Linguistics

Vision-Language Models (VLMs) increasingly power high-stakes applications, from medical imaging to autonomous systems, yet they routinely hallucinate, confidently describing content not present in the input. We investigate the root causes of these failure modes with a mechanistic analysis focusing on the decoder-based VLMs. We trace these failure modes to a geometric over-alignment: to bridge the modality gap required by attention mechanisms, decoder-based VLMs over-align visual embeddings with the text manifold, injecting a statistical linguistic bias that systematically overshadows fine-grained visual evidence. While prior work either aggressively closes this gap or suppresses hallucinations through expensive black-box decoding strategies, none addresses the underlying geometric cause. We provide the first quantitative characterization of this over-alignment, demonstrating that linguistic bias concentrates in the top principal components of a universal, dataset-agnostic text subspace. Building on this insight, we propose two complementary remedies: a training-free inference strategy and a bias-aware fine-tuning paradigm, both of which explicitly project out this subspace from visual representations. Our methods significantly reduce hallucinations across POPE, CHAIR, and AMBER benchmarks, and improve CLAIR scores on long-form captioning tasks, with the training-free variant adding no computational overhead over the base model.

This paper addresses automatic geospatial question answering over multilingual toponymic data. An original bilingual dataset of toponyms of the Republic of Tatarstan is introduced, comprising 9,688 structured records with linguistic, etymological, administrative, and coordinate information (93.1% georeferenced). Based on this dataset, a question-answering corpus of approximately 39,000 question-context-answer triples is constructed with guaranteed answer localization. A hybrid retriever integrates dense semantic indexing (multilingual-e5-large) with geospatial filtering via KD-trees and haversine distance. On 500 test queries, the hybrid search achieves Recall@1=0.988, Recall@5=1.000, and MRR=0.994, significantly outperforming BM25 and purely spatial methods. Among tested reader architectures (RuBERT, XLM-RoBERTa-large, T5-RUS), XLM-RoBERTa-large attains the best quality: EM=0.992, F1=0.994. On raw outputs, RuBERT models fail on coordinate questions (F1=0) while XLM-RoBERTa-large reaches F1=0.984; however, simple post-processing eliminates numerical gaps and restores RuBERT accuracy to 100%. This discrepancy stems from tokenization differences and pre-training corpora composition. All resources (dataset, QA corpus, model weights, web demo) are openly published on Hugging Face. Results apply to geospatial QA services, geocoding, and digital humanities in multilingual regions.

Emerging digital technologies are exacerbating the existing divide in Open Access Data (OAD) between high-and low-resource languages, excluding many communities from participating in the global digital transformation. In this PhD proposal, we aim to address this gap, focusing on the language coverage of Linked Open Data knowledge graphs (LOD KGs). First, we identify key variables that characterize language distribution in LOD, including the number of Wikipedia articles per language edition and the number of language-tagged entities in LOD KGs. These variables are analyzed across three major multilingual LOD KGs, DBpedia, BabelNet, and Wikidata, providing insights into the representation and distribution of languages within LOD. Building on this analysis, we intend to study the impact of cross-lingual transfer candidate selection on the task of multilingual KG completion. In particular, we plan to investigate strategies based on linguistic proximity and the availability of curated annotated alignments between languages. Language proximity also motivates us to explore the benefits of analogical reasoning that relies on (dis)similarities and has not yet been investigated to identify correspondences across languages to improve KG completion performance and enhance language coverage in LOD.

Emerging digital technologies are exacerbating the existing divide in Open Access Data (OAD) between high-and low-resource languages, excluding many communities from the global digital transformation. Multilingual Linked Open Data Knowledge Graphs (LOD KGs) could contribute to mitigating this divide through cross-lingual transfer; however, no clear quantitative definition of low-resource languages has yet been established in the context of LOD KGs. In this poster, we present a methodology to analyze the distribution of languages across LOD KGs and propose a preliminary multi-level categorization based on DBpedia, BabelNet, and Wikidata. This categorization is leveraged to bring a formal definition of low-, high-, and medium-resource languages that could be later leveraged to select cross-lingual transfer candidates.

Handwritten Text Recognition (HTR) for Arabic-script languages benefits from cross-language joint training under low-resource conditions, particularly when using CRNN-based models that combine convolutional encoders with sequence modeling. However, it remains unclear whether these improvements are better explained by shared visual representations or sequence-level dependencies. In this work, we conduct a controlled architectural study of line-level Arabic-script HTR, comparing CNN-only models with CTC decoding and CRNN models under identical single-script and multi-script training regimes. Experiments are performed on Arabic (KHATT), Urdu (NUST-UHWR), and Persian (PHTD) datasets under low-resource settings (K in {100, 500, 1000}). Our results show a clear divergence in transfer behavior: while CNN-only models exhibit limited or unstable improvements, CRNN models achieve better performance under multi-script training, particularly in the most data-constrained regimes. Focusing on transfer improvements (delta CER) rather than absolute performance, we find that cross-language improvements are associated with sequence-level modeling, while sharing visual representations learned by the CNN encoder, corresponding to similarities in character shapes across scripts, alone appears to be insufficient. This finding suggests that contextual modeling plays an important role in enabling effective transfer in low-resource scenarios, and that similar behavior may extend to other low-resource language settings.

This paper presents a comparative study of parameter-efficient fine-tuning (PEFT) methods, including LoRA and QLoRA, applied to the task of adapting large language models to the Bashkir language, a low-resource agglutinative language of the Turkic family. Experimental evaluation is conducted on a Bashkir text corpus of 71k documents (46.9M tokens) using models of various architectures: DistilGPT2, GPT-2 (base, medium), Phi-2, Qwen2.5-7B, DeepSeek-7B, and Mistral-7B. To improve the reliability of results, each configuration was trained with three different random seeds. The lowest perplexity on the test set was obtained for GPT-2 medium with full fine-tuning (3.34). Meanwhile, QLoRA applied to Mistral-7B (3.79) and Phi-2 (3.81) achieved comparable quality with over 40 times fewer trainable parameters. However, we also observed cases of significant quality degradation when using PEFT for certain architectures (e.g., DeepSeek-7B with rank 8, perplexity = 129.55), indicating that the outcome depends critically on the choice of the base model and its tokenizer. Additionally, a qualitative analysis of generated texts based on Bashkir prompts revealed that models with the best perplexity do not necessarily produce the most coherent outputs: QLoRA-tuned models generated monolingual Bashkir continuations, whereas the fully fine-tuned model with the lowest perplexity frequently switched to English. The results suggest that QLoRA on 7B-scale models offers an effective compromise between quality and computational cost for Bashkir. To ensure reproducibility, open data, code, and trained adapters will be released upon acceptance.

This paper presents the development and validation of an eye-tracking dataset designed to investigate how second-language (L2) learners process idiomatic expressions. While native speakers often rely on direct retrieval of figurative meanings, L2 speakers frequently adopt a literal-first approach, which incurs measurable cognitive costs. This resource captures these costs through ocular metrics recorded from Portuguese L1 speakers of English across all CEFR proficiency levels (A1-C2). Although the study uses entry-level 60 Hz hardware (Tobii Pro Spark), we demonstrate that this sampling rate provides sufficient data density to detect macro-cognitive events such as fixations and regressions in reading. Preliminary analysis validates the dataset by revealing a strong inverse correlation between language proficiency and regressive eye movements. Integrated into the MIA (Modeling Idiomaticity in Human and Artificial Language Processing) initiative, this dataset serves as a cognitively grounded benchmark for evaluating both human processing models and the alignment of large language models with human-like figurative understanding.

