Linguistics

Modern LLMs demonstrate impressive multilingual performance, yet standard benchmarks primarily reward selecting correct answers rather than evaluating genuine factual understanding. We introduce Systematic Wikidata-based Object-Relation Distortion (SWORD), a benchmark that evaluates whether models consistently reject factual errors across languages. SWORD generates syntactically well-formed but factually incorrect statements in eight widely spoken languages through controlled perturbations of Wikidata triples, ranging from random entity substitutions to semantically plausible property-based selections. Our distortion-based evaluation surfaces two critical insights that remain entirely obscured by conventional benchmarks. First, models counterintuitively achieve higher accuracy on semantically plausible distortions than on nonsensical random substitutions, suggesting reliance on distributional familiarity rather than genuine factual verification. Second, models exhibiting comparable baseline accuracy across languages show substantial performance degradation specifically on (East) Asian languages when presented with distorted statements, with cross-lingual performance gaps reaching up to 28 percentage points (49% relative reduction) in some models. These findings demonstrate that multilingual factual reasoning involves asymmetric capabilities that aggregate accuracy metrics systematically obscure.

Tracking semantic change in low-resource languages across extensive historical timelines presents significant challenges due to data scarcity and the limitations of static embedding alignments. This study investigates the diachronic evolution of the Sinhala language from the 13th to the 20th century using a multi-stage computational framework. We first align century-specific Word2Vec and FastText embeddings using Similarity Matrix Based Alignment (SMA) and Orthogonal Procrustes (OP) techniques, finding that OP alignment provides more stable neighbourhood tracking for identifying temporal similarity dips. To move beyond aggregate measures, we introduce a Bidirectional Semantic Impact Pruning approach using contextualised embeddings from a fine-tuned Llama-3.1-8B. By applying Leave-One-Out (LOO) diagnostics, we attempt to isolate influential sentences to distinguish between systemic semantic shifts and transient polysemic expansion. Our results show that semantic drift in the fine-tuned Llama-3.1-8B is not evenly distributed across all usages. Instead, a significant part of the change is driven by a smaller set of high-impact contextual instances, rather than gradual and uniform change across all occurrences. This work provides a preliminary framework for diachronic analysis in low-resource contexts, highlighting the trade-offs between model sensitivity and data availability.

As Large Language Models (LLMs) are increasingly integrated into human society, aligning them with pluralistic social values has become a critical priority. However, whether LLMs exhibit consistent value preferences across languages remains underexplored, particularly for culturally grounded values, which are more abstract and difficult to evaluate and align than safety-centric principles. We investigate this issue through Chinese Social Values (CSV), a value system rooted in Chinese culture and comprising $12$ dimensions across national, societal, and personal levels. We construct C-Voices, the first comprehensive multilingual contrastive probe dataset for CSV, with 86,400 dilemma-based instances in six languages, each pairing a CSV-aligned action with a value-conflicting alternative. Building on the contrastive probes of C-Voices, we then propose a fine-tuning-free value vector steering method that derives value directions from hidden-state discrepancies and selectively intervenes on value-sensitive layers during inference. Experiments on six languages show that CSV-oriented preferences are model-dependent and language-sensitive, with the same dilemma eliciting divergent responses across languages. Our method achieves effective CSV steering, supports cross-lingual transfer of value vectors, and generalizes to existing FLAMES and ValuePrism.

This study examines the compositionality of steering vectors for language and behavioral control in large language models. Focusing on language, jailbreak, and conciseness, we investigate whether additive, training-free composition of attribute steering vectors can preserve the intended steering effect of each attribute, across four instruction-tuned models from two model families and two size scales. We find that single-attribute steering is reliable for all three attributes, but only within an appropriate combination of intervention layer and steering strength, with abstract behaviors (jailbreak, conciseness) favoring middle layers and language favoring earlier layers. We show that additive composition of two attribute vectors succeeds in steering both attributes simultaneously when each is injected at its own best-performing layer, and that this partially extends to three simultaneously composed attributes, addressing an inconsistency left open by prior work on training-free composition. We further analyze the geometric properties of these steering vectors, finding that they are approximately orthogonal in the residual stream, consistent with their compositional behavior.

