning, preference optimization, and reasoning-oriented training, which can alter model outputs and shift the corresponding membership features, thereby reducing the separability between members and non-members. To address this problem, we propose CalibDCD, a broadly applicable calibration framework for feature-based DCD methods, comprising (1) Multi-View Shift Detection, which identifies recurring feature shifts associated with post-training, and (2) Bounded Feature Correction, which selectively mitigates their influence on membership prediction. Specifically, Multi-View Shift Detection evaluates controlled prompt variants on known non-member texts and consolidates the most informative views to identify recurring feature shifts. Bounded Feature Correction selectively adjusts feature components aligned with the detected shifts and controls the correction extent to preserve useful detection information. Experiments show that CalibDCD consistently improves existing feature-based detectors, with gains of up to 7.0% in AUC and 15.0% in TPR@5%FPR.

As information increasingly traverses linguistic boundaries, users require concise cross-lingual representations of long-form content. Nevertheless, long-document summarization research remains text-centric, whereas multilingual speech research has largely prioritized translation, preserving source content rather than compressing it. We address this methodological gap by formalizing joint speech summarization and translation (JSumT): the generation of a succinct, faithful target-language summary directly from a long spoken document in a source language. We additionally introduce VoxSumm, the first multilingual and cross-lingual benchmark for this task, comprising 10,045 BBC article-summary pairs across 24 languages and encompassing approximately 703 hours of speech data. Our evaluation of representative speech-language models reveals pronounced variation across models and generation settings: Gemini3.1-Pro demonstrates the greatest consistency, summarization into English generally surpasses generation into non-English target languages, and translating an entire document before summarization compounds instruction-following failures. Through the release of VoxSumm, we establish a foundation for developing and evaluating multilingual systems capable of jointly interpreting, compressing, and translating long-form speech.

Agent frameworks ship quality gates that compare text blocks by embedding-cosine similarity and decide at a fixed cutoff. Deduplication filters, semantic caches, drift guards, and answer grader gates deploy to answer the question: "Does this text still mean the same thing?" But the score answers a different question: "How much did the wording change?" We audit this gate class as a measurement instrument. In the cases these gates exist to catch, the two can run in opposite ways. Many times, reversing an instruction is a single word edit, while agreement often rephrases a sentence. The consequence is a safety check that fires backwards. The production drift guard we audited caught 0 of 56 meaning-breaking mutations, and one approved item, "withhold the study drug" -> "administer the study drug", came in at cosine 0.9608. We observed five shipped operating points, and balanced accuracy across 90 configuration-threshold-task cells never exceeded 0.700 (median 0.525). The same confounder also corrupted evaluations. A naively built corpus inherits this confounder and can return an inverted verdict, with a decision AUROC exactly 0.000 in 13 of 18 configuration-task cells (at most 0.040 in all 18) against 0.440-0.815 for the same nine configurations under a balanced 2x2 design. Twice in the effort it captured our own headline claims. Obvious repairs fail: an encoder swap and an overlap-conditioned gate (0.750 in-sample, 0.533 held-out) land at chance on separately authored held-out data, and an NLI drop-in did no better. Embeddings do still bear hope here, as the strongest two of nine configurations separated reversal from paraphrase at matched overlap (AUROC 0.79-0.90), but only a matched-pair audit reveals the deployment regime. We release the corpus method, harness, and frozen results, and contend that scores gated this way measure the wrong thing. We believe a valid instrument is buildable.

Social media has become a major venue for multilingual communication, where users frequently mix multiple languages within a single utterance. Although code-mixed corpora have been developed for several language pairs, Persian-English code-mixing remains relatively underexplored. Existing Persian resources lack Universal Dependencies (UD) part-of-speech (POS) annotations for code-mixed words, limiting both linguistic analyses and the development of syntax-aware NLP models. To address this gap, we introduce PERCEPT, the first publicly available large-scale Persian-English code-mixed corpus annotated with Universal Dependencies POS tags for code-mixed words. The dataset comprises 6,800 posts collected from X, Instagram, and Digikala. We further present an LLM-assisted annotation framework that automatically assigns POS tags and document-level topics. Human evaluation demonstrates high agreement between the automatically generated annotations and gold annotations, confirming the reliability of the annotations. Using PERCEPT, we conduct the first comprehensive linguistic analysis of Persian-English code-mixing across multiple social media platforms. Our analyses reveal that nouns are the predominant category for code-mixed words, while the distributions of other POS categories vary across platforms. We further find that the positional distribution of code-mixed words is remarkably consistent across platforms, whereas the triggering effect is substantially more pronounced in Digikala. PERCEPT is publicly available at https://github.com/kalhorghazal/PERCEPT.