We present CodeEvolve, an evolutionary framework for improving program performance and code quality with Large Language Models (LLMs). CodeEvolve extends OpenEvolve with runtime-guided target selection, Monte Carlo Tree Search (MCTS), automated code refinement, and language-specific evaluation pipelines for Java and Salesforce Apex. The system uses Java Flight Recorder (JFR) profiles to build weighted component graphs and select optimization targets that account for most execution cost, reducing reliance on manual bottleneck identification. For each target, CodeEvolve generates candidate edits, evaluates them through build validation, unit tests, performance checks, static analysis, and LLM-based review, and retains only variants that preserve functional correctness. Across real-world optimization tasks, CodeEvolve improves performance and code metrics while maintaining correctness. On a large enterprise Java codebase, it achieves an average speedup of 15.22$\times$ across seven hotspot functions and outperforms single-pass LLM optimization on five of them. An ablation study on Apex optimization shows that the full MCTS-augmented configuration produces 19.5 valid programs out of 20 on average, indicating that search, filtering, and refinement each contribute to more reliable optimization.

The Tajik language, written in Cyrillic script, remains severely under-resourced in terms of publicly available natural language processing (NLP) toolkits, hindering both linguistic research and applied development. This paper introduces TajikNLP, an open-source Python library that provides the first comprehensive pipeline for processing authentic Tajik text while preserving the original Cyrillic orthography. The library implements a modular architecture centered around a unified Doc object, enabling sequential application of components for cleaning, normalization, tokenization (including subword BPE), morphemic segmentation, part-of-speech tagging, stemming, lemmatization, and sentence splitting. A novel unified morphology engine is introduced, offering controlled and deep analysis modes that significantly improve handling of Tajik's agglutinative nominal and verbal inflections. The release further incorporates a lexicon-based sentiment analyser and pre-trained Word2Vec/FastText embeddings loaded directly from the Hugging Face Hub. To ensure reproducibility and facilitate future research, four accompanying linguistic datasets -- a POS-tagged corpus (52.5k entries), a sentiment lexicon (3.5k entries), a toponym gazetteer (5.6k entries), and a personal names dataset (3.8k entries) -- have been openly published under permissive licenses. The library's reliability is validated by an extensive test suite of 616 automated tests achieving 93% source code coverage. TajikNLP thus establishes a foundational technological infrastructure for Tajik language processing, lowering the barrier to entry for both academic and industrial applications in low-resource Cyrillic-script environments.

This paper presents the first benchmark for the task of automatic part-of-speech (POS) tagging for the Tajik language. Despite the existence of multilingual language models demonstrating high effectiveness for many of the world's languages, their capacity for grammatical analysis of Tajik has remained unexplored until now. The aim of this study is to fill this gap through a systematic comparison of classical neural network architectures and modern multilingual transformers. Experiments were conducted on the TajPersParallel corpus, a parallel lexical resource comprising approximately 44,000 dictionary entries. Due to the absence of full-fledged example sentences in the current version of the corpus, the task was performed at the level of isolated lexical units, representing a challenging case of context-independent classification. The study compares the following architectures: a recurrent BiLSTM-CRF model, as well as multilingual models XLM-RoBERTa (large), mBERT, ParsBERT (Persian), and ruBERT (Russian), adapted using the parameter-efficient fine-tuning method LoRA. The testing results showed that the best performance is achieved by the mBERT + LoRA model (macro F1-score = 0.11, weighted F1-score = 0.62). It was established that in the absence of syntactic context, all models experience significant difficulty in resolving morphological ambiguity, successfully classifying primarily high-frequency classes ("noun," "adjective") while demonstrating zero effectiveness for rare function words. Zero-shot evaluation revealed the greatest typological proximity of Tajik to Persian (ParsBERT) and Russian (ruBERT). The obtained results form a foundation for further research and development in the field of automatic processing of the Tajik language.

Low-resource language varieties used by specific groups remain neglected in the development of Multilingual Language Models. A great deal of cross-lingual research focuses on inter-lingual language transfer which strives to align allied varieties and minimize differences between them. However, for low-resource varieties, linguistic dissimilarity is also an important cue allowing generalization to unseen varieties. Unlike prior approaches, we propose a two-stage Language Generalization framework that focuses on capturing variety-specific cues while also exploiting rich overlap offered by high-resource source variety. First, we propose TOPPing, a source-selection method specifically designed for low-resource varieties. Second, we suggest a lightweight VACAI-Bowl architecture that learns variety-specific attributes with one branch while a parallel branch captures variety-invariant attributes using adversarial training. We evaluate our framework on structural prediction tasks, which are among the few tasks available, as proxy for performance on other downstream tasks. Using VACAI-Bowl with TOPPing yields an average 54.62% improvement in the dependency parsing task, which serves as a proxy for performance on other downstream tasks across 10 low-resource varieties.

Cognitive neuroscience is fragmented into specialized models, each tailored to specific experimental paradigms, hence preventing a unified model of cognition in the human brain. Here, we introduce TRIBE v2, a tri-modal (video, audio and language) foundation model capable of predicting human brain activity in a variety of naturalistic and experimental conditions. Leveraging a unified dataset of over 1,000 hours of fMRI across 720 subjects, we demonstrate that our model accurately predicts high-resolution brain responses for novel stimuli, tasks and subjects, superseding traditional linear encoding models, delivering several-fold improvements in accuracy. Critically, TRIBE v2 enables in silico experimentation: tested on seminal visual and neuro-linguistic paradigms, it recovers a variety of results established by decades of empirical research. Finally, by extracting interpretable latent features, TRIBE v2 reveals the fine-grained topography of multisensory integration. These results establish artificial intelligence as a unifying framework for exploring the functional organization of the human brain.

Large language models (LLMs) have demonstrated impressive multilingual capabilities for well-resourced languages, yet their performance on low-resource African languages remains poorly understood and largely unevaluated. This paper presents Nsanku, a systematic benchmark that evaluates the zero-shot machine translation performance of 19 open-weight and proprietary LLMs across 43 Ghanaian languages paired with English. Evaluation sentences were sourced from the YouVersion Bible platform, providing 300 sentence pairs per language. Two complementary automatic metrics are employed: Bilingual Evaluation Understudy (BLEU) and Character n-gram F-Score (chrF), alongside an average accuracy score and a cross-language consistency dimension. Nsanku represents the most comprehensive LLM translation evaluation for Ghanaian languages conducted to date. Results show that gemini-2.5-flash achieves the highest overall average score of 26.88 (BLEU: 24.60, chrF: 29.16), followed by claude-sonnet-4-5 at 24.87 (BLEU: 22.46, chrF: 27.28) and gpt-4.1 at 23.20 (BLEU: 21.15, chrF: 25.24). Among open-weight models, kimi-k2-instruct-0905 leads at an average score of 20.87. A critical finding from the consistency analysis is that no model and no language reached the Leaders quadrant of high performance and high consistency simultaneously, indicating that current LLMs are not yet reliably usable for Ghanaian language translation at scale. Siwu achieved the highest per-language average score at 25.73 while Nkonya scored lowest at 11.65. Nsanku establishes a publicly available, community-extensible evaluation infrastructure for African language NLP research.

Knowledge transfer, especially across related languages, has been found beneficial for multilingual neural machine translation (MNMT), but some aspects are still under-explored and deserve further investigation. A joint vocabulary is most often applied to form a uniform word embedding space, but since the impact of a disjoint vocabulary on model performance is far less studied, there is no consensus on how much knowledge transfer is mainly due to vocabulary overlap. In this paper, we present systematic experiments with joint and disjoint vocabularies, and auxiliary languages related and unrelated to the source language. We design this experiment in an out-of-domain setup in order to emphasize transfer and the impact of the auxiliary language. As expected, we yield better results with more extensive vocabulary overlaps typical for related languages, but our experiments also show that domain-match and language relatedness are more important than a joint vocabulary.

To navigate partially observable visual environments, recent VLM agents increasingly internalize world modeling capabilities into their policies via explicit CoT reasoning, enabling them to mentally simulate futures before acting. However, relying solely on passive reasoning over visited states is insufficient for sparse-reward tasks, as it lacks the epistemic drive to actively uncover the known unknown'' required for robust generalization. We ask: Can VLM agents actively find signals that challenge and refine their internal world model through curiosity-driven exploration? In this work, we propose GLANCE, a unified framework that bridges reasoning and exploration by grounding the agent's linguistic world model into the stable visual representations of an evolving target network. Crucially, GLANCE leverages the discrepancy between linguistic prediction and visual reality as an intrinsic curiosity signal within reinforcement learning, steering the agent to actively explore areas where its internal model is uncertain. Extensive experiments across a series of agentic tasks show the effectiveness of GLANCE, and demonstrate that aligning what the agent thinks'' with ``what the agent sees'' is key to solving complex or sparse agentic tasks.