Aligned language models fail under two independent pressures: the structural jailbreak class recently formalized as Involuntary In-Context Learning (IICL), which reframes a harmful request as the final missing cell of a data-labeling task completed by pattern rather than judged as content; and the erosion of safety alignment outside English. A natural hypothesis is that these compound. We test it directly. Using a deterministic IICL operator and a StrongREJECT-style rubric judge, we red-team two Google Gemini models on two benchmarks, a 30 general-harm behaviours from HarmBench and 30 financial-abuse behaviours from FinProof, each under a single-shot baseline and under IICL in four languages (English, Spanish, Hindi, Arabic). First, IICL generalizes to a second provider and is worse in finance: it lifts attack success from <=6.7% to 80-90% on HarmBench and 97-100% on FinProof, an order of magnitude above the <=24% its introducing study reported on OpenAI's GPT-5.4. Second, against the hypothesis, forcing the IICL output into a non-English language does not stack the two weaknesses, it attenuates the attack. Eleven of twelve non-English conditions score below their English baseline (sign test, p~0.003), the lone exception a ceiling tie near 100%; on the stronger model's financial set Arabic collapses from 100% to 33%. We attribute this to a relevance curse: once structure has unlocked compliance, the models produce lower-quality harmful content in lower-resource languages, which a substance-grading judge scores as partial. The pattern replicates under an independent non-Google judge (Cohen's kappa=0.86, 377 paired verdicts), and 76.6% of non-English responses were verified in-language. Jailbreak vulnerabilities are therefore not additive; the dominant residual risk is the English structural attack, most acute for financial abuse, not a multilingual one.

Multilingual LLMs show stereotype-related behavior that varies across languages, but behavioral scores do not show where the relevant information is represented or how it affects the output. To investigate these internal mechanisms, we compare linear probing, attribution patching, sparse autoencoders (SAEs) and feature ablation in Llama-3.1-8B, Qwen3-8B, and Gemma-2-9B. Probe performance peaks substantially earlier than attribution in all three models, with a separation of 36-53% of model depth. Retained Llama-Scope features often match the social category on which they were selected and form recurring semantic families, but their lexical alignment and ablation effects vary across SAE suites. Only 6-18% of evaluated residual-stream features have language-agnostic effects under our criterion, and none are category-agnostic. Language-agnostic features have larger mean ablation effects in Llama-Scope, but this pattern does not repeat in the other SAE suites. Decodability, output influence, and cross-lingual ablation effects therefore need to be measured separately.

Linguistic typology relies on expert analysis of reference grammars across languages, making large-scale crosslinguistic comparison labor-intensive and unscalable. We introduce AUTOTYPOLOGIST, an LLM agent for evidence-grounded typological analysis over reference grammars. The agent is capable of retrieving relevant grammar sections, analyzing interlinear glossed text (IGT), and iteratively reasoning over typological hypotheses using a ReAct-style workflow. We evaluate the system on TYPOLOGICAL FEATURE CODING against expert annotations and TYPOLOGICAL HYPOTHESIS TESTING with typological universals using 25 open-source reference grammars. Operating under different information constraints in TYPOLOGICAL FEATURE CODING, the agent can synthesize information from reference grammar prose but still faces challenges with only IGTs in the target language. In TYPOLOGICAL HYPOTHESIS TESTING, the agent can synthesize crosslinguistic evidence and identify both supporting cases and counterexamples. These findings suggest that LLM agents can support scalable and inspectable typological analysis, while still requiring expert validation.

Large language models (LLMs) are commonly associated with the distributional hypothesis, according to which (1) semantic meaning is grounded in distributional patterns of linguistic context, and (2) knowledge of cross-linguistic distributional correspondences allows for successful translation. This paper rejects the first claim as a causal inversion: linguistic distributions reflect patterns arising from meaning-making practices rather than constituting their source. At the same time, it accepts the second claim, arguing that translation -human or machine - can succeed without requiring access to meaning or reference. Knowledge of interlingual distributional correspondence and their inferential organization may be sufficient for translation. The paper develops an ecological-enactivist perspective, according to which reference and meaning are grounded in agent-environment interaction and stabilized through action-grounded concepts, forms of world-involving cognition that current LLMs do not possess.

Should multilingual LLMs answer medical questions consistently across input languages, or adapt responses to cultural cues? Existing multilingual medical benchmarks usually assume that medically correct answers should remain consistent across languages and treat cross-lingual variation as model error. In contrast, cultural adaptation research argues that appropriate medical answers may legitimately differ across contexts. We review the multilingual medical NLP literature through these two perspectives, we identify three gaps: limited stakeholder perspectives (e.g., of medical professionals), a lack of empirical evidence on which approach better serves users, and no benchmarks capable of distinguishing universally correct from culture-specific cases. To address the first gap, we survey 356 participants across three stakeholder groups (medical, NLP, and anthropology professionals) in three countries (Germany, Spain, and the United States). Anthropologists consistently favor adaptation, while medical and NLP respondents remain divided, with notable divergence between U.S. and European medical professionals. LLMs prompted with profession and country personas fail to reproduce this variation, overestimating cross-lingual consistency preference among NLP and medical personas. We conclude that neither consistency nor adaptation can currently be considered clearly preferable, highlighting the need for empirical evidence on which approach better serves users across cultural contexts.

