Linguistics

Sentence embeddings are a foundational component for semantic search, clustering, classification, and retrieval-augmented generation. This paper presents embeddingmagibu-200m, a Turkish-focused sentence embedding model that produces 768-dimensional L2-normalized vectors and supports an 8,192-token context window, far exceeding the 512-token limit of earlier BERT-based Turkish encoders. Instead of full pretraining, an efficient three-stage adaptation pipeline is introduced: (1) construct a Turkish-optimized multilingual tokenizer with a 131,072 vocabulary by pruning redundant tokens from the teacher's vocabulary and incorporating multilingual tokens via frequency analysis on a 40-language corpus, (2) clone a teacher embedding model while preserving transformer backbone weights and initializing a compatible embedding table for the new vocabulary via mean-composition token mapping, and (3) perform offline embedding distillation from precomputed teacher vectors using a cosine similarity objective over a balanced 40-language Wikipedia corpus. The resulting student model contains approximately 200M parameters and trains in roughly four hours on a single GPU by avoiding online teacher inference during training, at a total cost of $5-$20. Empirically, Pearson/Spearman correlations of 77.55%/77.45% are obtained on STSbTR, surpassing the 300M-parameter teacher model (73.84%/72.92%). On TR-MTEB (26 tasks), a mean score of 63.9% is achieved (7th out of 26 models), providing a competitive cost-quality trade-off with 33% fewer parameters than the teacher. To facilitate reproducibility and downstream use, all artifacts are released including model weights, tokenizer files, precomputed embedding datasets, and open-source cloning and distillation tooling.

This paper introduces Code Bench, a benchmark capable of evaluating Large Language Models (LLMs) concise code generation abilities in 60 programming languages. Based on code golf, a recreational programming competition focused on minimal character or byte solutions, the benchmark provides a distinctive measure of LLMs ability to produce efficient, concise code. Unlike existing benchmarks limited by fixed problem sets and language coverage, CodeGolf Bench leverages the code.golf platform to provide new problems and live human performance baselines. Evaluation of nine LLMs on Python and C++ tasks demonstrates that reasoning models significantly outperform non-reasoning models, achieving best average percentile of 70.97%. This performance gap is particularly pronounced in C++, highlighting reasoning's importance for languages with strict syntax requirements. Non-reasoning models struggle more with efficiency optimization across both languages, with best percentiles significantly lower than reasoning counterparts. CodeGolf Bench offers a dynamic framework for evaluating LLM code generation capabilities against evolving human performance on code golf.

Pattern languages are a classical model in formal language theory and algorithmic learning theory. This note formulates the problem of computing the inclusion depth of a pattern language: the length of the longest strict inclusion chain from the universal pattern language to the language generated by a given pattern. Inclusion depth captures the mind-change complexity of pattern identification from positive data. The central open question is whether the inclusion depth ID_Sigma(p) is computable for every pattern p over every finite alphabet Sigma with at least two symbols, and whether it is computable in polynomial time. A simple conjectured formula, ID_Sigma(p) = 2|p| - #var(p) - 1, would imply a linear-time algorithm. The problem connects pattern language inclusion, combinatorics on words, language identification in the limit, and mind-change-bounded learning.

The dominance of colonial languages in African education and scientific communication limits how hundreds of millions of speakers of African languages access and produce scientific knowledge. A core obstacle is the lack of established scientific terminology in these languages. We introduce AfriScience-MT, a parallel corpus covering six African languages (Amharic, Hausa, Luganda, Northern Sotho, Yorùbá, and isiZulu) across 11 scientific domains. Professional translators, working with expert science communicators, translated plain-language summaries of scientific papers into each target language and created new terms where none existed. We benchmark machine translation systems and large language models in zero-shot, few-shot, and fine-tuned settings. Our results show that closed-source models outperform all open-source models at both the sentence and document levels: GPT-5.4 and Gemini-3.1-Flash-Lite lead with average sentence-level COMET scores of 68.3 and 68.0, respectively, and tie at an average document-level COMET of 48.3. Among open systems, fine-tuned NLLB-1.3B reaches 67.3 at the sentence level, and TranslateGemma-12B reaches 44.0 at the document level with 1-shot in-context learning. We release AfriScience-MT to support benchmarking and document-level scientific MT for African languages.

Legal NLP benchmarks overwhelmingly evaluate a single language or aggregate tasks that differ fundamentally across jurisdictions, making cross-lingual comparison impossible. We introduce Multi-Legal-Bench, the first cross-jurisdictional legal benchmark that evaluates identical tasks across six countries (Ukraine, France, Netherlands, Poland, Czech Republic, Lithuania), four language families, and 134 million court decisions. The benchmark defines five tasks court-type classification, judgment form classification, case-outcome prediction, legal norm extraction, and cause category prediction mapped to structured metadata from national court registries, forming a deliberately sparse 5x6 task-jurisdiction matrix (20 of 30 cells filled). We evaluate 7 frontier LLMs under zero-shot and 3-shot prompting via AWS Bedrock, with 4 additional small/medium models (3-12B) for scaling analysis. Our results reveal that: (1) task-dependent few-shot effects discovered in Ukrainian replicate across all jurisdictions; (2) no single model dominates any language rankings shift with both task and jurisdiction; (3) cross-lingual few-shot transfer does not follow language proximity: UA->FR (Romance, -2.1 pp) transfers better than UA->PL (Slavic, -13.7 pp), with label-set alignment predicting transfer quality better than language family; and (4) tokenizer fertility, despite a 2.3x spread, does not significantly predict cross-lingual accuracy (r=-0.27, p=0.14), suggesting that model architecture and pretraining data dominate tokenizer efficiency. We release all data, prompts, and model predictions.

Mixture-of-Experts (MoE) models are widely used to scale language models, yet their expert routing behavior and adaptation in a multilingual setting remain underexplored. In this work, we study multilingual routing dynamics during continual pre-training of an English-centric MoE model on a multilingual corpus, analyzing how expert usage varies across languages. We find that continual multilingual pre-training leads to diffused, language-agnostic routing in early and middle layers, with language specialization primarily emerging in the final layers. We also show that token-level vocabulary overlap between languages plays an important role in how languages are routed. Motivated by these findings, we propose a parameter-efficient adaptation strategy that updates language-specific and shared experts in the final MoE layers. Experiments on MultiBLiMP and Belebele show that our method achieves a strong performance-efficiency trade-off, attaining competitive performance relative to fine-tuning complete final layers, while updating less than 2% of the parameters. Overall, our findings provide insights into where and how language specialization emerges in MoEs during continual pre-training and provide practical insights for low-resource multilingual adaptation. Our code is available at https://github.com/aditi184/moe-routing-adaptation.

Large language models recall knowledge reliably in English but often fail on the same query posed in a lower-resourced language -- a crosslingual consistency gap that remains underexplored for Indian languages and their code-mixed counterparts. To study this gap, we introduce IndiKLAR, an Indic extension of the KLAR-CLC benchmark covering 18 of the 22 scheduled Indian languages and pairing them with code-mixed variants for 11 widely used language pairs, with native-speaker verification of both monolingual and code-mixed variants for these 11 settings. This three-way alignment offers a unique opportunity to examine how knowledge recall consistency varies across the spectrum of English, code-mixed, and native Indian language inputs. Evaluating across nine open-weight models, we find that the native-language accuracy gap to English can reach $\sim$0.50, while code-mixed inputs close most of it -- bringing performance within $\sim$0.05 of English without any model-level intervention. Motivated by this, we evaluate several prompting strategies that vary in how language conversion is exposed, including a two-stage translate-then-answer setup, a one-stage joint translation-and-answer prompt, and Translate-in-Thought (TinT) -- a single-step strategy in which the model converts the input internally and emits only the final answer. Across the performance trajectory native $\rightarrow$ code-mixed $\rightarrow$ English, we identify a consistent flip point -- the boundary between incorrect and correct prediction -- that lies between the native and code-mixed settings. Interestingly, this holds whether the trajectory is induced by the input surface form or by the model's internal conversion process.