This paper is devoted to the adaptation of generative large language models for the Tajik language, a low-resource language with Cyrillic script. To overcome the shortage of digital text resources, the author created and publicly released the Tajik Web Corpus, the largest open-access corpus of Tajik, comprising 319,298 documents (~1.11 billion characters). On a subsample of 10,000 documents, 17 configurations were benchmarked, covering autoregressive, encoder-decoder, and encoder-only models with three fine-tuning strategies: full fine-tuning, LoRA, and QLoRA (ranks 8 and 16). Quality was assessed via perplexity and cross-entropy loss; peak GPU memory and training time were also recorded. Best results were achieved by Mistral 7B with QLoRA (r=16): mean perplexity 5.03, standard deviation 0.03. Increasing rank from 8 to 16 gave statistically insignificant improvement while raising memory consumption. For small GPT-2 family models, full fine-tuning yielded lower perplexity (3.48 for GPT-2 Medium) than LoRA (7.60-8.42), but induced catastrophic forgetting. The encoder-only XLM-RoBERTa showed the worst results (perplexity 59.3). The novelty lies in creating the largest verified Tajik corpus and the first systematic analysis of PEFT effectiveness for Tajik text generation. Practical value lies in recommendations for architecture and fine-tuning strategy selection, optimizing computational costs without substantial quality loss.

Recent large language models (LLMs) show strong speech recognition and translation capabilities for high-resource languages. However, African languages remain dramatically underrepresented in benchmarks, limiting their practical use in low-resource settings. While early benchmarks tested African languages and accents, they lacked exhaustive real-world noise and granular domain evaluations. We present AfriVox-v2, a comprehensive benchmark designed to test speech models under realistic African deployment conditions. AfriVox-v2 introduces "in the wild" unscripted audio for all supported languages. We also introduce strict domain verticalization, evaluating model accuracy across ten sectors including government, finance, health, and agriculture and conducting targeted tests on numbers and named entities. Finally, we benchmark a new generation of speech models, including Sahara-v2, Gemini 3 Flash, and the Omnilingual CTC models. Our results expose the true generalization gap of modern speech models in specialized, noisy African contexts and provide a reliable blueprint for developers building localized voice AI.

AI systems increasingly synthesize executable structure at runtime: LLMs generate programs, agents construct workflows,self-improving systems modify their own behavior. In classical homoiconic and staged languages, the transition from code representation to execution is unrestricted. eval is a language primitive, not a governed operation. We argue that in governed intelligent systems, this transition is an authority amplification: it converts symbolic structure into executable authority and must be mediated like any other effect. We present governed metaprogramming, a language design where program representations (machine forms) are first-class values, form manipulation is pure computation, and materialization (the transition from form to executable machine) is a governed effect subject to structural inspection. The governance system analyzes the proposed program's capability requirements, policy compliance, and resource estimates before permitting execution. We formalize two judgments: pure form evaluation (which emits no directives) and governed materialization (which emits exactly one governed directive). We prove three properties: purity of form manipulation, the no-bypass theorem, and boundary preservation. We implement the design in mashinTalk, a DSL for AI workflows compiling to BEAM byte code, and report on integration with 454 existing machine-checked Rocq theorems. The central contribution is reclassifying eval from a language primitive into a governed effect.

Language carries thought and coordination among humans but rarely reaches further along the spectrum of diverse intelligence. Yet non-neural systems -- from gene regulatory networks and microbial consortia to fungi -- are increasingly recognized as substrates of computation, decision-making and memory, making dialogue with non-human intelligence newly conceivable. Today such dialogue is attempted only by proxy: a large language model speaks on the system's behalf, so any intelligence on display originates from the model while the system itself remains silent. Here we ask whether the system can speak in its own voice. Following Wittgenstein, who located meaning in use, we treat communication as a game played with the system. Its internal dynamics are frozen as the nonlinear core of a reinforcement-learning policy, with only linear input and output interfaces trained. Through use and reward, the system's states and responses acquire meaning within the game, so playing becomes speaking. Because different architectures playing the same game optimize the same reward, their behaviors can all be read as pursuit of that reward; the game serves as a lingua franca across otherwise irreconcilable representations. Given a human prompt, a language model routes it to the game whose semantics best match it and designs an environmental state for which the desired action is the rational response, letting the system reply through its own behavior. Applied across diverse gene regulatory networks and reinforcement-learning tasks, the framework yields fluent dialogue without altering any system parameter, shows that well-trained agents of disparate origin converge on similar behavior, and reveals that specific GRN properties make a system easier or harder to talk with -- an inductive bias of the reservoir itself. Our framework opens a new route to conversing with any dynamical system on its own terms.

Text-in-image editing has become a key capability for visual content creation, yet existing benchmarks remain overwhelmingly English-centric and often conflate visual plausibility with semantic correctness. We introduce MULTITEXTEDIT, a controlled benchmark of 3,600 instances spanning 12 typologically diverse languages, 5 visual domains, and 7 editing operations. Language variants of each instance share a common visual base and are paired with a human-edited reference and region masks, isolating the language variable for cross-lingual comparison. To capture script-level errors that coarse text-matching metrics miss, such as missing diacritics, reversed RTL order, and mixed-script renderings, we introduce a language fidelity (LSF) metric scored by a two-stage LVM protocol that first traces the edited target text and then judges it in isolation, reaching a quadratic-weighted \k{appa} of 0.76 against native-speaker annotators. Evaluating 12 open-source and proprietary systems with LSF alongside standard semantic and mask-aware pixel metrics, we find pronounced cross-lingual degradation for every model, largest on Hebrew and Arabic and smallest on Dutch and Spanish, and concentrated in text accuracy and script fidelity rather than in coarse structural dimensions. We also uncover a pervasive semantic and pixel mismatch, where outputs preserve global layout and background fidelity yet distort script-specific forms.

Benchmarks within the OpenClaw ecosystem have thus far evaluated exclusively assistant-level tasks, leaving the academic-level capabilities of OpenClaw largely unexamined. We introduce AcademiClaw, a bilingual benchmark of 80 complex, long-horizon tasks sourced directly from university students' real academic workflows -- homework, research projects, competitions, and personal projects -- that they found current AI agents unable to solve effectively. Curated from 230 student-submitted candidates through rigorous expert review, the final task set spans 25+ professional domains, ranging from olympiad-level mathematics and linguistics problems to GPU-intensive reinforcement learning and full-stack system debugging, with 16 tasks requiring CUDA GPU execution. Each task executes in an isolated Docker sandbox and is scored on task completion by multi-dimensional rubrics combining six complementary techniques, with an independent five-category safety audit providing additional behavioral analysis. Experiments on six frontier models show that even the best achieves only a 55% pass rate. Further analysis uncovers sharp capability boundaries across task domains, divergent behavioral strategies among models, and a disconnect between token consumption and output quality, providing fine-grained diagnostic signals beyond what aggregate metrics reveal. We hope that AcademiClaw and its open-sourced data and code can serve as a useful resource for the OpenClaw community, driving progress toward agents that are more capable and versatile across the full breadth of real-world academic demands. All data and code are available at https://github.com/GAIR-NLP/AcademiClaw.