We identify and quantify temporal misgrounding: the systematic retrieval and citation of the currently in-force version of a legal article when the applicable version is an earlier or future one. Standard legal RAG treats the corpus as static; we argue legal question answering is a temporally-indexed retrieval problem. We introduce FiscalQA Pro, pairing a versioned corpus of 32,436 article-versions of the French tax code (93 years, 1938-2031) with an all-model-hard temporal-reasoning track: 209 scored, expert-reviewed questions across 33 CGI articles (221 released; twelve flagged out of the answerable scope). At selection time, no evaluated model recovered its date-applicable answer closed-book in any of four sampling draws, and the currently in-force text lacks the gold value for all but one of the scored questions. Answers are scored deterministically via atomic ground-truth "nuggets" (regex and numeric-with-tolerance), never LLM-as-judge: an LLM judge would inherit the temporal bias it is meant to score. Across eleven models (five frontier closed-API systems plus Gemini 2.5 Pro as a substitute entry, and five open-weight), parametric knowledge yields 3.0% mean strict accuracy and RAG over a static current-version corpus 2.7%. Static RAG retrieves the date-applicable version 0% of the time, confidently citing a real but inapplicable version. Our end-to-end retriever over a multi-version index, with no oracle, reaches 98.3% mean strict; an oracle-article ablation reaches 99.1%, locating the residual gap in first-stage recall, not version selection. We additionally release a version-aware jurisprudence dataset of 69,208 citation links, together with the corpus, benchmark, model responses, and pipeline code.

In this project, we delved into the pervasive challenge of bias detection within the text content. More specifically, our focus lies on the identification of subjective bias, a type of bias that introduces improper attitudes or portrays a statement at odds with the actual truth. The subjective bias can jeopardize the authenticity and reliability of texts, leading to misconceptions and potential social tensions, especially when expressed through offensive language. Following prior work [1], we tackled with three different types of subjective biases in text: (1) framing bias with the use of one-sided words or phrases containing a particular point of view; (2) epistemological bias which includes subtle linguistic features that can affect the believability of the texts; (3) demographic bias with word/phrase usage under presuppositions of a particular demographic factor (i.e., gender or religion). In terms of the data we utilize, the input consists of texts that may harbor subjective biases. The output is a classification or annotation that reveals the presence or absence of such biases within the provided content. More specifically, we detected three different types of multi-span biases in corpus WIKIBIAS [2] with more than 4,000 sentence pairs from Wikipedia edits. The data is labelled by bias type for span pairs with the following categories: (1) framing bias, (2) epistemological bias, (3) demographic bias, and (4) no bias. The project codes are released at https://github.com/HoningJade/LLM-Bias-Type-Classification.

How a document's arrangement is written down, its notation, is a training variable that no dataset card records. The field has established that text-extraction choices change model behaviour, and has never once measured the notation of what those choices put into the corpus. We define clean-window survival, a deterministic count of how much of a stream still demands the boundary inference, and measure notation on three fronts. What corpora carry: a census of thirteen public corpora, where survival falls to 0.153 in a vision-converted PDF slice against 0.889 in C4; the scarce resource is not unmarked text but long unmarked text; a pre-registered supply test finds what remains institutional, not consumer. Our own pre-registered prediction failed: converters do not fabricate structure on prose, and that null forced the reliability mechanism that survives it. What readers use: across five base models spanning 0.6B to 8.2B and two pipelines, deleting a structural announcement makes the following prose measurably harder to predict, while swapping its notation moves nothing. That zero does not make notation unimportant; it relocates the variable: the operative cue is the announcement, not the sigil. What writers impose: a bounded null. Base models do not impose the marked register above the authored baseline, and handed prose with every announcement deleted they do not put one back, at a rate indistinguishable from zero against an authored reference of zero. We ship the format those measurements imply: the pure frame, paragraphs in authored order, every announcement deleted into a reversible sidecar, mixed against the marked copy over announcement presence rather than notation. Choose format operators by the capability they train, not by the fidelity they preserve, and record extractor identity and survival on data cards.