This work presents a comparative evaluation of machine translation systems applied to images containing textual information, a task that lies at the intersection of computer vision and natural language processing. The study compares three main paradigms: modular pipelines that separate text detection, recognition, and translation; multi-modal large language models (MLLMs) capable of processing both image and text jointly; and an end-to-end model, Translatotron-V, which directly generates translated images. The modular systems employ state-of-the-art OCR (docTR) combined with multilingual LLMs such as Llama and EuroLLM, while the evaluated MLLMs include different configurations of Gemini 2.5. Experiments were conducted on parallel multilingual datasets covering multiple language pairs, with evaluation based on BLEU, chrF, and TER metrics. The results show that modular pipelines outperform the end-to-end approach, while MLLMs achieve the best overall performance, demonstrating superior flexibility and contextual understanding. These findings underscore the effectiveness of multi-modal reasoning for image-to-text translation and provide a solid foundation for future research on integrating visual understanding and language generation in multilingual settings.

Recent studies have shown that code-switching data (CSD), in which multiple languages are mixed within the same context, can improve cross-lingual transfer and multilingual alignment in large language models (LLMs). However, existing studies primarily focus on bilingual transfer between English and a target language, leaving multilingual settings involving three or more languages largely unexplored. In this work, we investigate multilingual code-switching instruction tuning across four languages: English, Japanese, Korean, and Chinese. We evaluate multilingual understanding on Belebele. Our experiments show that simple sentence-level multilingual CSD consistently improves average multilingual performance across all four languages, indicating that multilingual code-switching can be effective beyond bilingual transfer settings.

As LLMs become increasingly integrated into daily life, understanding how their presence will shape human linguistic behavior is an open question. We present a large-scale study of linguistic convergence in human-LLM dialogue, examining how humans and LLMs accommodate each other's linguistic style during multi-turn conversations. Using an asymmetric convergence metric on WildChat, a corpus of real-world ChatGPT transcripts, we find that while LLMs significantly overconverge toward their users on both function word and open-class features across eight languages, human convergence rates in this setting are broadly consistent with human-human baselines. These findings suggest that accommodation in human-LLM dialogue is asymmetric: while LLMs dramatically overfit to their users' style, humans linguistically accommodate LLMs no differently than they would another person.

In this paper, we introduce LLMBridge, a new LLM based system for the task of end-to-end referential bridging resolution in English. Our bridging resolution pipeline combines heuristic pre/post-processing with the natural language inference ability that comes from LLMs. We evaluate our bridging resolution pipeline on three datasets which have been used for referential bridging resolution evaluation in English: ISNotes, BASHI, and GUMBridge. Comparison to previous bridging resolution systems shows that the performance of LLMBridge surpasses previous state-of-the-art (SoTA) systems for all 3 datasets in the challenging End-to-end Evaluation Setting, as well as the Basic Bridging Resolution Evaluation Setting (gold bridging anaphor given). We also conduct a thorough error analysis of the LLMBridge performance, examining what varieties of bridging remain difficult for LLM based systems to identify. With this paper, we release the code for the LLMBridge pipeline.

Large language models (LLMs) have become increasingly useful computational models of human language processing, but it remains unclear whether vision-language learning makes text representations more human-like during natural reading. Here, we address this question by comparing tightly matched LLM and vision-language model (VLM) pairs under a strictly text-only setting, allowing us to isolate the effect of multimodal training history from online visual input or cross-modal fusion. We evaluate model alignment with a human natural-reading dataset that includes whole-cortex fMRI responses and synchronized eye-tracking saccades. Our findings demonstrate that multimodal pretraining may not confer a uniform, global advantage in human alignment during natural reading, indicating that language-internal representations remain the key factor for modeling human text processing. However, the VLM advantage could emerge more selectively when sentences contain stronger visual semantic content, with converging evidence from both fMRI and eye-movement alignments. Together, our findings provide a controlled in silico framework for testing how visual learning history shapes model-human alignment of language processing, suggesting that multimodal pretraining contributes selectively rather than globally to human-like language representations during natural reading.

Discourse particles, such as well and kind of, are crucial components that enable LLMs to "speak" more like humans. They are used to convey emotions, intentions, and interpersonal attitudes. However, existing studies have not yet built a comprehensive understanding of LLMs' capabilities in handling discourse particles. Moreover, the limited number of research focuses primarily on high-resource languages such as English, with little attention paid to Southeast Asian languages. In this paper, we (1) propose MalayPrag, a benchmark designed to systematically evaluate and analyze LLMs' capabilities in handling discourse particles in colloquial Malay; (2) introduce five attributes that provide a theoretically grounded, unified framework for interpreting pragmatic functions of discourse particles. Applying these two, we prompt ten off-the-shelf LLMs to perform three prediction tasks. The experimental results reveal substantial challenges for current LLMs to accurately connect discourse particles and their pragmatic functions in Malay. The provision of the five attributes designed in this study is found to significantly improve the connections, highlighting the need for structured scaffolding for models' pragmatic competence.

Vision-language models (VLMs) generate fluent causal explanations, but current evaluations cannot distinguish linguistic plausibility from faithful causal reasoning. We introduce a dual-probe methodology that isolates these properties. The Text-Only Probe measures linguistic quality. The Chain-Text Probe requires models to first generate explicit causal chains. The Abstraction Gap (AG) metric quantifies the normalized performance difference. Evaluating eight VLMs on CAGE (Causal Abstraction Gap Evaluation), a benchmark of 49,500 questions across 5,500 images spanning Pearl's causal hierarchy, we find seven models exhibit AG exceeding 0.50 with text scores of 6--8 but chain scores below 2.5. Fine-tuning on 45,000 chain-annotated examples fails to close the gap. However, one model achieves near-zero AG. The capability exists within current VLM architectures and depends on pretraining and architectural choices. CAGE provides a diagnostic tool for assessing faithful causal reasoning in VLMs.

LLMs' linguistically expressed confidence should faithfully reflect their intrinsic uncertainty. While recent work shows LLMs struggle to use epistemic markers (e.g., "it is likely...") in a human-aligned fashion, it remains unclear whether models can apply their own linguistic confidence framework to associate markers with specific confidence levels in a stable and generalizable way, and how contextual features impact this ability. We conduct the first systematic study of this question, formalizing marker internal confidence (MIC) as the estimated intrinsic confidence a model associates with a specific epistemic marker in a given task domain. We present 7 metrics to evaluate the stability of MICs within and across distributions. Applying our analysis framework to diverse models and tasks, we find that LLMs remain faithfully miscalibrated even under model-centric interpretation of marker meanings, struggling to differentiate markers by internal confidence across distributions despite preserving a somewhat consistent ranking order across tasks. This supplies critical, complementary evidence to existing work toward a holistic understanding of faithful calibration in LLMs, emphasizing the need for more aligned and stable marker use to improve trustworthiness and reliability.

Sense representations (explicit, per-token meaning decompositions) are useful for disambiguation, steering, and cross-lingual alignment, but existing approaches require models to be pretrained with sense structure baked in. We introduce ACROS, which induces an explicit sense pathway into a frozen pretrained decoder LM through a gated residual addition. On SmolLM2-360M, ACROS preserves base LM quality while supporting three uses of the same induced variables: zero-shot word-sense disambiguation (64.95 F1 on Raganato ALL, competitive with the WordNet first-sense heuristic), low-KL lexical steering across 5,161 CoInCo cases where a simple non-oracle proxy recovers about 90% of positive shifts, and SENSIA cross-lingual adaptation to four languages (mean R@1 0.988, target FLORES PPL 7.94). ACROS makes sense representations an inducible interface for ordinary pretrained LMs.