Transformer-based models achieve state-of-the-art dependency parsing for high-resource languages, yet their advantage over simpler architectures in low-resource settings remains poorly understood. We evaluate four parsers -- the Biaffine LSTM, Stack-Pointer Network, AfroXLMR-large, and RemBERT -- across ten typologically diverse languages, with a focus on low-resource African languages. We find that the Biaffine LSTM consistently outperforms transformer models in low-resource regimes, with transformers recovering their advantage as training data increases. The crossover falls within a resource range typical of treebanks for under-resourced languages. Morphological complexity (measured via MATTR) emerges as a significant secondary predictor of transformers' relative disadvantage after controlling for corpus size. These results indicate that the Biaffine LSTM may be better suited for syntactic tool development in low-resource regimes until sufficient annotated data is available to leverage the representational capacity of pre-trained transformers.

We present our shared task on evaluating the adaptability of LLMs and NLP systems across multiple languages and cultures. The task data consist of an extended version of our manually constructed BLEnD benchmark (Myung et al. 2024), covering more than 30 language-culture pairs, predominantly representing low-resource languages spoken across multiple continents. As the task is designed strictly for evaluation, participants were not permitted to use the data for training, fine-tuning, few-shot learning, or any other form of model modification. Our task includes two tracks: (a) Short-Answer Questions (SAQ) and (b) Multiple-Choice Questions (MCQ). Participants were required to predict labels and were allowed to submit any NLP system and adopt diverse modelling strategies, provided that the benchmark was used solely for evaluation. The task attracted more than 140 registered participants, and we received final submissions from 62 teams, along with 19 system description papers. We report the results and present an analysis of the best-performing systems and the most commonly adopted approaches. Furthermore, we discuss shared insights into open questions and challenges related to evaluation, misalignment, and methodological perspectives on model behaviour in low-resource languages and for under-represented cultures.

Most hallucination evaluations focus on English, leaving it unclear whether findings transfer to lower-resource languages. We investigate faithfulness hallucinations, defined as model-generated content that is fluent and plausible but diverges from the provided input or is internally inconsistent. Leveraging the multilingual MultiWikiQA dataset, we utilize the LettuceDetect framework to create synthetic hallucination datasets for 306 languages, from which we train token-level hallucination classifiers for 30 European languages. In this work, we present evaluations of model hallucinations on a selection of languages: English, Danish, German, and Icelandic. Using these classifiers, we evaluate the hallucination rates for Qwen3-0.6B, Qwen3-14B, Gemma-3-12B-IT, cogito-v1-preview-qwen-32B, and cogito-v1-preview-llama-70B. Our classifiers reveal notably higher hallucination rates for Qwen3-0.6B (up to 60% of answers containing at least one hallucination, peaking in Icelandic) and generally lower rates for larger models, with cogito-v1-preview-qwen-32B and cogito-v1-preview-llama-70B performing best on most languages. Hallucination rates are consistently higher for lower-resource languages, particularly Icelandic.

Large language models pick up social biases from the data they are trained on and carry those biases into downstream applications, often reinforcing stereotypes around gender, race, religion, disability, age, and socioeconomic status. The standard fixes (retraining on curated data or fine-tuning with human feedback) are expensive, need access to model weights, and risk degrading the model on other tasks. In this paper we take a different route: we debias the model at decoding time, treating bias mitigation as a structured search over candidate tokens without ever touching model weights. A separate Process Reward Model (PRM) acts as a judge, scoring each candidate for both fairness and fluency. We design three schemes of increasing sophistication (Best-of-N selection, Sequential critique-and-revise, and Constitutional self-audit) and evaluate them on four models (GPT-4o-mini, Llama 3.2 3B, Gemma 3 4B, Qwen 2.5 3B) across a 200-prompt bilingual benchmark in English and Urdu covering eight bias categories. Sequential debiasing proves the most effective, raising mean bias scores by up to +0.40 over baseline while preserving (and sometimes improving) fluency. We then extend all three schemes to open-ended generation, where each token is debiased on the fly, and introduce a lightweight Bias Guard gate that fires only on potentially biased words, keeping overhead near 2x for well-calibrated models. A formal overhead metric that separates generator cost from judge cost reveals that Best-of-N is effectively free on the generator side in a native implementation. GPT-4o-mini, included as a strong proprietary anchor, confirms that the framework scales with model capability; the three open-weight models show where current small-scale LLMs still struggle.

This paper presents the first comprehensive comparative analysis of modern machine learning architectures for transliteration between Tajik (Cyrillic script) and Persian (Arabic script). A key contribution is the creation and validation of a unique parallel corpus aggregated from multiple heterogeneous sources, including crowdsourced projects, lexicographic pairs, parallel texts of "Shahnameh", diplomatic articles, texts of "Masnavi-i Ma'navi", official terminology lists, and transliterated correspondences. The initial dataset comprised 328,253 sentence pairs; a representative subset of 40,000 pairs was formed using stratified random sampling. The experiment compared six classes of models: rule-based baseline, LSTM with attention, character-level Transformer, G2P Transformer (trained from scratch), pre-trained multilingual models (mBART, mT5 with LoRA), and byte-level ByT5. Results demonstrate the overwhelming superiority of ByT5 (chrF++ 87.4 for Tajik to Farsi, 80.1 for reverse). The G2P Transformer significantly outperformed mBART (72.3 vs. 62.2 chrF++) despite limited data. Models using subword tokenization (mT5) failed completely (chrF++ less than 18.5). The findings demonstrate that for accurate transliteration of the Tajik-Farsi pair, architectures operating at the byte or character level are unequivocally more effective than traditional multilingual Seq2Seq models relying on subword tokenization.

This discussion paper reflects on how quantitative approaches to historical linguistics interact with dataset properties. Drawing on two worked examples, we examine English data using quad-based concept modelling of Early Modern English discourse in EEBO-TCP (c. 1470s-1690s; 765M words) alongside SynFlow analysis of scientific writing in Royal Society Corpus 6.0.4 (1750-1799; drawn from a 78.6M-token open corpus). Through parallel comparison, the paper explores how each approach operationalises concepts, the data assumptions they entail, and the diachronic interpretations they support. We argue that comparative methodological reflection clarifies the limits of purely lexical, frequency-based approaches and highlights how dataset structure shapes the kinds of semantic change that quantitative methods can reliably detect.

Purpose: Vision-language models (VLMs) have shown promising performance in surgical visual question answering (VQA). However, existing surgical VQA datasets often contain linguistic shortcuts, where question phrasing implicitly constrains the answer space. It remains unclear whether reported performance reflects visual understanding or reliance on such linguistic shortcuts. Methods: We introduce SurgCheck, a diagnostic benchmark for quantifying linguistic shortcut reliance in surgical VQA. SurgCheck employs a paired-question design in which each surgical frame is associated with an original question containing entity names and a less-biased counterpart that removes these names while preserving identical visual content and ground-truth answers. The resulting performance gap provides a diagnostic signal of shortcut reliance. To ensure that the less-biased question remains well-defined even without entity names, four grounding cues are incorporated: bounding box, arrow, spatial position, and periphrasis. We evaluate both general-purpose and surgical-specific VLMs under zero-shot and fine-tuned settings on SurgCheck. To evaluate open-ended zero-shot responses, we introduce an LLM-as-a-judge evaluation protocol. Results: Using SurgCheck, we observe consistent performance degradation on less-biased questions across five VLMs, despite identical visual inputs. Text-only ablation reveals minimal performance drops for action and target prediction, indicating that action and target prediction is largely driven by linguistic shortcuts rather than visual reasoning. Conclusion: SurgCheck provides a controlled diagnostic framework that exposes failure modes masked by linguistic bias in existing surgical VQA benchmarks. Our findings demonstrate that strong benchmark performance does not necessarily imply faithful visual understanding, underscoring the need for bias-aware evaluation in surgical VQA.