Public infant cry corpora are small, label-incompatible, and almost always evaluated one corpus at a time. We ask what this practice hides and what fixes it. Across four cry corpora screened by a multi-level leakage audit (byte-level and embedding-level deduplication plus a within-corpus train-test near-duplicate audit), we probe four frozen encoders and a handcrafted baseline under a unified five-class need ontology and shared task formulations. The audit exposes what single-corpus evaluation conceals: within-domain macro-F1 swings by 0.57-0.80 for the same encoder, cross-corpus transfer is negative on average (negative-transfer ratio 0.19-0.35, significant in 18 of 30 directed cells, BH-FDR), and 349 content-identical clip groups carry conflicting metadata labels across corpus distributions. The same audit, however, reveals a consistent way forward. Transfer into the noisiest corpus is consistently positive in effect size at matched training size and after near-duplicate removal, offering a practical recipe for small, noisy corpora. Frozen probes saturate at modest label budgets, while stabilized fine-tuning wins with full labels; domain-adaptive pretraining significantly beats stabilized fine-tuning at 5-10-shot (the 1-shot advantage is not robust to optimization-seed variance) but shows no significant advantage at 50-shot or beyond. In the tested binary, shared-label settings, ontology-mapped joint training wins in all four encoder-by-target combinations, whereas naively merging unmapped labels costs up to 37 F1 points. We release the ontology, mapping code, and audit pipeline, turning incompatible cry corpora into a usable joint-training resource.

Agent Skills---structured packages of instructions and scripts that augment LLM-based agents---are rapidly proliferating, yet their security properties remain under-explored. We present \textsc{SkillsMetric}, a five-stage static analysis framework that scores skill packages along pattern density, statistical anomaly, dataflow taint, import anomaly, and capability mismatch dimensions. We construct an adversarial evaluation dataset of 2{,}266 skills spanning 16~attack types across code-level, system-level, and semantic-level threats, and evaluate on the full SkillMD-138K corpus. Our framework achieves an AUC of 0.93 and 5-fold cross-validated F1 of 73.4%$\pm$0.5%, with strong detection of data exfiltration (93%) and steganographic payloads (93%). Crucially, we identify fundamental blind spots: \emph{host destruction} attacks using common shell commands evade all five stages (0% detection), and \emph{prompt injection} via natural-language manipulation achieves only 42% detection. These findings establish that static analysis alone is insufficient for skill security, motivating defense-in-depth architectures that combine fast static pre-screening with semantic review.

We extend the DS@GT ARC working-note submission to the Touché 2025 Retrieval-Augmented Debate task. The task has two subtasks: generating the next utterance in a simulated debate, and evaluating debate responses according to the Gricean maxims of Quantity, Quality, Relation, and Manner. The DS@GT ARC submission consisted of six leading LLMs from three providers through a retrieval-augmented prompting pipeline. We summarize the results from the working paper and explore whether multi-LLM evaluator agreement is a reliable proxy for official evaluation performance. The analysis shows that frontier LLM systems are strong response generators, and as evaluators they agree strongly within model families. However this consensus does not reliably track the official evaluation target, with the largest gap on the Quality maxim. The accompanying source code for this paper is located at https://github.com/dsgt-arc/touche-2025-rad and https://github.com/dsgt-arc/touche-2025-rad-analysis.

Post-Editing (PE) of Machine Translation (MT) output often involves repeating the same lexical and terminological corrections across many segments, especially in specialised and highly repetitive documents. Despite substantial work on Automatic Post-Editing (APE), most available corpora operate at the sentence level, others are synthetic, and overall not designed to study how corrections propagate in realistic Computer-Assisted Translation (CAT) workflows. This paper presents the APEX-VW (Automatic Post-Editing eXperiments on Virtual Wards) Corpus, a new open English-Spanish (EN-ES) dataset built from recent NHS virtual-ward documents and professional PE in Trados Studio, with controlled MT, terminology, and quality assurance settings. The corpus contains seven document-coherent source texts totalling 42k words, translated with four MT systems representing different paradigms and then post-edited by professional translators. Unlike prior resources such as WMT APE corpora, eSCAPE, MLQE-PE, or LangMark, the dataset preserves document order and CAT-tool context, making it suitable for research on terminology normalisation, correction propagation, and human-in-the-loop translation support. The paper describes the corpus design, data preparation, PE setup, and initial corpus statistics, and positions the resource as a benchmark for document-level APE and propagation-aware assistive tools.