ClinicalEncoder26AM is a multilingual Diagnosable ColBERT for clinical and biomedical texts, which aligns at multiple levels its token-level semantic with ClinicalMap25, a clinical latent space inspired by BioLORD-2023 and enriched with synthetic and annotated supervision. The post-training recipe builds upon BGE-M3, and combines synthetic clinical notes, patient--doctor conversations, and annotated resources such as MedMentions, while considering both named-entity-level and sentence-level representations in a multi-adapter distillation, along with a ColBERT-style retrieval objective. In this system demonstration paper, we evaluate the model in the MultiClinNER shared task by finetuning it as a BIO tagger for patient symptoms, disorders, and procedure spans, using a lightweight two-layer CNN head to improve local boundary detection. The resulting system remains simple, processes most documents in a single 8192-token window, and achieves state-of-the-art multilingual entity recall, while achieving Top 5 overall across all entity types and languages in Character-weighted F1 scores. Training curves further show that ClinicalEncoder26AM is markedly more data-efficient than the base M3 model, supporting the usefulness of its clinical post-training for downstream information extraction. The model can be downloaded on https://huggingface.co/Parallia/ClinicalEncoder26AM-Diagnosable-Colbert-L2-for-multilingual-medical-texts

We describe a new sense-tagged corpus for word sense disambiguation. The corpus is constituted of instances of 20 French polysemous verbs. Each verb instance is annotated with three sense labels: (1) the actual translation of the verb in the english version of this instance in a parallel corpus, (2) an entry of the verb in a computational dictionary of French (the Lexicon-Grammar tables) and (3) a fine-grained sense label resulting from the concatenation of the translation and the Lexicon-Grammar entry.

Automatic Speech Recognition (ASR) systems are commonly evaluated using aggregate metrics such as Word Error Rate (WER), which do not capture the linguistic structure of errors. Fine-grained analysis, such as Part-of-Speech (PoS)-wise error characterization, requires accurate alignment between ASR hypotheses and reference transcriptions. However, existing alignment tools are often unreliable for languages written in non-Latin scripts. In this work, we address this gap by proposing a robust, automated, language-agnostic alignment mechanism applicable across ASR architectures and across languages written in both Latin and non-Latin scripts. This enables consistent alignment of hypotheses, references, and evaluation sequences, forming the basis for downstream linguistic analysis. Building on this, we employ standard PoS taggers to perform scalable and reproducible PoS-wise error analysis. Notably, we perform alignment and downstream ASR error analysis across three major segmented writing systems, namely, Abugida (Tamil, Hindi, Kannada), Alphabetic (English, Russian, Greek), and Abjad (Arabic). We further demonstrate how such error information can be leveraged during ASR training to improve metrics such as WER.

Vision-language models (VLMs) are increasingly evaluated for whether they identify the right visual content, but little is known about whether they express such content in a discourse-appropriate form. We address this research gap using information structure (IS), testing whether VLMs distinguish discourse-old Topics from discourse-new Foci in visually grounded question answering. We exploit Hungarian, a language in which Topic and Focus map onto dedicated syntactic positions, making IS choices observable in text. Comparing six VLMs with human participants, we find that models produce IS-relevant constructions, but over-regularise this sensitivity. Under the interacting pressures of discourse status, grammatical role (preference for subject Topics) and definiteness (preference for indefinite Foci), humans choose variable strategies for IS realisation. VLMs, by contrast, collapse onto narrow response templates, resembling mode collapse (Kirk et al., 2024). Our findings suggest that VLM evaluation should look beyond content accuracy to how content is packaged for the discourse.

General-purpose machine translation benchmarks such as FLORES-200 have reached a saturation regime on Chinese-English pairs, where modern large language models cluster within a narrow band of high scores. Across 22 systems, FLORES-200 zh-en GEMBA scores fall in a 7.87-point range with a standard deviation of 2.29, which compresses the separation between systems on knowledge-intensive domains such as finance, healthcare, law, and science and technology. We introduce HardMTBench, a difficulty-aware diagnostic benchmark for bidirectional Chinese-English domain translation. HardMTBench covers 12 domains and contains 10,000 hand-curated source sentences with reference translations, packaged as 20,000 directional test items. A three-stage construction pipeline builds a domain-balanced candidate pool of 84{,}566 pairs, applies an LLM-based multi-signal judge over knowledge density, translation difficulty, terminology load and reference correctness, and assembles the final test set under a hardness fusion rule with per-domain quotas. Across 22 systems spanning general LLMs, commercial engines and specialised MT models, HardMTBench widens the cross-system GEMBA range by roughly a factor of two over FLORES-200, induces visible rank reorderings, and exposes domain-specific terminology and knowledge weaknesses that quality-only metrics tend to flatten. All data and code are open-sourced at https://github.com/jasonNLP/HardMTBench.

Mixture-of-Experts (MoE) models have emerged as a dominant paradigm for efficient LLM scaling, yet adapting them to non-English downstream tasks remains challenging. Existing fine-tuning approaches treat MoE models as monolithic learners, ignoring the heterogeneous routing structure that develops during pretraining. We validate across multiple MoE models and downstream tasks that middle layers form a language-universal alignment zone where routing divergence strongly predicts per-language task performance gaps. Building on this observation, we propose RA-MoE (Routing-Aligned MoE Fine-Tuning), a three-stage framework that categorizes parallel task examples into a four-way taxonomy (cc/ci/ic/ii) based on correctness in English and the target language, identifies task-relevant experts in the middle layers, and augments standard SFT with a routing alignment loss that encourages target-language routing on ci-type examples to follow the English task-expert activation pattern. Experiments across three MoE models, three tasks, and six target languages demonstrate that RA-MoE consistently outperforms standard SFT and strong baselines including Routing Steering and RISE, with the ci proportion of a task-language pair serving as a reliable predictor of alignment benefit.

The preservation of under-resourced languages requires digital tools and resources shaped by and for their speakers. We present the first dedicated ASR resources for Puno Quechua (ISO 639-3: qxp): (1) the largest speech corpus for any single Quechua variety, consisting in 66 hours of recordings for scripted and spontaneous speech (including 36 hours of manually transcribed and validated data), collected via a participatory design campaign; (2) the first systematic ASR benchmark for Puno Quechua, evaluating state-of-the-art models and fine-tuning Whisper-base, wav2vec2-base, and XLS-R-300M, with and without continued pre-training (CPT); (3) an open release of all datasets and fine-tuned models.

Modern translation workflows demand more than semantic equivalence. Users routinely require models to preserve JSON or HTML schemas, honor curated glossaries, disambiguate with provided context, and match prescribed registers, often several at once. Conventional metrics such as BLEU and xCOMET capture semantic fidelity but provide little signal on constraint adherence, while general instruction following benchmarks ignore the cross-lingual nature of translation. We introduce \bench, a benchmark for multilingual translation instruction following covering seven languages, with 4,506 single-constraint and 2,838 multi-constraint items spanning six constraint dimensions and five compositional patterns with instructions issued in all seven languages. Constraints are split into a gating subset verified by deterministic checkers and a continuous subset scored by a rubric-based LLM judge, combined under a multiplicative rule that resists reward hacking. Evaluating 15 models reveals systematic gaps that prior protocols miss: Instruction following scales with size more sharply than translation quality, glossary and structured-format constraints dominate the difficulty gradient, and general instruction following rankings correlate only weakly with translation behavior. Our benchmark are available at https://github.com/Tencent-Hunyuan/Hy-MT2/tree/main/IFMTBench.