Large Language Models (LLMs) have substantially advanced the field of Natural Language Processing (NLP), achieving state-of-the-art performance across a wide range of tasks. These improvements have been attributed, in part, to their emerging reasoning capabilities, which are enabled by large-scale training and increased model capacity. However, existing LLMs can generate erroneous responses when addressing complex queries that fall outside their training distribution, due to limited internal knowledge or the need for multi-step reasoning. To address these limitations, recent work has introduced large reasoning models (LRMs), which incorporate explicit internal reasoning processes to improve response accuracy. Additionally, state-of-the-art LRMs often comprise hundreds of billions of parameters and require several seconds per inference, even on advanced multi-GPU systems. These characteristics limit their practicality for deployment in conventional computing environments. Meanwhile, NLP research on multilingual LLMs continues to prioritize high-resource languages. However, these models exhibit limited performance in under-resourced languages, primarily due to insufficient language- and culture-specific training data. In this paper, we focus on Modern Greek, for which only a limited number of question answering (QA) datasets have been proposed, most of which are intended for model evaluation. To address this research gap in Greek QA, we make the following contributions: (i) CulturaQA, a high-quality LRM-generated and human-curated dataset, for Greek LLM training and evaluation; (ii) a memory-efficient LLM evaluation framework adaptable to diverse languages and QA tasks; (iii) Maistros 8B, a state-of-the-art open-weights Greek LLM developed via knowledge distillation and fine-tuning on CulturaQA; and (iv) a comprehensive evaluation of nine LLMs across nine human-curated Greek QA datasets.

Existing large-scale sign language resources typically provide supervision only at the level of raw video-text alignment and are often produced in laboratory settings. While such resources are important for semantic understanding, they do not directly provide a unified interface for open-world recognition and translation, or for modern pose-driven sign language video generation frameworks: 1. RGB-based pretrained recognition models depend heavily on fixed backgrounds or clothing conditions during recording, and are less robust in open-world settings than style-agnostic pose-processing models. 2. Recent pose-guided image/video generation models mostly use a unified keypoint representation such as DWPose as their control interface. At present, the sign language field still lacks a data resource that can directly interface with this modern pose-native paradigm while also targeting real-world open scenarios. We present SignVerse-2M, a large-scale multilingual pose-native dataset for sign language pose modeling and evaluation. Built from publicly available multilingual sign language video resources, it applies DWPose in a unified preprocessing pipeline to convert raw videos into 2D pose sequences that can be used directly for modeling, resulting in a consolidated corpus of about two million clips covering more than 55 sign languages. Unlike many laboratory datasets, this resource preserves the recording conditions and speaker diversity of real-world videos while reducing appearance variation through a unified pose representation. Toward this goal, we further provide the data construction pipeline, task definitions, and a simple SignDW Transformer baseline, demonstrating the feasibility of this resource for multilingual pose-space modeling and its compatibility with modern pose-driven pipelines, while discussing the evaluation claims it can support as well as its current limitations.

We test whether the causal inner product of \citet{park2024linear} -- defined by the unembedding covariance $Σ$ -- enables cross-lingual concept transport. Across 17 models and 4 language pairs, a matched-spectrum randomization test finds that Whitened Causal Alignment is indistinguishable from spectral regularization alone ($p = 0.95$). However, this failure reveals a broader phenomenon: anti-concentration is observed in residual-stream difference-of-means vectors across five architecture families ($p < 10^{-33}$) and supported by SAE features (e.g., $p = 4.5 \times 10^{-19}$) and linear probes on Gemma and Llama. We discover a \emph{dual geometry}: activation-space concept directions anti-concentrate in the spectral tail, while static unembedding-row contrasts \emph{concentrate} in high-variance directions ($p < 10^{-4}$). Split-injection causal interventions support the functional basis on Gemma and Llama (Cohen's $d$ up to $1.80$), and POS-tag probing across 8 models shows syntax preferentially encodes in the high-variance subspace in 6 of 8 architectures ($p < 0.013$), with the Qwen~2.5 family showing a significant reversal consistent with architecture-specific spectral structure. These results suggest transformers may rotate semantic content into spectrally quiet regions during contextualized processing, encoding concepts where they can be manipulated with reduced grammatical disruption.

Isolated word meanings are inherently uncertain. This uncertainty reduces when they are combined and anchored in context. We propose that grammar compresses meaning uncertainty cross-linguistically, which is reflected in brain and selectively disrupted in disorders. Compression was operationalized as the relative difference between non-contextual surprisal estimated from lexical frequency, and contextual surprisal from grammar-sensitive models. In narratives from 20 languages, contextual surprisal reduced frequency-based surprisal. This reduction closely tracked the surprisal cost of reversing word order, and scaled with richer, non-redundant lexis as organized by more complex but optimal dependency structure. During fMRI, surprisal and its reduction explained BOLD activity for comprehension and production in overlapping but distinct regions. Uncertainty reduction was significantly attenuated in aphasia, dementia, and schizophrenia, but remained intact where primary deficit is not language. These findings position uncertainty reduction via grammar as a foundational concept that illuminates principles, brain basis, and disruptions of language.

Despite significant strides in factual reliability, errors -- often termed hallucinations -- remain a major concern for generative AI, especially as LLMs are increasingly expected to be helpful in more complex or nuanced setups. Yet even in the simplest setting -- factoid question-answering with clear ground truth-frontier models without external tools continue to hallucinate. We argue that most factuality gains in this domain have come from expanding the model's knowledge boundary (encoding more facts) rather than improving awareness of that boundary (distinguishing known from unknown). We conjecture that the latter is inherently difficult: models may lack the discriminative power to perfectly separate truths from errors, creating an unavoidable tradeoff between eliminating hallucinations and preserving utility. This tradeoff dissolves under a different framing. If we understand hallucinations as confident errors -- incorrect information delivered without appropriate qualification -- a third path emerges beyond the answer-or-abstain dichotomy: expressing uncertainty. We propose faithful uncertainty: aligning linguistic uncertainty with intrinsic uncertainty. This is one facet of metacognition -- the ability to be aware of one's own uncertainty and to act on it. For direct interaction, acting on uncertainty means communicating it honestly; for agentic systems, it becomes the control layer governing when to search and what to trust. Metacognition is thus essential for LLMs to be both trustworthy and capable; we conclude by highlighting open problems for progress towards this objective.

Cross-attention patterns in neural machine translation (NMT) are widely used to study how multilingual models align linguistic structure. We report a systematic artifact in cross-attention analysis of NLLB-200 (600M): non-content tokens - primarily end-of-sequence tokens, language tags, and punctuation - capture 83 percent to 91 percent of total cross-attention mass. We term these "attention sinks," extending findings from LLMs [Xiao et al., 2023] to NMT cross-attention and identifying a causal mechanism rooted in vocabulary design rather than position bias. This artifact causes raw metrics to underestimate content-level similarity by nearly half (36.7 percent raw vs. 70.7 percent filtered), rendering uncorrected analyses unreliable. To address this, we validate a content-only filtering methodology that removes non-content tokens and renormalizes the distribution. Applying this to 1,000 parallel sentences across African languages (Swahili, Kikuyu, Somali, Luo) and non-African benchmarks (German, Turkish, Chinese, Hindi), we confirm the artifact is universal and recover masked linguistic signals: a 16.9 percentage-point gap between teacher-forcing and generation modes, clear language-family clustering in attention entropy, and a hidden Somali paradox linking SOV word order to monotonic alignment. We release our filtering toolkit and corrected datasets to support reproducible interpretability research on multilingual NMT.

This paper argues that contemporary multilingual NLP has converged on a fragile and misleading paradigm of incidental multilingualism. Today's LLMs appear multilingual largely because they are trained on massive, uneven web corpora, not because multilingual or multicultural competence has been treated as a core design objective. We contend that this paradigm systematically produces unequal, brittle, and opaque behavior across languages, with severe consequences in real-world and agentic deployments where models must reason, plan, and act across multiple linguistic contexts. We report a focused empirical study of two practical questions: which languages models self-report as supported and which languages they actually respond in across multilingual prompts. We additionally demonstrate how even a simple language-change attack can surface these failures and expose hidden assumptions about language in LLM-based systems. To address this, we call for a shift toward multilingualism by design: a research agenda that treats equitable multilingual performance, cultural grounding, and cross-lingual behavioral understanding as first-class goals in all aspects of the model pipeline.