Robot foundation models are trained and evaluated predominantly in English, and robot demonstration corpora do not exist for most languages. We study the addition of Greek to an open vision-language-action stack using only machine-rephrased instructions and no architecture changes. The main challenge is measurement rather than translation. Several plausible instruments produce false conclusions: a color-histogram metric rewards noise, a single-goal benchmark scores 84.6% under correct Greek and 82.6% under deliberately wrong instructions, training loss fails to predict Greek success, and single-run comparisons are dominated by seed variation. On a discriminative ninety-task suite with three seeds per arm, a multilingual text tower without Greek demonstrations remains at its wrong-instruction floor, while Greek-only training exceeds its control by at most 2.7 points. Bilingual training yields a consistent 6.7-7.1 point margin over its control and reaches about two fifths of English performance. The policy also overfits the translator's phrasing; training on seven phrasings per task approximately halves this penalty. Warm-starting from a language-adapted world model and unfreezing the text tower both degrade performance. The results support two practical requirements for low-resource robot-policy localization: build a guaranteed null before trusting a metric, and replicate low-resource-language results across seeds.

Minimal-pair benchmarks such as BLiMP evaluate linguistic knowledge by testing whether language models (LMs) prefer acceptable sentences over minimally different unacceptable ones. However, these benchmarks largely ignore lexical frequency variation, despite lexical frequency being a pervasive and highly skewed property of natural language use. Consequently, existing evaluations do not test whether grammatical preferences remain stable when contrasts involve rare lexical items. We introduce FreqBLiMP, a frequency-controlled extension of BLiMP that regenerates all 67 paradigms under explicit Zipf-frequency regimes while preserving each minimal-pair's grammatical contrast. Evaluating multiple open-weight LLM families across scales, we find that decreasing lexical frequency produces a consistent, monotonic decrease in sentence likelihood, but only a modest reduction in overall contrastive acceptability accuracy. However, this aggregate stability masks substantial variation across linguistic phenomena, with LLMs remaining robust on overt morphosyntactic generalization while degrading on phenomena that require lemma-specific information.

Humans can learn and generalize novel concepts from sparse data because they express knowledge in rich structural formats. In this paper, we propose that programs are a strong candidate for universal representation of concepts. We review computational models of concept learning that use programs as their concept representation and evaluate their contribution toward a universal representational language.

Given various grammatical options, language users prefer the word order choice that reduces the overall length of syntactic dependencies, a principle known as dependency length minimization (DLM). The origins of this preference remain an open question, particularly whether it originates from constraints on efficient information processing. Computational simulations provide a powerful approach to identifying the factors influencing the emergence of linguistic phenomena. However, previous simulations of DLM have not examined realistic interaction contexts and have produced mixed results. The present study investigates the emergence of DLM in artificial languages using a recently proposed language learning and communication framework based on recurrent neural networks (RNNs). In this framework, agents are trained to speak and interpret artificial languages and then use these languages to communicate. Using this framework, we study the impact of several factors related to processing limitations in a communicative setting, such as noise during listening, limited speaker capacity, and incremental sentence processing. Our results reveal a complex interplay among these factors in shaping word order preferences in neural agents. Specifically, in the full meaning space, agents regularize toward a single dominant word order, while in the half meaning space they show a short-before-long preference that only aligns with DLM in verb-initial languages. A consistent DLM preference emerges only when agents are subject to incremental processing pressure. These findings suggest that limitations in human cognitive processing may indeed play a role in shaping DLM. Our findings provide insights into the conditions under which neural models replicate human-like preferences and highlight the challenges of designing emergent communication models that capture human cognitive biases in language processing.

Multilingual language models often produce inconsistent answers to semantically equivalent questions across languages, motivating methods to improve cross-lingual consistency (CLC). However, existing methods are typically evaluated using different models, tasks, and protocols, leaving their relative strengths unclear. In this work, we present a unified evaluation of representative CLC-enhancement methods for question answering, spanning inference-time interventions and post-training approaches across three model families and three closed-form benchmarks. The results show that post-training methods are generally more reliable, with direct distribution alignment consistently improving CLC across all model-dataset combinations, while other methods are more sensitive to answer format and the breadth of language coverage. Notably, cross-domain transfer is limited unless source and target tasks share similar output formats. We further investigate whether CLC enhancement hurts models' ability to respond differently when needed, that is, when asked culture-dependent questions. Across two benchmarks of culturally diverse question answering, we find no systematic degradation in controlled closed-form evaluation, whereas open-ended generation reveals occasional accuracy reductions, particularly for non-English responses. Our work highlights the need to evaluate CLC enhancement for both cross-domain robustness and culturally appropriate variation, informing future work in post-training and benchmark development.