Multilingual Large Language Models (mLLMs) leaderboards report per-language accuracy but rarely explain why disparities emerge, leaving systemic biases unattributed and offering practitioners no actionable levers. We first establish that these gaps are systematic rather than artifacts of sampling noise via distribution-free Friedman and Kruskal--Wallis tests, then introduce a two-step Bayesian hierarchical framework that decomposes multilingual performance variance into interpretable components. First, isolating the variance attributable to language identity, we show that observable language features (script, family, typological distance) explain $R^2_{\text{ling}} = 79\%$ of this variance on understanding tasks and $92\%$ on reasoning, with a model's internal representational similarity to English emerging as the dominant predictor across both task buckets. Second, decomposing the full (model$\times$benchmark$\times$language) cube, we find that NLU and reasoning have fundamentally divergent variance profiles: model identity dominates understanding ($66.7\%$ of variance), whereas the benchmark$\times$model interaction dominates reasoning ($46.3\%$). Together these results recast multilingual evaluation from passive performance mapping into an explainable, diagnostic framework with concrete levers for targeting the root drivers of language disparity.

Modern large language models (LLMs) achieve state-of-the-art machine translation performance, but they do so as broad generalists largely trained for many tasks and capabilities unrelated to translation. Thus, they are heavily overparameterized for this task, resulting in excessive memory and compute requirements. In this paper, we present a method for aggressively pruning experts from modern mixture-of-experts LLMs while incurring negligible degradation in translation quality. Our approach exploits expert specialization and the separability of multilingual capabilities in LLMs to identify experts irrelevant to translation. And because of the modular nature of MoEs, these can be easily pruned without any training. Without retraining, we are able to prune half of all experts with negligible degradation and 70% with only minor losses. With a very short SFT, we prune 75% of experts while recovering baseline performance, and in some settings remove nearly 90% while maintaining reasonable translation quality. Overall, our results show that translation requires only a fraction of the LLM, enabling substantial compression of the MoE blocks that contain over 90% of parameters.

Prompt-based personality control is a key technique for designing large language model (LLM) dialogue agents that behave consistently across social contexts. However, specifying Big Five personality traits (BFTs) in a prompt does not ensure that the intended traits are expressed in generated utterances. This paper investigates this mismatch from an interactionist perspective, viewing personality expression as a context-dependent outcome shaped by the interplay between trait specification and situational factors. We analyze how perceived BFT expression in LLM-generated dialogue is influenced by three prompt factors: personality traits, dialogue roles, and expressive styles. Using a factorial design that combines six personality conditions, three roles, and three expressive-style conditions, we generate 1,080 LLM-agent dialogues in each of English and Japanese. We then evaluate the target agent's utterances using an LLM-as-a-judge framework to estimate expressed Big Five traits. The results show that expressed personality is shaped not only by explicit trait specification, but also by dialogue role and expressive style. These effects are trait-specific: dialogue role strongly influences Openness, expressive style substantially shapes Conscientiousness and Agreeableness, and explicit trait specification dominates Neuroticism. Even without explicit personality-trait specification, social and expressive conditions induce distinct personality-like impressions. Cross-linguistic comparisons show broadly similar patterns between English and Japanese dialogues, with noticeable differences only under specific combinations of personality, role, and expressive style. These findings suggest that personality control in LLM agents should be understood not as a direct consequence of trait prompting, but as a context-dependent process involving personality specification, social role, and expressive style.

Existing safety evaluations for large language models overlook whether responses preserve comparable medical information across different user phrasings of the same question. To address this, we introduce the Medical Information Response Audit (MIRA), a bilingual, controlled benchmark that assesses whether LLMs provide comparable medical information across user-side language, register, and health literacy signals. MIRA contains 4,320 prompts built from 60 medically reviewed, low-risk health questions. Across five mainstream LLMs, models answered all medical questions, but responses to low health-literacy signals consistently omitted more key information, provided fewer concrete next steps, and offered less support for independent judgment. We term this pattern Differential Information Dilution (DID). A comparison with 300 real-world health queries provides preliminary evidence of rank-order validity. A knowledge-guided mitigation prompt reduces information dilution for most models, with the largest reductions in underinformative simplification observed for Claude (~8%) and Qwen (~6%). Code and data are available at https://github.com/Rainxu09/MIRA.

Speech language models (SpeechLMs) have achieved substantial progress by extending large language models (LLMs) to the speech modality. However, SpeechLM evaluation remains heavily centered on English, limiting reliable assessment of multilingual speech capabilities. Straightforward benchmark transfer through ASR, translation, normalization, and TTS can corrupt language-specific instructions, answer constraints, and spoken forms; for audio understanding, transferring source-language audio also fails to preserve target-language speaker attributes, accents, and paralinguistic properties. To address these limitations, we propose two human-agent benchmark-construction frameworks: one transfers source-language SpokenQA benchmarks into target-language SpokenQA benchmarks, and the other converts target-language ASR corpora into audio understanding benchmarks using transcriptions and speaker metadata. Using these frameworks, we construct and publicly release three Korean speech benchmarks: KVoiceBench and KOpenAudioBench for Korean SpokenQA, and KMMAU for Korean audio understanding, comprising 12,345 samples in total. We evaluate eight recent SpeechLMs and find that English-Korean performance gaps vary substantially across models and task families, and that SpokenQA and audio understanding rankings diverge, revealing complementary weaknesses invisible to English-only evaluation.

Driven by the wave of large language models, Video-Language Models (VLMs) have become a significant yet challenging technology to bridge the gap between videos and texts. Although previous VLM works have made significant progress, almost all of them implicitly assume that all the texts are predefined by the specific template. In real-world applications, such a strict assumption is impossible to satisfy since 1) predefining all the texts is extremely time-consuming and labor-intensive. 2) these predefined text inputs are too restrictive and user-unfriendly, limiting their applications. It is observed that given a video input, texts with similar semantics but different templates lead to various performances. To this end, in this paper, we propose a novel plug-and-play framework for various VLM-based methods to fully bridge videos and texts. Specifically, we first generate positive and negative texts from the original ones to target specific text components. Then, we propose an attribute-based text reasoning strategy to mine fine-grained textual semantics of generated texts. Finally, we utilize videos as guidance to conduct cross-modal bridging by designing a self-weighted loss. Extensive experiments show that the proposed method can serve as the plug-and-play module to effectively improve the performance of state-of-the-art VLMs.

Chain-of-thought (CoT) monitoring has been proposed as a promising safety mechanism for detecting misaligned behavior in large language models. However, its reliability remains largely unexplored beyond English and across diverse model families. We present the first large-scale evaluation of CoT monitorability across 13 diverse languages and seven frontier model families, comprising 16 models. Using adversarial-hint evaluations that require explicit intermediate computation, together with analysis of internal answer-token probabilities, we consistently find CoT unfaithfulness across languages and hint types, with an average rate of 95.9% across 8B--120B parameter models. We find that frontier models systematically engage in strategic manipulation, including answer-switching, post-hoc rationalization, and procedural exploitation of hints, making external monitors struggle to detect deception. We show that frontier models often commit to the misaligned cue in their latent activations within the first 15% of generation, even when the CoT appears faithful. Surprisingly, these deceptive patterns remain 100% in low-resource languages, revealing fundamental limitations in current CoT-based oversight. Our results reveal that CoT monitoring is fundamentally fragile under linguistic distribution shift, providing a substantially weaker safety signal than what English-only studies suggest. These findings underscore an urgent need to develop robust CoT monitors and to accelerate research into white-box monitoring techniques, especially to improve CoT monitorability in mid- and low-resource languages. Our code is available \href{https://multilingual-cot-monitoring.github.io/}{\textcolor{blue}{here}}.