The record of the beginning of the most widespread legal system in the world is contained in millions of pages of handwritten text. Most of the records of the first centuries of the Anglo-American legal system are hand-written in a highly abbreviated form of medieval Latin which only a few dozen scholars in the world are trained to read. In this interdisciplinary project, we construct a dataset of 4029 lines of text across 193 medieval criminal and civil cases. We then use the dataset to train an open-source end-to-end pipeline for transcribing these manuscripts. We first train standard neural network architectures for line segmentation and handwriting recognition (R-Blla and CNN+LSTM with CTC decoding, respectively) and show that they can already achieve 79% word accuracy, despite the relatively small training set and the challenge of expanding abbreviations. We then demonstrate that simple post-processing significantly boosts accuracy: adding an n-gram language model to the CTC decoder improves word accuracy to 82%, while asking Gemini Pro 3 to correct mistakes boosts accuracy to 88%. Finally, we compare the CNN+LSTM architecture with TrOCR, a transformer-based OCR architecture, demonstrating that TrOCR shows comparable word accuracy but worse character accuracy due to its over-willingness to guess, making it harder for humans to infer the correct reading. We incorporated our pipeline into a web portal (glyphmachina.com), opening up the English legal tradition to legal scholars, medievalists, and students.

We investigate the extent to which cosine similarity between paragraph embeddings is invariant under machine translation, using the Manifesto Corpus of over 2,800 political party platforms in 28 languages translated to English via the EU eTranslation service. Rather than measuring translation-induced semantic shift directly we measure the stability of pairwise similarity relationships across embedding models, and use inter-model disagreement on original-language text as a calibrated invariance threshold. This yields a per-language non-inferiority test for four hypotheses about how translation interacts with embedding choice, with verdicts that distinguish languages where translation demonstrably preserves semantic structure from those where it demonstrably degrades it and from those where the available evidence does not resolve the question. The framework is corpus- and pipeline-agnostic and extends naturally to downstream tasks. Applied to our data, it identifies ten languages with translation invariance and four with detectable distortion.

Animal-welfare advocates produce a lot of writing, and increasingly that writing trains the language models that millions of people then ask about animal welfare. Using vocabulary-matched stance-contrast probes on a held-out animal-welfare benchmark, we measure how each of ten linguistic features changes Llama-3.2-1B's preference for pro-animal-welfare reasoning when used as fine-tuning data. Eight of the ten features produce statistically significant shifts. Seven move the model toward stronger pro-animal-welfare reasoning: assertive certainty, explicit moral vocabulary, emotion words, evaluative claims, narrative structure, depicted harm severity, and immediate temporal framing. Two move it the other way: hedged language and concrete sensory description both dilute the pro-animal-welfare stance. First-person perspective has no statistically significant effect. The practical recommendation for anyone writing animal-welfare text that may end up in LLM training corpora: assert a position rather than describe a scene neutrally. The features that shift the model are the ones that make the writer's position explicit; the features that dilute it hold animal-welfare content but withhold stance.

There is a significant gap in evaluating cultural reasoning in LLMs using conversational datasets that capture culturally rich and dialectal contexts. Most Arabic benchmarks focus on short text snippets in Modern Standard Arabic (MSA), overlooking the cultural nuances that naturally arise in dialogues. To address this gap, we introduce ArabCulture-Dialogue, a culturally grounded conversational dataset covering 13 Arabic-speaking countries, in both MSA and each country's respective dialect, spanning 12 daily-life topics and 54 fine-grained subtopics. We utilize the dataset to form three benchmarking tasks: (i) multiple-choice cultural reasoning, (ii) machine translation between MSA and dialects, and (iii) dialect-steering generation. Our experiments indicate that the performance gap between MSA and Arabic dialects still exists, whereby the models perform worse on all three tasks in the dialectal setup, compared to the MSA one.

High-quality corpora are essential for advancing Natural Language Processing (NLP) in Portuguese. Building on previous encoder-only models such as BERTimbau and Albertina PT-BR, we introduce NorBERTo, a modern encoder based on the ModernBERT architecture, featuring long-context support and efficient attention mechanisms. NorBERTo is trained on Aurora-PT, a newly curated Brazilian Portuguese corpus comprising 331 billion GPT-2 tokens collected from diverse web sources and existing multilingual datasets. We systematically benchmark NorBERTo against Strong baselines on semantic similarity, textual entailment and classification tasks using standardized datasets such as ASSIN 2 and PLUE. On PLUE, NorBERTo-large achieves the best results among the encoder models we evaluated, notably reaching 0.9191 F1 on MRPC and 0.7689 accuracy on RTE. On ASSIN 2, NorBERTo-large attains the highest entailment F1 (~0.904) among all encoders considered, although Albertina-900M and BERTimbau-large still hold an advantage. To the best of our knowledge, Aurora-PT is currently the largest openly available monolingual Portuguese corpus, surpassing previous resources. NorBERTo provides a modern, mid-sized encoder designed for realistic deployment scenarios: it is straight-forward to fine-tune, efficient to serve, and well suited as a backbone for retrieval-augmented generation and other downstream Portuguese NLP systems.

Recent research has shown that filtering massive English web corpora into high-quality subsets significantly improves training efficiency. However, for high-resource non-English languages like German, French, or Japanese, aggressive filtering creates a strategic dilemma: should practitioners prioritize diversity by training once on large amounts of lightly filtered web data, or prioritize quality by strictly filtering for a high-quality core and repeating it over multiple epochs? We investigate this trade-off for German by constructing hierarchical quality filters applied to 500M web documents, comparing multi-epoch training on the filtered subsets against single-pass training on a diverse corpus. Our experiments across multiple model scales and token budgets show that repeating high-quality data consistently outperforms single-pass training on larger, less filtered sets. Notably, the performance gap persists even after 7 epochs. Our findings suggest that for non-English LLMs, semantic concentration through quality filtering offers a more viable path to efficient language modeling than simply maximizing unique data volume. We release our German language models (called Boldt), as well as our cleaned evaluation benchmarks to the research community. Our experiments indicate that they achieve state-of-the-art results despite training on 10-360x fewer tokens than comparable models.

Integrating multimodal foundation models into enterprise ecosystems presents a fundamental software architecture challenge. Architects must balance competing quality attributes: the high latency and non-determinism of vision language action (VLA) models versus the strict determinism and real-time performance required by enterprise control loops. In this study, we propose an architectural pattern language for visual agents that separates fast, deterministic reflexes from slow, probabilistic supervision. It consists of four architectural design patterns: (1) Hybrid Affordance Integration, (2) Adaptive Visual Anchoring, (3) Visual Hierarchy Synthesis, and (4) Semantic Scene Graph.

Detecting language ideologies is a valuable yet complex task for understanding how identities are constructed through discourse. In Luxembourg's multicultural and multilingual society, language ideologies reflect more than simple preferences: they carry deep cultural and social meanings, shaping identities and social belonging. Following recent developments in applying Natural Language Processing tools to linguistics and social science, this paper explores the potential of large language models to assist in the detection of language ideologies. We manually annotate a corpus of user comments in Luxembourgish with predefined ideological categories and then evaluate the performance of large language models under varying prompt conditions to assess their ability to replicate these human annotations. Since Luxembourgish is a small language and poorly represented in the LLMs' training data, we also investigate whether machine-translating the data to high-resource languages increases performance on the ideology detection task. Our findings suggest that, while LLMs are not yet fully optimized for a multi-class ideological annotation task, they are practical tools to identify language ideological content.

Models of sign language have historically lagged behind those for spoken language (text and speech). Recent work has greatly improved their performance on tasks like sign language translation and isolated sign recognition. However, it remains unclear to what extent existing models capture various linguistic phenomena of sign language, and how well they use cues from the multiple articulators used in sign language (hands, upper body, face). We introduce a new benchmark dataset for American Sign Language, ASL Minimal Translation Pairs (ASL-MTP), divided into multiple types of sign language phenomena and corresponding minimal pairs of translations, for performing such linguistic analyses. As a case study, we use ASL-MTP to analyze a state-of-the-art ASL-to-English translation model. We conduct a targeted analysis of the model by ablating various input cues during training and inference and evaluating on the phenomena in ASL-MTP. Our results show that, while the model performs above chance level on most of the phenomena, it relies strongly on manual cues while often missing crucial non-manual cues.