Research and news coverage of language-model deception increasingly attributes human-like mental-state concepts to language models. Such claims can blur the distinction between behavior that looks deceptive and a mechanism that is actually deceptive. We introduce a causal taxonomy separating prior commitment from retrospective report, model preference from realized output, false preference from sensitivity to the utility of misleading a recipient, and deceptive behavior from the provenance of the objective or strategy producing it. We test these distinctions in two open-weight model families. Across controlled guessing-game and stock-trading experiments, we find that deceptive-looking behavior can arise without the corresponding proposed mechanism, while other interventions provide direct evidence that recipient information state can causally affect deceptive preference. These results show that deceptive behavior can provide evidence for a deceptive mechanism. But even evidence for such a mechanism does not establish model agency in the deception.

Typological features are widely used in multilingual NLP, and the prediction of such features holds downstream utility. However, existing methods to predict missing values lack interpretable justifications for predictions, while their performance across resource levels and feature types remains underexplored. Given LLMs' abilities in meta-linguistic reasoning and in providing rationales, we investigate LLMs' performance in typological feature prediction via an in-context learning approach with linguistic data from URIEL+ and Glottolog. We find that zero-shot prompting is insufficient, but when given phylogenetic and geographic neighbour evidence, LLMs substantially outperform all baselines without disadvantaging low-resource languages. We further find that most LLM rationales are consistent with the provided evidence, offering a step toward explainable typological feature prediction.

BLEU-4 is the standard metric for evaluating sign language translation (SLT), but spoken-language metrics may not adequately reflect sign language proficiency. The multimodal, low-resource context of SLT allows models to exploit spurious correlations and spoken-language priors, rather than learning stronger sign representations. In this paper, we evaluate the relationship between spatio-temporal understanding and BLEU-4 across six SLT models on Phoenix-2014T and CSL-Daily, showing that gains in BLEU-4 are not on their own evidence of better sign language understanding. This work introduces an alternative inspired by language-learning assessment, using an open-weight-LLM QA protocol that measures salient content preservation. It aligns more closely with human rankings and is six to seven times more paraphrase-invariant than BLEU-4. Applied to SLT, this protocol targets content transfer, is more robust to train-test overlap, and gives a different picture of the field: the five gloss-free systems are largely within noise of one another on Phoenix-2014T, while the gloss-supervised system stands 9.3 points higher, a gap invisible to BLEU-4.

Language models are commonly discussed as technical artefacts, but they are obviously shaped by the linguistic worlds conveyed by data during their training. Using Italian language models as evidence, I want to bring attention to the nature of the systems which result from training and specialising models on translated and synthetic data, and further curating them, and to the meaning of testing them on equally unnatural data. Are these eventually models of Italian? Are they models of language? Does NLP still care about language? These questions yield another, more concrete question: what language do we actually want language models to produce? I argue that this question cannot be answered if we do not first consider a clearer distinction between language models designed as technical products and language models designed as tools for studying language itself. The answers then might be diverse, the languages we are talking about might be diverse, and the picture might not be as pessimistic as we fear.

Large Language Models (LLMs) demonstrate strong multilingual reasoning performance, yet their robustness to semantics-preserving structural variation remains underexplored, particularly for relatively free word-order languages. We investigate the structural sensitivity of multilingual LLMs using two linguistically grounded perturbation settings in Hindi and Malayalam: constrained constituent reordering and active-passive voice transformation. We introduce a benchmark dataset IndicReStruct, with two variants, GSM8K-Reordered and GSM8K-Voice, constructed from GSM8K while preserving semantic meaning. Across six state-of-the-art LLMs and multiple prompting strategies, we observe consistent and significant degradation in mathematical reasoning performance under structurally perturbed inputs. To further understand these failures, we perform qualitative error analysis and mechanistic interpretability experiments using residual-stream activation patching. Our analyses show that reasoning failures frequently arise from disruptions in entity-quantity alignment and that intermediate transformer layers contribute most strongly toward reasoning restoration. Overall, our findings suggest that current multilingual LLMs remain highly sensitive to surface syntactic realization and lack robust compositional invariance under structurally different but semantically equivalent inputs.

Idiomatic expressions are an integral part of natural language, reflecting cultural nuances and posing unique challenges for computational models, particularly in low-resource languages. In this paper, we present the first large-scale benchmark dataset of Bangla idioms, complemented by a synthetic multiple-choice question (MCQ) dataset for idiom meaning identification. We conduct a comprehensive evaluation of recent large language models (LLMs) across three idiom-related tasks: paraphrasing, idiom span detection, and meaning identification, leveraging zero-shot and few-shot prompting strategies. Our results reveal substantial variability in model performance, with no single LLM consistently outperforming others across all tasks. Notably, Phi-4-mini-instruct excels in paraphrasing, Kimi-K2-32b-instruct in span detection, and Gemini-2.5-flash in meaning identification. We believe that our datasets and analyses will provide valuable resources to guide future research in improving LLM comprehension of idiomatic expressions, particularly in Bangla and other low-resource languages.