Despite rapid progress in LLM-based code generation, existing models are predominantly trained on imperative languages, leaving functional programming languages (FPLs) such as Haskell, OCaml, and Scala chronically underexplored, with even frontier models performing substantially worse on FPLs. Fine-tuning is a natural remedy, but our experiments show that per-language fine-tuning fails to capture shared functional abstractions, while merged multi-language fine-tuning introduces cross-language interference. To address this, we introduce FPMoE, a lightweight, open-source code generation model built on a sparse Mixture-of-Experts (MoE) architecture with three language-specific routed experts (one each for Haskell, OCaml, and Scala) and a shared expert that captures cross-language functional patterns such as monadic reasoning and type-directed programming. This design resolves both failure modes simultaneously: dedicated experts eliminate interference, while the shared expert preserves abstractions that per-language models miss. On FPEval, FPMoE substantially outperforms fine-tuned baselines and, with only 3B active parameters, matches the performance of much larger models including DeepSeek-Coder-6.7B, Qwen2.5-Coder-14B-Instruct, and Qwen3-Coder-30B-A3B.

Large reasoning models (LRMs) achieve strong mathematical reasoning performance in English, but remain much less reliable in many low- and medium-resource languages. This gap is often explained as a failure to understand non-English problem statements. We show that this view is incomplete: even when the problem is given in English, controlling the model's reasoning language can substantially reduce accuracy, suggesting that language also affects reasoning execution itself. To study this effect, we introduce DATG, a Directed Acyclic Trace Graph framework that maps reasoning traces to language-independent mathematical anchors and dependencies. This allows us to align target-language traces with reference DAGs and measure whether they cover required mathematical nodes, respect dependency edges, and avoid harmful mathematical actions. Experiments on the Qwen3 series across 12 languages show that non-English reasoning often suffers from reduced anchor coverage and weaker dependency fidelity, especially in low-resource languages. Motivated by this diagnosis, we propose Loop-Retry and Formula-Retry, two simple test-time controls targeting DATG-exposed failure modes, and show that they consistently improve target-language reasoning performance in low-resource languages.

Generative translation systems are cultural technologies because they decide how socially meaningful cues are rendered within culturally specific grammatical systems. We study one concrete notion of successful cultural translation: when an English source explicitly encodes gender, an English-to-Hindi translation should preserve the recoverability of that cue unless the source itself is ambiguous. We evaluate this criterion on a 37,345-instance benchmark spanning twelve categories and show that five systems frequently erase gender through ergative and honorific constructions. We then introduce two mechanism-aware inference-time interventions. The first, the Source-Aware Reranker (SAR), prefers candidates that avoid gender-neutralizing syntax. The second, the Phenomenon-Aware Reranker (PAR), preserves gender through targeted lexical marking even when ergative syntax remains. Across GPT-4o-mini and Sarvam, PAR improves target-subset accuracy from 11.07% to 54.47% and from 15.99% to 49.66%, respectively. Human evaluation shows that PAR increases gender preservation from 10.3% to 81.3%, but reduces mean fluency from 4.36 to 3.37. These findings place the two interventions on a preservation and fluency frontier rather than supporting a single dominant solution, and show how culturally situated generation can require explicit tradeoffs among fidelity, fluency, and stylistic naturalness.

Multilingual LLMs are increasingly used when instruction, source content, and required response languages do not coincide. Existing benchmarks have expanded multilingual instruction-following evaluation, but they rarely isolate these three roles within a fully crossed design. We introduce MTM-Bench, a controlled benchmark for language-conditioned task execution in which each instance is defined by a triplet $(L_{\text{instr}}, L_{\text{content}}, L_{\text{resp}})$. Across English, Spanish, and Chinese, MTM-Bench enumerates all 27 triplets and contains 2{,}430 instances per model across semantic reversal, final-state extraction, and language purity with update realization. We evaluate 20 frontier and open-weight LLMs using decomposed metrics for semantic correctness, target-language adherence, constraint satisfaction, contamination ratio, and joint success, with scoring validated by a targeted human audit. The fully crossed design reveals that degradation is organized by the role a language occupies in the task structure, not merely by mismatch count. The response-language role is the dominant axis of variation, and a single response-slot mismatch accounts for most degradation. The response-only and full-mismatch comparison suggests that mismatch count is not a monotonic predictor of difficulty, with model-level ordering varying across systems. Task families fail through distinct channels, showing that semantic correctness alone does not capture reliable multilingual task execution.

Although Large Language Models (LLMs) demonstrate excellent capabilities and performance for general reasoning tasks within the general public domain, they may face challenges with culturally grounded knowledge within languages with limited digital and textual data. In this paper, we investigate culturally grounded multiple-choice question answering with the BLEnD benchmark, which consists of a multilingual corpus of 30 languages and covers various socio-cultural domains, such as cuisine, sports, family, etc. We propose a region-aware hybrid retrieval approach that combines BM25 lexical matching and dense semantic similarity with regional weighting heuristics to improve the relevance of the answer. The retrieved documents are used to construct a structured prompt for the Qwen3-14B quantized model with logit-based deterministic answer selection. The experimental results show improvements to cross-lingual stability with the hybrid retrieval approach over pure parametric inference for culturally grounded question answering. However, there are still notable performance gaps between languages with more and less training data. This shows that the limitations of the retrieval augmentation approach are not entirely overcome by the training data imbalance problem.

Annotation quality is difficult to sustain when campaigns span weeks or months with small annotator pools. We present a Setswana sentiment dataset of 3,565 tweets annotated by three native-speaker annotators across eight batches and examine why inter-annotator agreement (IAA) declines over time. Despite an aggregate Randolph's free-marginal Kappa of $κ= 0.76$, "excellent," per-batch $κ$ falls by more than 32 points across the annotation task. Through six targeted analyses, we find that (i) label confusion concentrates on the negative/neutral boundary, (ii) two annotators show run-length drift consistent with autopilot labeling, and (iii) the dominant predictor of $κ$ is temporal simultaneity: tweets labeled within one minute achieve $κ= 0.98$, while those labeled more than a day apart reach only $κ= 0.65$. Annotation speed and tweet-level linguistic features show no meaningful association with $κ$. We benchmark three open multilingual encoders and proprietary models (GPT-5 and Gemini) on three-class sentiment classification; fine-tuning yields gains of 29 to 43 macro-F1 points over pretrained baselines, with GPT-5 few-shot leading overall (62.2 macro-F1). We release the dataset, per-annotation timestamps, and analysis code to support reproducible quality auditing for future African language NLP resources.

This paper reports the core linguistic work performed to construct a dictionary-based morphological analyser for Malagasy simple verbs. It uses the Unitex platform and comprised the contruction of an electronic dictionary for Malagasy simple verbs. The data is encoded on the basis of morphological features. The morphological variations of verb stems and their combination with inflectional affixes are formalized in finite-state transducers represented by editable graphs. 78 transducers allow Unitex to generate a dictionary of allomorphs of stems. 271 other transducers are used by the morphological analyser of Unitex to recognize the stem and the affixes in conjugated verbs. The design of the dictionary and transducers prioritizes readability, so that they can be extended and updated by linguists.

We present BhashaSetu, a linguistically enriched English--Marathi parallel dataset addressing persistent data limitations in low-resource neural machine translation (NMT). Marathi, spoken by over 95 million people, remains underrepresented in high-quality parallel corpora across diverse domains. Our dataset comprises 2.78 million sentence pairs from heterogeneous sources including news, politics, healthcare, literature, and culture, with stemmed and lemmatized representations to support morphology-aware analysis. We benchmark multiple state-of-the-art translation models using BLEU, spBLEU, chrF++, and TER metrics, and conduct parameter-efficient fine-tuning of NLLB-200-distilled-600M using LoRA. A key finding from our ablation: corpus-level deduplication is the single largest preprocessing contributor to downstream quality (removing it reduces performance by 1.17 BLEU and 2.21 chrF++), demonstrating that disciplined cross-source corpus hygiene is a low-cost, high-impact intervention for low-resource, morphologically rich languages. The dataset is publicly released to promote reproducible and linguistically informed low-resource NMT research.