Reasoning modes of language models outperform their non-reasoning counterparts on multi-step agentic tasks, but pay a 3-6x premium in output tokens on every episode -- much of it spent re-deriving procedures that are shared across episodes of the same domain. We show this recurring cost can be amortized: a coding agent analyses a small corpus of existing trajectories from a training split and compiles a compact natural-language skill that is injected into the non-reasoning model's system prompt. Across four agentic benchmarks (ALFWorld, tau$^2$-bench telecom and retail, and SpreadsheetBench-Verified), skills recover 55%-100%+ of the reasoning gap for GPT-5.4-mini on held-out tasks -- exceeding the reasoning mode outright on two of four -- while emitting 2.7-6x fewer output tokens and zero reasoning tokens. Notably, reasoning traces are not a prerequisite: skills distilled from non-reasoning trajectories alone remain competitive with skills distilled from paired reasoning/non-reasoning corpora, with domain-dependent differences between the two sources. We interpret these results through a search lens: test-time reasoning is deep search inside a single episode, re-paid at every deployment, while corpus distillation is wide search across episodes, paid once. The two recover overlapping procedural knowledge, and width over cheap trajectories is often the better buy -- with the residual gap on some domains (telecom, SpreadsheetBench) delineating where genuinely per-instance deep search remains necessary.

We introduce the workbook time machine, a pipeline that automatically creates benchmarks evaluating the ability of language models to create derived objects in spreadsheets (formulas, charts, pivot tables, and conditional formatting). Applied to public workbook corpora, it produces wtmcorpus--a collection of (input workbook, output workbook, query) triples spanning four artifact types and varying complexity. From this corpus we curate wtmbench, a 150-task evaluation benchmark with queries at three levels of specificity. We evaluate existing spreadsheet manipulation agents and baselines on wtmbench across artifact types, step complexity, and instruction granularity. Our evaluations show that query specificity, agent orchestration, and interface API used to control spreadsheets play a big role in LLM performance on Excel tasks.

The dominant approach to mechanistic interpretability trains proxy dictionaries such as sparse autoencoders and labels features from max-activating text. The best such atlases identify con- cepts, but that identity lives in the learned dictionary rather than in the network weights them- selves. We propose extracting mechanism mounts directly from linear sites by column-tiled SVD: each mount is a triple (v,u,σ) read as trigger, write, and strength. Identity is the weight rule. We evaluate mounts with a pre-registered suite judged on full-write energy lift rather than tile-local lift. On Gemma-2-2B with WikiText-2 (16,384-token subsample), all seven linear maps are scored: residual writes (mlp.down, attn.o) receive full A/B/C with steer after post-sublayer RMSNorm and pass 52/52 site-layers; other maps receive A/B only (mlp.gate/attn.q/attn.k/effective mlp.up/attn.v 26/26 each). Aggregate: 182/182 GO. We release library code, the corpus builder, the experiment entrypoint, and unit tests.

We describe DAEP, team BIGC's submission to NLPCC 2026 Shared Task 1 Track 3: Difficulty-Aware Temporal Answer Grounding in Video Corpus (DA-TAGVC). The task requires retrieving the target video from 50 candidates and localizing the answer-supporting span. DAEP ranks videos with subtitle, visual, and procedural-context evidence, expands high-scoring anchors into temporal spans, and reranks spans for final output. Its main design is to convert the task-provided simple/complex input label into an inference-time evidence plan controlling modality weights, Top-K aggregation, boundary threshold, expansion length, and reranking strength. In the official evaluation, BIGC ranks first among ten systems with an Average score of 0.2728. Validation ablations show that visual evidence, procedural context, and difficulty-aware planning improve ranking quality, with the largest gain on complex questions.

Conceptual metaphors guide our thinking and actions by allowing us to reason about more abstract experiences (e.g., paying taxes) in terms of more concrete or embodied experiences (e.g., carrying a physical load) (Lakoff and Johnson, 2011). It follows that different conceptual metaphors can result in different reasoning: framing paying taxes as an investment in a community rather than a physical load leads to a very different outlook on taxation. Identifying the conceptual metaphors guiding a speaker or writer thus helps to reveal their framing of events. Though these metaphors can't be observed directly, groups of linguistic metaphors, metaphorical expressions as they appear in language, serve as evidence for them. Motivated by this, we present an unsupervised method that extracts linguistic metaphors from a corpus and uses a structured clustering approach to form groups corresponding to conceptual metaphors. Using this method, we point to key topical and framing differences in left- vs. right-leaning podcasts. For example, left-leaning podcasts tend to conceptualize media stories as a weapon, while right-leaning sources commonly discuss the economy as a system subject to vertical changes.