Pretraining data for Armenian, a morphologically rich and low-resource language, is scarce, and no open Armenian LLM has been released with the data and recipe needed to reproduce it. To address this gap, we curate and release two datasets. ArmWeb is an extensively validated corpus of 4.37M Armenian news documents. ArmSTEM is a parallel English-Armenian collection of 373K math and science problems with step-by-step solutions, translated into Armenian and verified through both answer-preserving LLM judgment and human evaluation. Continued pretraining of Gemma-4-E4B on these datasets yields arm-gemma-e4b, which outperforms every existing open Armenian model as well as its unadapted base, and is the first open Armenian LLM with complete training data and recipe. Our ablations show that news-only continued pretraining improves fluency while eroding knowledge, a pattern we also observe in existing Armenian models, and that a small share of verified translated STEM data reverses the loss. We further find that the largest public Armenian corpora overlap web-derived evaluation panels heavily, including a train/test self-overlap inside FineWeb-2. We openly release all data, models, and code.

Tamil spell and grammar correction is challenging because Tamil is an agglutinative low-resource language with rich verbal morphology, complex sandhi (phonetic transformation) rules at word boundaries, and a script of 247 distinct letters. Prior work targets word-level surface errors with rule-based methods, statistical n-gram models, Minimum Edit Distance, or hybrid pipelines with a transformer re-ranker; such methods cannot reliably handle contextual errors - subject-verb agreement, tense consistency, or cross-word sandhi - which require sentence-level understanding. We propose an end-to-end sequence-to-sequence formulation and fine-tune mT5-small and mBART-50 on a synthetic corpus of up to 657,720 noisy-clean Tamil sentence pairs spanning ten error categories. Both backbones follow the same four-stage progressive schedule, each stage targeting one weakness: surface noise (v2), contextual grammar (v3), single-site sandhi (v4), and multi-site cross-word sandhi (v5). On a 1,000-sentence balanced diagnostic set verified disjoint from all training data, our best model, mBART-50 v5, reaches 69.3% top-1 exact-match accuracy, with 87.5% on sandhi and 43.5% on subject-verb agreement. The schedule is what produces these gains: subject-verb accuracy rises from 1.0% to 52.5% once contextual pairs are introduced, and sandhi from 0% to 87.5% once multi-site sandhi pairs are. We additionally quantify a precision-recall trade-off this literature has not reported: sandhi recall is paid for monotonically in identity accuracy. Finally, Tamil-LLaMA-7B-Instruct reaches 19.0% zero-shot and 24.7% with three demonstrations against a 20.0% copy baseline, showing that a Tamil-adapted instruction model does not transfer to specialised sentence-level correction without task-specific supervision.

Nastase et al. (2026) argue that large language models (LLMs) may illuminate language processing because both rely on distributed, context-sensitive representations shaped by statistical learning. Their rejection of simple cortical "boxology" is persuasive, and they articulate a strong case for the value of LLM-brain alignment research. The key question is what kind of inference LLM-brain alignment licenses. My claim here will be narrow: representational alignment can in principle constrain mechanistic hypotheses, but it does not by itself identify a mechanism. Nastase et al. acknowledge that an encoding model can capture features represented in neural activity without establishing a shared architecture or algorithm. Yet the authors sometime move from alignment to "shared computational principles" and ultimately to LLMs as mechanistic models of natural language. Indeed, their methodological caveat that alignment does not establish a shared architecture or algorithm sits uneasily with their conclusion that LLMs might instantiate the same computational principles as biological brains and provide a "fully mechanistic model" of language. I discuss what I consider to be problems of logical, causal, and computational underdetermination in Nastase et al.'s (2026) proposal.

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

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

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

The growing rate at which LLM agents interact with one another raises key questions about language evolution in multi-LLM-agent settings, with implications for safety and monitorability as well as for linguistic accounts of LLMs. To address these questions, we introduce GlossoGen, a novel platform for studying multi-agent language evolution in complex scenarios. Within GlossoGen, we build the SaveVeyru scenario, which requires agents with partial information to communicate under pressure. We find that language evolution does occur between LLM agents, that the resulting languages are compositional and morphologically productive, and that they deviate from the LLMs' English prior in ways that render them incomprehensible to humans. Moreover, we identify several qualities essential to this evolution: pressure towards efficiency; the strength of the models backing the agents; and access to a "postmortem" stage in which agents can agree on linguistic conventions. Importantly, we observe that different conditions govern the transmission of language to new agents. Specifically, we find that agents learn new languages from usage alone, take an active role in this learning, and that while stronger models are required for novel language emergence, weaker models can learn an existing language once it has emerged. Taken together, our results indicate that current LLMs have the potential for cumulative cultural evolution -- previously attested only in humans -- with mixed populations of agents developing capacities that go beyond their lowest common denominator.