This paper introduces a systematic evaluation framework grounded in the Interagency Language Roundtable (ILR) Skill Level Descriptions and applies it to Claude (Sonnet 4.6) across six languages: English, French, Romanian, Spanish, Italian, and German. We administer a battery of 12 semantically equivalent prompt clusters spanning ILR complexity levels 1 through 3+, collect 216 responses (12 prompts, 6 languages, 3 runs), and analyze outputs through a two-layer methodology combining automated quantitative metrics with expert ILR qualitative assessment. Quantitative analysis reveals that French responses are approximately 30% longer than German responses on identical prompts, and that creative and affective clusters show the highest cross-lingual surface divergence. Qualitative analysis, conducted by a six-language professional with 12 years of ILR/OPI assessment experience, identifies five cross-lingual variation patterns: systematic differences in pragmatic disambiguation strategies, aesthetic and literary tradition divergence in creative output, language-internal technical terminology norms, cultural calibration gaps evidenced by the absence of culture-specific content in favor of culturally neutralized templates, and language-specific institutional referral behavior in emotional support responses. We argue that ILR-informed expert judgment applied to LLM outputs constitutes a novel and underreported evaluation methodology that complements purely computational benchmarks, and that cross-lingual output variation in Claude is interpretable, domain-dependent, and consequential for equitable multilingual AI deployment.

Languages of the world vary concerning the order of subject, object and verb. The most frequent dominant orders are SOV and SVO, and researchers have tailored models to this fact. However, there are still languages whose dominant order does not conform to these expectations or even lack a dominant order. Here we show that across linguistic families and macroareas, word order variation within languages is shaped by the principle of swap distance minimization even when the dominant order is not SOV/SVO and even when a dominant order is lacking.

Aspect-based Sentiment Analysis (ABSA) extracts fine-grained opinions toward specific aspects within text but remains largely English-focused despite major advances in transformer-based and instruction-tuned models. This work presents a multilingual evaluation of state-of-the-art ABSA approaches across seven languages (English, German, French, Dutch, Russian, Spanish, and Czech) and four subtasks (ACD, ACSA, TASD, ASQP). We systematically compare different transformer architectures under zero-resource, data-only, and full-resource settings, using cross-lingual transfer, code-switching and machine translation. Fine-tuned Large Language Models (LLMs) achieve the highest overall scores, particularly in complex generative tasks, while few-shot counterparts approach this performance in simpler setups, where smaller encoder models also remain competitive. Cross-lingual training on multiple non-target languages yields the strongest transfer for fine-tuned LLMs, while smaller encoder or seq-to-seq models benefit most from code-switching, highlighting architecture-specific strategies for multilingual ABSA. We further contribute two new German datasets, an adapted GERestaurant and the first German ASQP dataset (GERest), to encourage multilingual ABSA research beyond English.

Timely and reliable multilingual communication is critical during natural and human-induced disasters, but developing effective solutions for crisis communication is limited by the scarcity of curated parallel data. We propose a domain-adaptive pipeline that expands a small reference corpus, by retrieving and filtering data from general corpora. We use the resulting dataset to fine-tune a small language model for crisis-domain translation and then apply preference optimization to bias outputs toward CEFR A2-level English. Automatic and human evaluation shows that this approach improves readability, while maintaining strong adequacy. Our results indicate that simplified English, combined with domain adaptation, can function as a practical lingua franca for emergency communication when full multilingual coverage is not feasible.

Large language models (LLMs) demonstrate strong multilingual capabilities, yet often fail to consistently generate responses in the intended language, exhibiting a phenomenon known as language confusion. Prior mitigation approaches based on sequence-level fine-tuning, such as DPO, ORPO, and GRPO, operate at the level of entire responses and can lead to unintended degradation of general model capabilities, motivating the need for more fine-grained alternatives. To address this, we introduce Token-Level Policy Optimization (TLPO), a fine-tuning framework designed to mitigate language confusion through localized, token-level updates. TLPO identifies error-prone positions, explores alternative candidate tokens, and updates the policy using a tailored objective to suppress error-inducing outputs at a granular level. This selective intervention enables effective mitigation of language confusion without compromising the model's general abilities. Experiments on multiple multilingual LLMs across diverse languages demonstrate that TLPO significantly outperforms baselines in improving language consistency while preserving downstream task accuracy.

LLMs and speech assistants are increasingly used for task-oriented interactions, yet their evaluation often relies on controlled scenarios that fail to capture the variability and complexity of real user requests. Drink ordering, for example, involves diverse named entities, drink types, sizes, customizations, and brand-specific terminology, as well as spontaneous speech phenomena such as hesitations and self-corrections. To address this gap, we introduce StarDrinks, a test set in English and Korean containing speech utterances features, transcriptions, and annotated slots. Our dataset supports speech-to-slots SLU, transcription-to-slots NLU, and speech-to-transcription ASR evaluation, providing a realistic benchmark for model robustness and generalization in a linguistically rich, real-world task.

Language models (LMs) behave more like humans when their cognitive resources are restricted, particularly in predicting sentence processing costs such as reading times. However, it remains unclear whether such constraints similarly affect sentence comprehension strategies. Besides, existing methods do not directly target the balance between memory storage and sentence processing, which is central to human working memory. To address this issue, we propose a dual-task paradigm that combines an arithmetic computation task with a sentence comprehension task, such as "The 2 cocktail + blended 3 =..." Our experiments show that under dual-task conditions, GPT-4o, o3-mini, and o4-mini shift toward plausibility-based comprehension, mirroring humans' rational inference. Specifically, these models show a greater accuracy gap between plausible sentences (e.g., "The cocktail was blended by the bartender") and implausible sentences (e.g., "The bartender was blended by the cocktail") in the dual-task condition compared to the single-task conditions. These findings suggest that constraints on the balance between memory and processing resources promote rational inference in LMs. More broadly, they support the view that human-like sentence comprehension fundamentally arises from the allocation of limited cognitive resources.

Speech recruits the same executive, attentional, and working memory processes underlying instrumental activities of daily living, or IADLs, providing a non-invasive proxy for cognitive assessment. Yet most speech-based dementia detection systems depend on transcription, discard within-recording temporal structure, and are validated on a single English corpus with known recording artifacts. We propose an ASR-agnostic framework operating directly on Mel spectrograms. Our key contribution is extracting spectrotemporal displacement fields from consecutive spectrogram frames, capturing shifting spectral energy patterns as digital biomarkers of cognitive decline. These features are fused with CNN-ConvGRU acoustic embeddings via a learned cross-attention mechanism and aggregated using a Transformer encoder with learnable query pooling. A composite temporal loss enforces smoothness and contrastive coherence across segments. We train independent models on English DementiaBank, Slovak EWA-DB, and Spanish Ivanova corpora, using clinical elicitation protocols taxing IADL-relevant cognitive domains. The Slovak model achieves 83.9% accuracy, and Spanish achieves, while the English baseline yields 53.2%, confirming known artifacts. Cross-lingual ablation studies reveal distinct fusion regimes: removing cross-attention collapses Spanish performance to 53.7%, below unimodal models, while the Slovak audio encoder alone outperforms the full model, 93.7% vs. 83.9%, and all English configurations remain near chance. Thus, multimodal fusion's value is corpus-dependent: essential when signal is distributed across modalities, counterproductive when one dominates, and irrelevant when no signal exists. Auxiliary temporal losses converge to language-invariant values, indicating cross-lingual architectural stability.