This paper presents an end-to-end, IIIF-based pipeline that turns the digitised Armenian press of France into an interactive map of the 20th-century Parisian Armenian commercial community. On each page, commercial advertisements are located, read, and parsed into structured records, which are then geocoded and placed on the map. Western Armenian is under-resourced and unsupported by off-the-shelf layout and OCR models, so the pipeline uses vision-language models (VLMs) as a data-bootstrapping strategy: they produce usable structured records at a scale hand annotation could not reach, and stay reliable on the strongly curved scans where conventional line-level CRNN OCR breaks down. The contribution includes a 500-page Western Armenian press corpus with 3,270 advertisement-level annotations, a Label Studio template that captures detection and semantic fields in a single annotation pass, and a reproducible workflow transposable to other under-resourced historical corpora. More broadly, the work shows that VLM-driven data bootstrapping is an effective lever for under-resourced historical languages such as (Western) Armenian.

We present MameLoshnLM, the first open-source 8B-parameter language model built specifically for Yiddish. Despite Yiddish's rich textual tradition, its limited digital presence and the scarcity of reliable evaluation resources have constrained progress in Yiddish language modeling. Existing multilingual corpora and benchmarks are often poor proxies for the language, containing substantial amounts of noisy, machine-translated, and misclassified text. We address these gaps by introducing Oytser, a high-quality Yiddish pretraining corpus that combines contemporary web-native sources with literary materials, and Kashes, a multi-task benchmark spanning translation, linguistic analysis, information extraction, and language understanding. Using these resources, we continue pretraining Llama 3.1 8B to obtain MameLoshnLM. Across the tasks in the benchmark, MameLoshnLM outperforms open baselines of similar scale. Our analyses show that these gains are not only quantitative: relative to general-purpose multilingual models, MameLoshnLM better captures language-defining lexical and morphological patterns, pointing to a broader failure mode of noisy web-scale multilingual data for low-resource languages. Our results provide both a foundation for Yiddish NLP and a practical template for language model development in historically rich but digitally underrepresented languages.

Computer-use agents pay full frontier inference to re-derive routines their user has already performed, because an agent's memory today records what the user said, not what the user did. We compile passively captured screen activity into agent memory with a deterministic, zero-model pipeline: it segments a local capture stream into typed activity frames, bounded episodes carrying application, site, timing, input volume, and evidence pointers back to the raw rows, with no model in the loop, so the output is byte-identical, cacheable, and mechanically auditable. On one professional's single-user corpus of 128,756 frames over 51 active days, the compiler reduces a day of raw capture to a prompt-ready context block 86x smaller in 68 ms, and an agent reading that block answers questions about the day at 98.4% accuracy (Wilson 95% CI 91.7-99.7%) against an independent oracle, versus 66-80% for an LLM summary of the same capture, a mid-tier model reading the block matching a frontier one. The same compiler doubles as a demand-side cost instrument. Read off passive, pre-delegation human activity rather than agent rollouts, it supplies two parameters that agent-cost models assume but, to our knowledge, have not measured: the Routine Overhead Ratio R and the routine recurrence h. We report first values of R, a modeled upper bound, at 60-343x, and a delegable recurrence of 9.0% in-sample and 7.7% out-of-sample, for a realistic all-fleet token ceiling near 8%; a compiled routine replays deterministically with the model out of the loop, demonstrated live at zero model tokens on a guard-matched hit. Schema, compiler, and evaluation harness are open.