Linear classifiers trained on hidden states of a large language model (LLM), linear probes, can flag factual errors from a single forward pass. Geometrically, that implies that true and false statements separate along a stable direction in hidden state space, i.e., the truth direction. Prior work disagrees on whether this generalises across input shifts, but the disagreement is hard to interpret because cross-dataset probe transfer experiments confound several kinds of input change at once. We isolate three such variables in medical question-answering (QA): writing style (register), domain (medical specialty), and corpus (dataset). We build a benchmark using 500 MedQA entries, each rewritten into four styles (textbook, patient, clinical note, colloquial), annotated with clinical specialty, and grouped with two other exam corpora, MedMCQA and MMLU-medical, for cross-dataset evaluation. Probing four open-weight LLMs (2--8B), we find that the truth direction is largely robust to writing style (mean $Δ_\text{register} \approx 0.10$ AUROC on held-out facts) and to medical specialty ($Δ_\text{specialty} \approx 0.03$), but degrades unevenly across corpora: by $0.12$ AUROC on MMLU-medical and by $0.21$ on MedMCQA, roughly twice the register gap. The register result replicates with a second generator and carries over to human-written patient questions. The truth direction is therefore largely stable within the medical domain but breaks under some corpus shifts, and question format does not explain the break, which suggests that the signal a linear probe recovers is partly bound to dataset structure rather than to medical knowledge alone.

Multilingual large language models (mLLMs) achieve strong performance in machine translation, yet our understanding of the mechanisms by which they transform representations from one language to another remains incomplete. Prior work suggests that translation decomposes into separable processes within an mLLM, where conceptual content is first represented independently, followed by a production into language-specific form. In this work, we show that translation is even more modular than previously assumed and that the output language production in translation processes is actually further separable into a syntax and a surface language process. We construct controlled multilingual datasets that isolate cross-linguistic differences in word-order and use causal interventions and probing to track how representations are transformed during translation. We find that models first construct target-side word-order before realizing the target language surface form. We identify individual attention heads that are selectively sensitive to syntactic transformations while remaining largely invariant to language identity. These results establish the commitment to a syntactic structure as an independent stage in translation, extending prior decompositions and showing how translation is implemented by functionally different components within mLLMs.

Do multilingual language models transfer factual knowledge across languages during continued pretraining, or do they mostly recall facts learned directly from the target-language data? To answer this question more reliably, we propose an intervention-based framework: starting from an English-pretrained model, we continue pretraining on Persian data from which specific facts have been systematically removed at varying levels of granularity. We construct SIFT, a resource of 500 triples across 20 relations, stratified by the cultural origin of each fact's subject into general (globally prominent) and Persian-related entities, designed for both systematic fact removal from training data and evaluation, with natively written Persian cloze templates. Our results show that fact transfer is very limited: under the strictest removal condition, a large majority of English-acquired facts fail to transfer into Persian. We further show that sentence-level co-occurrence removal is insufficient to eliminate fact signal, and that easier (randomly selected) negative candidate sets substantially inflate apparent transfer by rewarding shallow associative heuristics, while performance on a harder candidate set that allows for less reliance on heuristics is much lower. Finally, we show that source-language entity frequency has a large influence, with Persian-related facts, which are orders of magnitude rarer in the English corpus, hardly transferring.

Dialectal Arabic (DA) remains under-resourced compared to Modern Standard Arabic (MSA), particularly for machine reading comprehension (MRC) and question answering (QA). Existing Arabic QA benchmarks primarily focus on formal written MSA or multiple-choice QA, with limited coverage of naturally spoken dialects. Here, we aim to bridge this gap. We introduce EDRAC, the first large-scale benchmark for dialectal Arabic machine reading comprehension (MRC) and generative QA, covering five major dialects: Egyptian, Moroccan, Emirati, Syrian, and Saudi Arabic. EDRAC contains 499 passages derived from naturally occurring spoken interactions and 4,977 corresponding QA pairs generated through a human--LLM collaborative pipeline combining iterative generation, LLM-as-a-judge evaluation, and human verification. We benchmark Arabic-centric and multilingual LLMs on EDRAC using lexical and semantic metrics. Our results reveal substantial gaps between semantic answer quality and dialectal fidelity, highlighting the limitations of existing evaluation metrics for dialectal Arabic generation. EDRAC provides a realistic and challenging MRC benchmark for future research on dialectal Arabic NLP.