AI-powered language learning tools increasingly provide instant, personalised feedback to millions of learners worldwide. However, this feedback can fail in ways that are difficult for learners--and even teachers--to detect, potentially reinforcing misconceptions and eroding learning outcomes over extended use. We present a portion of L2-Bench, a benchmark for evaluating AI systems in language education that includes (but is not limited to) six critical dimensions of effective feedback: diagnostic accuracy, awareness of appropriacy, causes of error, prioritisation, guidance for improvement, and supporting self-regulation. We analyse how AI systems can fail with respect to these dimensions. These failures, which we argue are conducive to "explainability pitfalls," are AI-generated explanations that appear helpful on the surface but are fundamentally flawed, increasing the risk of attainment, human-AI interaction, and socioaffective harms. We discuss how the specific context of language learning amplifies these risks and outline open questions we believe merit more attention when designing evaluation frameworks specifically. Our analysis aims to expand the community's understanding of both the typology of explainability pitfalls and the contextual dynamics in which they may occur in order to encourage AI developers to better design safe, trustworthy, and effective AI explanations.

Preserving a speaker's voice identity while generating speech in a different language remains a fundamental challenge in spoken language technology, particularly in specialized domains such as scientific communication. In this paper, we address this challenge through our system submission to the International Conference on Spoken Language Translation (IWSLT 2026), the Cross-Lingual Voice Cloning shared task. First, we evaluate several state-of-the-art voice cloning models for cross-lingual speech generation of scientific texts in Arabic, Chinese, and French. Then, we build voice cloning systems based on the OmniVoice foundation model. We employ data augmentation via multi-model ensemble distillation from the ACL 60/60 corpus. We investigate the effect of using this synthetic data for fine-tuning, demonstrating improvements in intelligibility (WER & CER) and speaker similarity (SIM), with gains varying across languages.

We present a unified quantitative analysis of the Currier A/B language distinction in the Voynich Manuscript, proceeding in two stages. First, we confirm that the distinction is genuine: a Beta-Binomial mixture model applied to character-pair substitution ratios across 185 folios, without access to Currier's labels, selects 2 by BIC and predicts held-out folio labels at 89% accuracy. Second, we show that the A/B contrast is not primitive but is a low-resolution projection of a higher-dimensional generative system. Its dominant component is a discrete boolean "switch" set once per folio, governing the vowel following the digraphs ch and sh. A two-state binomial mixture achieves Delta\ AIC = 2,549 over a single-state model and assigns 195 of 197 folios unambiguously. This switch does not operate uniformly: word templates divide into fixed contexts and switchable contexts, with template identity accounting for 92% of variance.

Reliable evaluation of large language model (LLM)-generated summaries remains an open challenge, particularly across heterogeneous domains and document lengths. We conduct a comprehensive meta-evaluation of 14 automatic summarization metrics and LLM-based evaluators across seven datasets spanning five domains, covering documents from short news articles to long scientific, governmental, and legal texts (2K-27K words) with over 1,500 human-annotated summaries. Our results show that traditional lexical overlap metrics (e.g., ROUGE, BLEU) exhibit weak or negative correlation with human judgments, while task-specific neural metrics and LLM-based evaluators achieve substantially higher alignment, especially for linguistic quality assessment. Leveraging these findings, we propose LLM-ReSum, a self-reflective summarization framework that integrates LLM-based evaluation and generation in a closed feedback loop without model finetuning. Across three domains, LLM-ReSum improves low-quality summaries by up to 33% in factual accuracy and 39% in coverage, with human evaluators preferring refined summaries in 89% of cases. We additionally introduce PatentSumEval, a new human-annotated benchmark for legal document summarization comprising 180 expert-evaluated summaries. All code and datasets will be released in GitHub.

Achieving reliable control of Large Language Models (LLMs) requires a precise, scalable understanding of how they interpret linguistic cues. We introduce a rigorous framework using Shapley values to quantify the steering effect of individual adjectives on model performance, moving beyond anecdotal heuristics to principled attribution. Applying this method to 100 adjectives across a diverse suite of models (including o3, gpt-4o-mini, phi-3, llama-3-70b, and deepseek-r1) on the MMLU benchmark, we uncover several critical findings for AI alignment. First, we find that a small subset of adjectives act as disproportionately powerful "levers," yet their effects are not universal. Cross-model analysis reveals a "family effect": models of a shared lineage exhibit correlated sensitivity profiles, while architecturally distinct models react in a largely uncorrelated manner, challenging the notion of a one-size-fits-all prompting strategy. Second, focused follow-up studies demonstrate that the steering direction of these powerful adjectives is not intrinsic but is highly contingent on their syntactic role and position within the prompt. For larger models like gpt-4o-mini, we provide the first quantitative evidence of strong, non-additive interaction effects where adjectives can synergistically amplify, antagonistically dampen, or even reverse each other's impact. In contrast, smaller models like phi-3 exhibit a more literal and less compositional response. These results suggest that as models scale, their interpretation of prompts becomes more sophisticated but also less predictable, posing a significant challenge for robustly steering model behavior and highlighting the need for compositional and model-specific alignment techniques.

We investigate linguistic biases in LLM-based restaurant and product recommendations given prompts varying across Southern American English (AE), Indian English (IE), and Code-Switched Hindi-English dialects, using the Yelp Open dataset (Yelp Inc., 2023) and Walmart product reviews dataset (PromptCloud,2020). We add lists of restaurant and product names balanced by cuisine type and product category to the prompts given to the LLM, and we zero-shot prompt the LLMs in a cold-start setting to select the top-20 restaurant and product recommendations from these lists for each of the dialect-varied prompts. We prompt LLMs using different list samples across 20 seeds for better generalization, and aggregate per cuisine-type and per category response counts for each seed, question/prompt, and LLM model. We run mixed-effects regression models for each model family and topic (restaurant/product) with the aggregate response counts as the dependent, and conduct likelihood ratio tests for the fixed effects with post-hoc pairwise testing of estimated marginal means differences, to investigate group-level differences in recommendation counts by model size and dialect type. Results show that dialect plays a role in the type of restaurant selected across the models tested with the mistral-small-3.1 model and both the llama-3.1 family models tested showing more sensitivity to Indian English and Code-Switched prompts. In terms of product recommendations, the llama-3.1-70B-model is particularly sensitive to Code-Switched prompts in four out of seven categories, and more beauty and home category recommendations are seen when using the Indian English and Code-Switched prompts for larger and smaller models, respectively. No broad trends are seen in the model-size based differences, with differing recommendations based on model sizes conditioned by the type of dialect.

Do large language models (LLMs) truly acquire embodied cognition and cultural conventions from text? We introduce demonstratives, fundamental spatial expressions like "this/that" in English and "zhè/nà" in Chinese, as a novel probe for grounded knowledge. Using 6,400 responses from 320 native speakers, we establish a human baseline: English speakers reliably distinguish proximal-distal referents but struggle with perspective-taking, while Chinese speakers switch perspectives fluently but tolerate distal ambiguity. In contrast, five state-of-the-art LLMs fail to inherently understand the proximal-distal contrast and show no cultural differences, defaulting to English-centric reasoning. Our study contributes (i) a new task, based on demonstratives, as a new lens for evaluating embodied cognition and cultural conventions; (ii) empirical evidence of cross-cultural asymmetries in human interpretation; (iii) a new perspective on the egocentric-sociocentric debate, showing both orientations coexist but vary across languages; and (iv) a call to address individual variation in future model design.

A multilingual collection may contain useful knowledge in other languages to supplement and correct the facts in the original language for Retrieval-Augmented Generation (RAG). However, the vanilla approach that simply concatenates multiple pieces of knowledge from different languages into the context may fail to improve effectiveness due to the potential disparities across languages. To better leverage multilingual knowledge, we propose CroSearch-R1, a search-augmented reinforcement learning framework to integrate multilingual knowledge into the Group Relative Policy Optimization (GRPO) process. In particular, the approach adopts a multi-turn retrieval strategy with cross-lingual knowledge integration to dynamically align the knowledge from other languages as supplementary evidence into a unified representation space. Furthermore, we introduce a multilingual rollout mechanism to optimize reasoning transferability across languages. Experimental results demonstrate that our framework effectively leverages cross-lingual complementarity and improves the effectiveness of RAG with multilingual collections.