Sparse word embedding pipelines can avoid dense co-occurrence matrix materialization, dense factorization, and gradient training while still relying on sparse global corpus statistics. This paper studies Random Indexing (RI) vectors refined by weighted averaging on a sparse Positive Pointwise Mutual Information (PPMI) graph. On a fairytales corpus, the covered semantic analogy set consists of 272 Google family- category questions. On this family subset, PPMI top-K graph averaging repairs a weak RI initialization, improving accuracy from 19.4+-0.7% to 30.7+-2.9% across five seeds. Under the single tested runs, the same neighborhood averaging reduces family- subset analogy accuracy for PPMI+SVD (singular value decom- position), Binary+SVD, CBOW, and Skip-gram. Thus the method is not competitive with neural baselines on text8 and gives near- zero strict similarity correlation on SimLex-999. While Bloom filter sketches underperform RI in the tested configuration, we find that PPMI graph averaging with top-K pruning is a useful non-gradient repair for weak RI embeddings. On the fairytales dataset, PPMI top-K=50 graph averaging improves RI with accuracy going from 19.4+-0.7% to 30.7+-2.9%, and performing best with a seed42 of 34.6%.

Context lengths of language models (LMs) have dramatically increased, driven by the demands for in-context learning, self-improvement, and long-horizon agentic workflows. Existing long-context corpora, however, are dominated by books, academic articles, and code repositories, which are finite resources and often scarce in long-distance dependencies. In this work, we introduce OctoLong, a context engineering pipeline that instruments an AST parser, a language server backend, and a package manager to facilitate the recursive retrieval of code references, enabling the curation of dependency-rich code contexts of millions of tokens in length. We then train OctoLong-Instruct, a suite of capable long-context open LMs, derived from base models ranging in size from 600M to 14B parameters, via context-extension mid-training on a ~50B-token mixture containing ~6.2B tokens of OctoLong code contexts, followed by ~10B tokens of instruction tuning. Our training ablations and experimental evaluations against 18 state-of-the-art open-weight long-context LMs show that supplanting just 12% of traditional context-extension corpora with OctoLong data yields substantial gains in long-range retrieval, long-term state tracking, repository-level code understanding, and downstream agentic tasks, while also enhancing API usage in short-context coding scenarios.

Modern Greek is absent from NVIDIA's Nemotron retrieval models and from major multilingual retrieval benchmarks, despite being important for retrieval-augmented generation (RAG) in legal, energy, financial, and medical applications. We present an end-to-end adaptation of the Nemotron retrieval stack for Modern Greek, including corpus mining, synthetic supervision, retrieval model training, reranker adaptation, reader fine-tuning, and a new benchmark called HERA. Our study shows that a parameter-free BM25 baseline outperforms several off-the-shelf multilingual dense retrieval models on specialist Greek corpora. After fine-tuning on 65,773 Greek retrieval pairs, a Nemotron 1B embedder improves nDCG@10 from 0.362 to 0.835 and substantially outperforms its unadapted counterpart. The learned language competence transfers to general-domain Greek, although the advantage over BM25 remains domain-dependent. We further adapt a cross-encoder reranker and demonstrate consistent improvements across specialist domains. Finally, we LoRA-tune a Nemotron 30B-A3B mixture-of-experts reader for grounded generation, increasing judged answer correctness from 29.4% to 66.9% while significantly improving faithfulness and citation quality. We also introduce HERA, the first large-scale Greek benchmark for retrieval-augmented generation, and release our adapted models and benchmark to support future research on Greek-language RAG systems.

This study examines the grammatical status of preverbal uninflected and underived roots in Mapudungun, with particular focus on wuno 'return/re-'. Through a critical review of scholarly classifications--auxiliaries (Smeets, 2008), modal prefixes (Longkon, 2011), and preverbal particles/complex verb stems (Zúñiga, 2006)--we demonstrate the limitations of existing frameworks. A diachronic corpus analysis spanning four centuries (1606-present) reveals that these elements exhibit three distinct profiles: stable V1 compounds (kim, shinge), volatile V1 rates reflecting orthographic shift (pepi, wuno), and a true particle (kalli). The discovery of V2 attestations for kim and kupa confirms their status as full lexical verbs. We propose a prosodic-orthographic hypothesis: apparent "variable binding" results from the fossilization of prosodic pauses transcribed by early missionaries as spaces, a convention later reanalysed by speakers as syntactic boundaries. The evidence supports Zúñiga's radical concatenation as the correct grammatical model, with implications for the study of languages with no pre-contact written tradition.