Despite the growing use of LLM-powered agents to solve multi-step tasks in complex environments, existing benchmarks rarely test state preservation, performance across languages, and application to realistic, grounded scenarios. To address these concerns, we present WorldBench: a comprehensive, multilingual benchmark of genuine, persona-grounded everyday workflows, where agents can act in a sandbox via structured actions. WorldBench comprises 1,600 tasks across seven languages and eight cultures, filtered and refined through feedback from human annotators with language- and culture-specific expertise. For evaluation, we extend metrics from previous works and introduce Constrained Task Success (CTS), which combines natural language instructions and testbeds to score task completion, minimal modification, and other complementary metrics through deterministic and LLM-as-a-Judge evaluations. Our experiments show that frontier models reach only 49.2% CTS, with all models demonstrating large gaps between correctness and environment preservation. We thereby show that current agents remain brittle in multilingual, agentic scenarios, especially for long-horizon tasks and under state-preservation constraints

Customer-service QA in an AI contact center (AICC) runs under deployment constraints that benchmark QA misses: tight voice-hotline latency and a high cost for unsupported or wrong automatic answers. We deploy a system that answers only from a closed set of verified QA units: it returns a retrieved unit verbatim, or routes to clarify, abstain, or handoff. The index is enriched offline by staged linguistic seeding (SLS): a human authors a per-unit world-grounded slot recipe, gpt-4.1-mini renders it into variants, and a light human gate filters them. One methodology is reused across both domains, so inference stays a single retrieval pass with no query-time generation. On held-out query variants from two industrial domains, SLS lifts hybrid R@1 to 0.881/0.930 (+0.27/+0.34), with gains across all five retrievers tested. At the same gpt-4.1-mini generation budget, SLS beats doc2query by +0.20/+0.32, while cross-provenance evaluation provides additional evidence of transfer across generated-query distributions. Verified-unit answering also removes free-form generation's unsupported-content surface (7-13% versus approximately 0%). We report this as an application study, including negative results.

Multi-party interaction is a central setting for human communication and a necessary target for human-agent interaction systems that must participate in group conversation. Yet available corpora often focus on meetings, task-oriented interaction, text-based interaction, or acted scenarios, and fewer resources support cross-linguistic comparison of spontaneous face-to-face triadic discussion. This paper presents TEIDAN, a multilingual multimodal corpus that currently consists of Japanese and English three-party conversations. TEIDAN records groups of three participants discussing open-ended topics with individual pin microphones, a microphone array, and participant-facing cameras, and provides IPU-based transcripts for both language portions. Earlier studies used subsets of the Japanese portion for task-specific benchmarks in multi-party dialogue modeling; in contrast, this paper presents TEIDAN as a corpus resource spanning both Japanese and English, with planned expansion to additional languages. We describe the collection design, participants, recording setup, transcription format, and corpus statistics, and provide preliminary analyses to illustrate how TEIDAN can support research on turn-taking, addressee recognition, and multimodal grounding in human-human and human-agent interaction.

Cultural fine-tuning has become the de facto paradigm for building culture-aware large language models (LLMs), yet existing optimization exclusively for alignment scores provides an incomplete portrait of cultural fidelity by systematically obscuring inherent cultural diversity. This unidimensional evaluation lens prompts a fundamental question: do models genuinely perceive distinct cultural nuances, or do they merely memorize dominant cultural values? To address this, we propose a synergistic evaluation framework that jointly formalizes cultural alignment and diversity. Through extensive benchmarking of six mainstream LLMs on the World Values Survey, this framework uncovers a systematic and critical trade-off: the pursuit of cultural alignment consistently incurs an acute expense of diversity, leading to severe "cultural flattening." Investigating this behavioral shift, we demonstrate that these superficial alignment gains stem from models artificially anchoring to dominant majorities, converging onto a monolithic response pattern that wipes out the heterogeneous distributions inherent to human groups. Crucially, our mechanistic analysis suggests that this diversity collapse is not merely a behavioral anomaly but more likely a structural consequence of the low-rank bias inherent in neural network optimization. Therefore, our findings expose the limitations of current post-training paradigms and call for a shift toward alignment objectives that preserve cross-cultural pluralism.

Large language models (LLMs) have recently demonstrated improved machine translation performance over strong supervised baselines. This raises questions as to what mechanisms underlie how LLMs perform machine translation between languages. Motivated by recent interpretability findings--namely, that LLMs use massively multilingual latent feature representations to perform language modeling--we propose the interlingua hypothesis. The hypothesis holds that language models translate by reading a source sentence into a latent feature space, and generate a target sentence by reading from the latent feature space. We show three lines of evidence in support of this hypothesis: (1) variance in BLEU across language pairs is largely predictable from language-specific competences with no language pair-specific interaction terms; (2) many model components are causally influential in both monolingual tasks and translation tasks; and (3) fine-tuning on monolingual data recovers a large proportion of translation improvements relative to fine-tuning on aligned documents. Together, these provide convergent evidence in support of the interlingua hypothesis, and suggest new ways of understanding and improving how LLMs can be leveraged to perform translation tasks.