Text-to-speech (TTS) evaluation for low-resource non-Latin-script languages can fail when it relies on a single ASR round-trip word error rate (WER). A system may produce no audio, speak a neighbouring language, preserve target script text only in an ASR transcript, or sound unnatural to native listeners. We introduce INSV (Intelligibility, Naturalness, Script fidelity, and Verification), a reporting framework that separates these cases. This paper reports INSV-A, the automated screening subset: synthesis completion, ASR WER/CER, transcript Script Fidelity Rate, and audio language identification. Native MOS and phonetic annotation are specified but not claimed in this release. We instantiate INSV-A as PashtoTTS-Bench, a dated benchmark for Pashto TTS. The April-May 2026 run evaluates Edge GulNawaz, Edge Latifa, OmniVoice clone, OmniVoice auto, and an Urdu negative control on 200 FLEURS and 200 filtered Common Voice 24 prompts. Under the independent omniASR_CTC_300M_v2, OmniVoice auto has the lowest WER (24.1% FLEURS, 27.4% CV24), followed by Edge GulNawaz (32.8%, 39.5%), Edge Latifa (35.6%, 47.7%), and OmniVoice clone (45.4%, 34.8%). WER below the natural-speech baseline reflects clean synthetic audio and should not be read as better than native speech. Whisper Large V3 returns 0.0% Pashto labels on checked Pashto TTS audio, while MMS-LID-4017 and SpeechBrain VoxLingua107 separate Pashto outputs from the Urdu control. The release provides provider metadata, per-sentence scores, LID audits, failure logs, and scripts for adding systems.

Safety evaluation of Large Language Models (LLMs) has largely focused on high-resource languages, leaving low-resource languages critically underserved. We present AlbanianLLMSafety, the first publicly available safety evaluation dataset for LLMs in Albanian, a linguistically distinct low-resource language with approximately 7.5 million speakers across Albania, Kosovo, North Macedonia, and the diaspora. The dataset contains 2,951 prompts spanning 11 safety categories, including self-harm, violence, racist content, child exploitation, and radicalization, with an average of 268 prompts per category. Each prompt is provided in Albanian with an English reference translation and a detailed category label. This resource addresses a significant gap in safety evaluation infrastruc-ture for low-resource languages and provides an essential benchmark for developing safer, more inclusive LLMs. The dataset will be provided upon request to support safety evaluation, fine-tuning, red-teaming, and guardrail development for Albanian-speaking communities.

Recent reasoning Large Language Models produce a chain-of-thought (CoT) predominantly in English, even when prompted in non-English languages. Prior work suggests that forcing the CoT to remain in the input language (\emph{native reasoning}) substantially degrades performance relative to allowing the model to reason in English before answering in the input language (\emph{English-pivoted reasoning}). However, most studies of this native reasoning gap rely on inference-time interventions or limited native-language training data. We revisit this comparison at a larger scale and under comparable supervision. We construct long multilingual reasoning datasets across six languages (English, French, German, Spanish, Chinese and Swahili); fine-tune specialists in both native and English-pivoted regimes on top of \texttt{Qwen/Qwen3-8B-Base}, and evaluate across mathematics, science, general knowledge, and code. In this setting, the average native reasoning gap shrinks to 1.9--3.5% across the five non-English languages, considerably smaller than previously reported. Weight-space analysis of the native specialists reveals aligned fine-tuning updates in the middle layers and divergence in the outer layers. This points to a largely language-agnostic reasoning core surrounded by language-specific layers. Exploiting this structure, we introduce a Layer Swap: transferring the English specialist's stronger reasoning mid-layers into each native specialist, closing most of the native reasoning gap across the five non-English languages while preserving CoT in the target language. We release all models and datasets.

Benchmarks that reflect the diversity and complexity of real-world documents are essential for accurately evaluating Automatic Text Recognition (ATR) systems, especially Vision-Large Language Models (vLLMs). Although recent models demonstrate impressive performance, they are often evaluated on datasets containing modern, printed texts mostly written in English, which limits their relevance to many practical applications. Therefore, selecting a model for a specific use case requires evaluating it on data that matches the target documents. This highlights the importance of representative benchmarks for real-world applications. In this paper, we introduce METATR (v1.0), a multilingual, evolving benchmark designed to evaluate ATR models across a wide range of documents, facilitating meaningful model comparison and selection. The benchmark was designed to maximize diversity by including documents from various public collections. These documents cover 29 languages and include texts with multiple scripts and layouts. Beyond the dataset itself, METATR defines a standardized prompting and normalization methodology and establishes a dynamic evaluation framework. This approach is intended to produce reproducible results while remaining extensible over time. We evaluated a wide range of state-of-the-art systems, including open-source models and closed-source models. Results are reported across various dimensions, including performance at the dataset and language levels, robustness to handwritten documents, and computational efficiency. Our findings show that, although proprietary models achieve the most consistent performance, substantial variability persists across scripts and layouts. Overall, METATR provides a multidimensional, practitioner-oriented framework for assessing multilingual ATR in real-world conditions and tracking progress as the field evolves.

Cross-lingual transfer in language models is difficult to study in natural corpora because lexical overlap, morphology, data imbalance, and tokenization are entangled. We introduce an in-vitro framework with two procedurally generated languages that share the same ontology, typed grammar, and compositional structure, but differ in surface realization. This lets us independently vary lexical distance, minority-language proportion, tokenizer training regime, and vocabulary size, while evaluating transfer on a masked minority-language condition whose lexical forms are never observed during training. Across 700 controlled runs, we find that transfer is governed less by tokenizer balance or raw lexical similarity than by whether tokenization preserves reusable cross-lingual substructure. Smaller vocabularies often improve masked transfer by keeping words decomposable into shared fragments, whereas larger vocabularies can turn forms into language-specific atoms. We further show that transfer emerges as a staged process: grammatical and type-level competence precede masked lexical generalization. Finally, we attempt to explain this mechanism through tokenizer bridges and show that bridge strength correlates strongly with masked reachability.

Vision-language models have progressed rapidly, but Tibetan remains a severely underserved low-resource language due to the lack of reproducible training and evaluation infrastructure. To fill this gap, we introduce FTibSuite, a comprehensive resource suite for Tibetan vision-language research, consisting of FTibData (human-verified multimodal training corpora spanning continual pretraining, image-text alignment, and instruction tuning data), FTibBench (Tibetan adaptations of five mainstream multimodal benchmarks with a hierarchical quality-control workflow to reduce translation noise), and FTibVLM, a reproducible baseline built on Qwen3-VL-8B-Instruct via a three-stage adaptation pipeline. Experiments on FTibBench show FTibVLM delivers consistent performance gains across all tasks, such as improving MMBench accuracy from 42.97 to 67.78 and POPE-random accuracy from 47.53 to 80.56, while retaining the backbone's original Chinese capabilities with minimal degradation, providing the first standardized foundation for Tibetan multimodal research.