Scaling supervision for multi-turn medical agents is difficult because expert dialogue annotation is costly and clinical conversations are privacy-restricted. We introduce Guideline-as-Oracle (GAO), which compiles American Academy of Ophthalmology guidance into a 70-row operational rule table and uses it as the sole source of instance-level supervision for 3,000 training dialogues, reserving human labeling for evaluation. Because converting rules into dialogues is itself a design problem, we catalog eight construction strategies, including cited-row tier assignment, one-fact boundary pairs, metadata-only repair, and label repair, and characterize the evidential status of each: labeling mechanism, null, confounded, or evaluated only as a package. Fine-tuning a 9B backbone on this corpus yields GAO-Triage, improving agreement with a 201-case operational reference from 61.7% to 74.1% (exact McNemar p=0.0046) and emergent-case recall from 9.5% to 69.0%; the gains persist across a second seed and patient simulator. None of the seven general-purpose systems we test dominates GAO-Triage on both metrics, and GAO-Triage requires no frontier model at inference time. Permuting label-dialogue assignments collapses the model to a constant-routine predictor, indicating that the signal lies in guideline-derived assignment rather than dialogue surface form. Label repair coincides with the disappearance of a late-training safety degradation.

Prior work on LLM conformity largely measures discrete answer flips under verifiable labels. Open-ended revisions require a different measurement strategy because answer quality is graded, latent, and judged imperfectly. We introduce an experimental protocol implemented across a pooled main peer-condition corpus and separately constructed decomposition corpora, allowing us to separate ordinary re-answering, candidate-content exposure, a bundled peer-presentation residual, and directional judge sensitivity to visible peer context. Across four open-weight generators and three benchmarks, all-wrong peer input produces the lowest-quality revisions in every generator-dataset cell. Blind and informed ratings of identical answers also differ by evaluator: one judge shifts toward the peer-endorsed position, two shift away, one is approximately neutral, and GPT-4o and GPT-5.4-mini audits are likewise non-neutral. Finally, an anchor audit shows that terse correct anchors can be misread often enough to destabilize the latent scale unless calibration is checked explicitly. These results support four conclusions: flip rates are insufficient as a complete measure of open-ended conformity, wrong peers harm open-ended revision, evaluators are not neutral, and anchor calibration is necessary.

We present MERaLiON-GR, a speech gender recognition system that performs binary classification (female / male) on English and Southeast Asian (SEA) languages. The model finetunes MERaLiON-SpeechEncoder-2, a large conformer based transformer pre-trained on a broad speech corpus, and applies parameter efficient fine-tuning via Low-Rank Adaptation (LoRA) to adapt the encoder to the gender recognition task, and appends a multi-scale ECAPA-TDNN down stream network with attention pooling and a lightweight linear classifier. Extensive evaluations across multilingual Singaporean and Southeast Asian languages (English, Chinese, Malay, Tamil, Thai, Vietnamese, Indonesian, and Khmer) show that MERaLiON-GR consistently surpasses the state-of-the-art gender recognition model Vox-Profile and a large Audio-LLM, in both full-utterance and segment level evaluation modes. The results underscore the value of dedicated speech models in achieving accurate paralinguistic understanding and strong cross-lingual generalization.

Large language models can generate fluent Arabic answers while introducing factual errors that are difficult to identify and verify. Existing Arabic hallucination resources often assign a binary label to an entire response, indicating whether it is hallucinated or non-hallucinated, but provide limited information about the exact erroneous content, the reason for the error, or the correct factual answer. We present HalluTruthQA-4K, an expanded version of the HalluTruthQA resource containing 4,000 expert-curated Arabic question-answering instances across four knowledge-intensive domains: Islamic knowledge, history, science, and geography. Serving as the official dataset for Track 2 of the HalluScoring 2026 shared task, HalluTruthQA-4K extends our original corpus to 4,000 instances. Each instance pairs an Arabic question with a model-generated response, a verified reference answer, and five plausible distractors. Hallucinated responses are additionally annotated with character-level erroneous spans, human-written explanations, and hierarchical hallucination types. The corpus contains 1,643 hallucinated and 2,357 non-hallucinated responses, with 1,843 annotated erroneous spans. We describe the resource construction and annotation methodology, including question selection, controlled answer generation, candidate construction, expert annotation, independent verification, adjudication, and quality control. We also document the annotation guidelines, taxonomy, data format, inter-annotator agreement, and corpus statistics. HalluTruthQA-4K provides a reusable resource for hallucination detection, span-level error localization, explanation generation, factual verification, and the broader evaluation of factual reliability in Arabic language models.