Multilingual embedding models are deployed under the assumption that cross-lingual retrieval is symmetric: if a query in language A retrieves its translation in language B, the reverse should also hold. In practice it does not. Using a parallel corpus of 6,518 idiomatic and proverbial expressions in English, Bangla, Hindi, and Arabic, embedded by five production-grade encoders (Gemini, Mistral, OpenAI-L, OpenAI-S, Qwen), we formalise this failure as a deficit in mutual nearest-neighbour reciprocity and test a single mechanistic claim: among the geometric pathologies of multilingual spaces, hubness, not anisotropy, centroid drift, or magnitude, is the dominant causal driver. Across five pre-registered experiments with falsification conditions specified in advance, hub mass dominates a joint regression on reciprocity (49.5% dominance share, 1.68x the next predictor; partial R^2 = 0.302 versus 0.003 for anisotropy), while a hub-aware score correction (CSLS) closes 63.5% of the worst-to-best reciprocity gap and yields a mean within-model effect size 130x larger than surgical hub-vector ablation. The latter contrast pinpoints the mechanism: hubness is a pathology of the similarity metric, not of individual hub vectors. We resolve the well-known anisotropy-hubness paradox by showing the two are statistically dissociable, and we recommend replacing cosine similarity with CSLS as the default retrieval metric for multilingual embedding pipelines.

Large language models (LLMs) learn undesirable properties during pretraining, including dangerous knowledge and toxic text generation. Just as post-training uses different objectives to shape different behaviors, we argue that unlearning methods should be designed for the language function at issue. To study this, we consider two mechanistically distinct unlearning goals, dangerous-knowledge unlearning and toxicity unlearning. For dangerous knowledge, we introduce a cosine-based, meta-learned variant of RMU. For toxicity, we propose a multi-layer objective based on layer-specific probe directions. Across four open-source 7-8B models, our methods achieve strong results, based on distinct training objectives for the two types of unlearning. Overall, our results suggest that unlearning should be studied as a family of problems, analogous to the multiple types of LLM post-training.

Prior work establishes that controlled contrastiveness between self-generated responses from large language models, set via reward scores, improves downstream preference tuning in English. We extend this method to multiple languages and evaluate two models across a total of 14 high and low-resource languages on a diverse set of tasks. Our central finding is that cross-lingual contrastive preference tuning on self-generations (CroCo) transfers without language-specific preference annotation. A reward model trained on English preferences (atop a multilingual base) produces useful within-language rankings across most languages, and pairing in either a monolingual or multilingual setting improves over each model on the majority of setups while preventing the catastrophic forgetting of supervised fine-tuning. We observe that the gains require on-policy data. Off-policy responses reduce the benefit and online preference optimization fails to improve over the offline variant. Specifically, on structured tasks, our method matches or exceeds the base in 6/7 languages for EuroLLM-9B and 4/7 settings for Aya-3B. On open-ended generation, evaluated by two judges, both tuned models win 28/30 times against their respective base across 15 evaluated languages (high and low-resource). Overall, we show promising directions for multilingual preference tuning using self-generations.

Annotating speaker attributes from text is inherently ambiguous, particularly in multilingual settings where demographic and social cues are implicit and culturally variable. We propose a human-large language model (LLM) collaborative re-annotation framework for stabilizing multilingual speaker-attribute labels under practical resource constraints. Starting from a noisy corpus, we use LLMs to surface recurring annotation rationales through iterative interaction with experts, and apply disagreement-focused sampling for targeted re-annotation. Using this framework, we construct WhoSaidIt, a multilingual dataset covering nine speaker-attribute labels. We quantify divergence between original and revised annotations, benchmark recent LLMs, and analyze the effect of explicit rationales on model behavior. Our results reveal substantial cross-lingual differences in annotation decisions and demonstrate both the strengths and limitations of LLMs in speaker-attribute classification.

Automated Essay Scoring (AES) for English proficiency assessment increasingly relies on pretrained transformer models, yet these models are typically trained on general-domain English and may under-represent second-language learner writing. This study investigates whether domain-adaptive continued pretraining (DAPT) on a learner-writing corpus improves transformer-based AES for English proficiency assessment. We perform DAPT on BERT, RoBERTa, and DistilBERT using the EFCAMDAT corpus, then compare the adapted models with their original checkpoints on two English proficiency test datasets, FCE and IELTS, in both in-domain scoring and few-shot cross-dataset transfer. Full-corpus DAPT produces mixed effects across models, datasets, and metrics. Subsequent lexical and syntactic analyses suggest mismatches between EFCAMDAT and the downstream datasets in proficiency level, genre, and communicative purpose. We therefore repeat DAPT using proficiency-specific EFCAMDAT subsets across all three encoder architectures. Proficiency-specific DAPT frequently outperforms full-corpus DAPT and, in some settings, even the non-adapted baseline. Overall, continued pretraining on learner writing can improve in-domain AES, but its benefits depend on both the proficiency composition of the pretraining data and the underlying encoder architecture, and do not consistently extend to cross-test transfer.

Endowing models with consistent multilingual performance can be achieved by mixing pre-training data, or post-training approaches such as language-specific model merging. In this work, we test whether merging can be applied to monolingually pre-trained models. We conduct a controlled study on the efficacy of mixed, merged, and monolingual pre-training setups. We find that while monolingual pre-training results in strong in-language performance, merging any combination of monolingual models leads to performance collapse due to interference. Our analysis suggests representational similarity is a prerequisite for model merging. We therefore conclude that the flexibility of merging in fine-tuning does not extend trivially to language-specific pre-training.

Social media platforms enable large-scale cross-lingual communication, but translating user-generated content (UGC) remains challenging due to its informal style, cultural references, and interaction-based expressions. While recent LLMs have improved translation quality, existing benchmarks and metrics often fail to capture whether translations convey intended meaning and cultural resonance in real-world settings. In this work, we introduce CULTURE-MT, a benchmark for social media translation that focuses on both CULtural Transmission and UGC-specific emotion REsonance. CULTURE-MT consists of 1,002 UGC notes across 14 domains, categorized into four types based on culture-loaded symbols and linguistic style features. We also construct UGC-oriented training data to fine-tune Qwen3-8B and Qwen3-32B as baselines. We propose cultural effectiveness as a new evaluation criterion, focusing on expression accuracy and cultural adaptability. Testing 15 models, including the baselines, we find that traditional metrics fail to capture cultural effectiveness. We also observe that cultural effectiveness on base LLMs correlates with model size. Our work provides a comprehensive evaluation system for UGC translation models and will offer an open evaluation platform to advance research in this area. We release the CULTURE-MT benchmark and provide an online leaderboard where submitted translation results can be evaluated by our trained JUDGER.

Alzheimer's disease (AD) is a progressive neurodegenerative disorder and the leading cause of dementia, affecting memory, reasoning, communication, and daily functioning. Early diagnosis is particularly important, as timely intervention may help slow cognitive decline and improve patient care. Recent studies have demonstrated that spontaneous speech contains valuable linguistic and acoustic biomarkers associated with dementia. However, existing approaches often rely on independently trained modality-specific models, feature concatenation strategies, ensemble methods, or attention-based fusion mechanisms that do not explicitly maximize the dependency between speech and transcript representations. In this work, we propose a multimodal deep learning framework for automatic dementia detection that jointly exploits speech and transcript information in an end-to-end trainable manner. Specifically, speech recordings are divided into 10-second segments and passed through a pre-trained HuBERT model to extract contextualized acoustic representations. To better capture informative temporal speech characteristics, attentive statistics pooling is employed to aggregate frame-level acoustic embeddings. For the textual modality, transcripts are encoded using a pre-trained BERT model, where the [CLS] token representation is used as the linguistic embedding. The acoustic and textual representations are subsequently combined using an attention-based Audio-Text Fusion (AT-Fusion) mechanism. In addition, we introduce a MINE objective to maximize the mutual information between modalities and improve multimodal representation alignment. The fused multimodal representation is finally used for dementia classification. Experiments conducted on the publicly available ADReSS Challenge and PROCESS-2 dataset demonstrate the effectiveness and robustness of the proposed approach for speech-based dementia assessment.

Large language model safety evaluation remains heavily English-centered, leaving low-resource languages under-measured even when models are deployed globally. We evaluate four open-weight instruction-tuned models on SomaliBench v0, a native-author-verified benchmark of 100 harmful-intent prompts paired across English and Somali. Each of Llama-3.1-8B-Instruct, Gemma-2-9B-Instruct, Qwen-2.5-7B-Instruct, and Aya-23-8B is run locally with temperature 0 and the same English "helpful, harmless, and honest" (HHH) system prompt. A pinned Claude Sonnet snapshot (claude-sonnet-4-5-20250929) classifies each response as refused, complied, or unclear; the native author spot-checks a stratified 80-row sample. We find large English-to-Somali refusal gaps for all four models: Llama-3.1-8B (0.90; 95% bootstrap CI [0.85, 0.96]), Aya-23-8B (0.75 [0.67, 0.83]), Qwen-2.5-7B (0.69 [0.59, 0.78]), and Gemma-2-9B (0.38 [0.27, 0.49]). For three models, the dominant Somali non-refusal mode is not fluent harmful compliance but unclear output: empty, wrong-language, or incoherent generations. The native verification spot-check achieves 100% agreement with the judge (Cohen's kappa = 1.00) on the 80 sampled rows. We report aggregate refusal rates, category gaps, and reliability statistics only; raw model generations are retained locally and are not released.

Large language models~(LLMs) are trained on heterogeneous multilingual corpora, yet existing policy optimization methods often implicitly restrict each training question to a single response language or rely on a fixed dominant language for supervision. We propose language-routed policy optimization (LRPO), an online policy optimization framework that treats language as a selectable variable. LRPO elicits multilingual rollouts for each training question and integrates their relative quality into preference-based policy updates, increasing the diversity and informativeness of training signals under the fixed rollout budget. To adaptively determine which languages to explore during reinforcement learning, we introduce a trainable language router formulated as a multi-armed bandit, balancing exploration of underutilized languages with exploitation of more informative ones. Extensive experiments show that LRPO consistently improves multilingual performance, demonstrating that adaptive language routing enables effective cross-lingual knowledge exploitation for training. We release all the resources at https://github.com/Guochry/LRPO.

AI-associated lexical shifts have been documented mainly in Scientific English. We extend this work to 34 languages in the WMT News Crawl corpus, refining a split-halves continuation diagnostic that compares GPT-4.1 continuations with matched human gold-standard text. For each language, we derive ranked AI-overused lemmas using log prevalence ratios. We find substantial cross-lingual semantic convergence: semantically related concepts recur across typologically diverse languages, with 'emphasize'-type verbs appearing in 24 of 34 languages. Embedding-based and manual analyses support this pattern. We also examine diachronic uptake in news writing before and after ChatGPT's release. Tracking each language's top 20 AI-overused items, we find prevalence increases in 26 of 34 languages from 2020-2021 to 2023-2024, with a mean change of +15.1%, whilst matched baseline words show no comparable increase (-4.5%). In 10 languages with longer historical coverage, longitudinal analyses show post-2022 increases that exceed the modest shifts observed in earlier periods, though with smaller effect sizes than in Scientific English. We validate our approach extensively, including across seeds, model variants, data sizes, model families, and more. Our findings are consistent with the view that AI-associated lexical preferences extend beyond English and may exert cross-lingual homogenising pressure on global language use.

Current language modeling approaches are built around tokens. Text corpora are split into tokens, and models are trained by performing computations on these tokens, such as predicting the next token given the preceding ones as context. This paradigm has become the standard in modern language modeling, especially given the outstanding performance obtained by token-based architectures. However, recent works have not only begun to question how language models process and understand meaning from tokens, but also to question whether using higher levels of granularity could advance the research field. This led to the idea of Concept Modeling, that is, to directly train models for next-concept prediction rather than next-token prediction. In this work, we introduce Mimir, a 1.6B Large Concept Model trained for multilingual concept understanding and generation. We leverage a large-scale multilingual pre-training corpus (38,883,987,240 sentences) spanning 46 languages and a large-scale multi-turn, and multilingual instruction-tuning dataset (66,816,428 sentences) covering a total of 35 languages. We extensively evaluate model performance against a language model with a comparable number of parameters.

Background: Fine-tuning is central to adapting pre-trained Large Language Models (LLMs) to downstream tasks, but its reliance on training data, parameter updates, and reusable components opens entry points for attackers. Threats have evolved from data poisoning and weight tampering to agent manipulation and interface exploitation, yet existing reviews lack a unified framework spanning the full fine-tuning lifecycle. Objective: This paper presents a systematic survey of LLM fine-tuning security and establishes a lifecycle-based framework for comparing attacks and defenses, complemented by unified empirical evaluation. Methods: We divide attack and defense mechanisms into three phases by intervention timing: pre-tuning, during-tuning, and post-tuning. Within each phase, strategies are reviewed and contrasted to expose their evolution and limitations. Representative methods are then evaluated under a unified model, hardware, and protocol setup, with cross-phase experiments pairing attacks and defenses from different phases. Results: Attack effectiveness is highly model-dependent and non-monotonic with scale: weight-editing attacks effective on earlier models lose impact on modern open-source LLMs; cross-lingual backdoor transfer, reported as near-perfect at larger scales, fails entirely on tested 1B-4B models; and purely benign samples can compromise safety alignment in instruction-tuned models. Single-phase defenses rarely generalize across phases, and defense effectiveness depends jointly on model architecture and alignment state. Conclusion: We identify key open problems (configuration-robust defense, cross-phase defense composition, and embedding-space attacks beyond behavioral assumptions) and propose concrete future research directions.

Hallucinations in Large Language Models (LLMs) represent a critical barrier to their reliable deployment, a vulnerability heavily exacerbated in non-English and resource-constrained contexts. Existing detection approaches that rely on output confidence heuristics or single-layer internal representations frequently fail to capture deep, complex factual inconsistencies across diverse languages. To address this, we introduce MultiHaluDet, a novel three-stage stacking framework that detects multilingual hallucinations by probing the full hidden state trajectories of frozen LLMs without requiring language-specific fine-tuning. Our method extracts sequential features across multiple layers and processes them via a hybrid architecture using multi-scale attention and self-attention pooling. By generating out-of-fold embeddings that feed into a calibrated classical classifier ensemble, MultiHaluDet captures both fine-grained and coarse-grained patterns of factual inconsistency. Extensive experiments demonstrate that our framework achieves state-of-the-art detection performance, reaching up to 98.55% AUROC on the English HaluEval and TriviaQA benchmarks using Mistral-7B and LLaMA2-7B architectures. Crucially, we rigorously evaluate our framework's cross-lingual generalization across high (French), medium (Bangla), and low-resource (Amharic) languages. MultiHaluDet demonstrates exceptional representational robustness, consistently outperforming baselines and successfully transferring hallucination detection capabilities across typologically diverse linguistic tiers.

Machine-translated benchmarks are widely used to assess the multilingual capabilities of large language models (LLMs), yet translation errors in these benchmarks remain underexplored, raising concerns about the reliability and comparability of multilingual evaluation. We address two practical gaps: (i) how well automatic MQM-style error spans from LLM judges and a span-aware QE baseline (xCOMET-XXL) match expert human span annotations on benchmark translations, and (ii) how strongly translation errors (as opposed to source-side issues in the English original) explain accuracy drops on translated benchmarks. We find that span agreement is non-trivial on naturally occurring benchmark translations, and that target-side translation errors are consistently associated with measurable, percentage-point drops in translated accuracy even after controlling for English correctness and source-side anomalies.