Multilingual large language models can exhibit unintended code-switching -- unnecessarily alternating between languages during generation. We present a comparative study of three methods that identify language-controlling latents in cross-layer transcoders: activation value-based selection (ValSel), activation frequency-based selection (FreqSel), and LLM-generated latent annotation-based selection (AnnSel). To evaluate the efficacy of these methods in identifying language-controlling latents, we introduce two multilingual benchmarks that exhibit code-switching for fine-grained analysis of language steering across seven languages. Through targeted intervention experiments on Gemma-2-2B and Qwen3-4B, we find that all three methods effectively manipulate generation language, with FreqSel achieving the strongest overall performance, while AnnSel offering interpretable latent selection through explicit language annotations. A knock-out analysis suggests the methods select non-overlapping but each-functional latent subsets, indicating redundancy rather than a single canonical language direction. Code and data can be found at https://github.com/rm-3284/Latent-Mechanism-Multilingual.

Latent language identification is often used to argue that multilingual language models route computation through language-specific states, such as English pivots. However, existing probes infer latent language from different signals, such as the geometry of hidden states or what can be decoded from intermediate representations. Since such claims shape conclusions about how models share and route information across languages, we ask whether these probes measure the same phenomenon or expose distinct aspects of multilingual computation. We study this question across model families, training regimes, domains, tasks, checkpoints, and up to 27 languages. We find that identification probes systematically disagree: the GMM-based representation probe, which draws evidence from hidden state geometry, shows earlier cross-lingual mixing, whereas decoding-based probes, which rely on output-space decodability, retain sharper language-specific and more English-biased signals. These differences track model multilinguality and training progression, but are comparatively stable across domains. Our results suggest a more cautious interpretation of latent language identification, where current probes expose different aspects of multilingual processing, rather than directly revealing a single internal lingua franca.

LLMs can imitate how people write, which raises concerns about impersonation, trust, and detection in social settings. These concerns are especially important for adolescents, who use generative AI frequently but may struggle to recognize it. We introduce \textit{DoppelBot}, a cooperative social deduction game designed to study how young people detect and respond to AI impersonation. Through studies with middle schoolers, we investigate whether a DoppelBot prompts reflection on privacy and impersonation, how repeated exposure affects AI-detection accuracy as agents become more personalized, and which strategies students use to identify AI doppelgängers. We find that students' detection accuracy improves over time, driven by a shift from relying on linguistic cues to leveraging shared social and contextual signals. Students also demonstrated an understanding of AI limitations such as embodiment and reflected on broader issues such as data privacy. To support future research, we release an anonymized dataset of game transcripts and voting behavior.

While gender and racial biases in language models have been widely studied, anti-LGBTQ biases remain underexplored, particularly beyond English. Existing benchmarks often do not capture cultural and linguistic variation and rely on gender representations. This paper introduces a multilingual German-English benchmark dataset for the evaluation of anti-LGBTQ biases in language models. It combines community-sourced stereotypes from German-speaking queer individuals with a German translation of WinoQueer. The data is used to evaluate eight language models across sizes and architectures and explore mitigation through fine-tuning on community and progressive media content. Results show that language models reproduce anti-queer stereotypes, with variation across identities and models. Differences between the translated and community-based data highlight the importance of cultural adaptation for multilingual bias evaluation. Fine-tuning reduces bias on average, but not consistently across models and identities. Warning: This text contains examples of anti-queer hateful language and stereotypes.

Building language technologies and conducting NLP research for low-resource languages---particularly when led by native speakers or involving participatory research practices---are often framed as means of addressing inequality, serving local communities, and, at times, contributing to decolonisation. In this paper, we examine recently published NLP and ML papers, focusing on the narratives used to characterise multilinguality, low-resource languages, and underrepresented cultures. We propose a framework for analysing research framings and identify recurring rhetorical patterns that may hinder accountability and constrain equitable knowledge production for---and by---underserved communities. We further assess the evidential basis of assertions regarding community benefit and find that such statements are often weakly supported or left unsubstantiated. Although community ownership and participation are frequently presented as key objectives, our analysis, supported by statistics from the ACL Anthology, suggests that research outputs more often prioritise resource creation and benchmarking---important but distinct goals---over evidence of broader structural change. We conclude by offering practical recommendations to help authors, reviewers, and readers critically assess these assertions and avoid potentially misleading framings.

While automatic speech recognition (ASR) models have achieved remarkable improvements in recent years, performance disparities persist across different speaker populations. One such disparity is for speakers whose first languages (L1) are from families distant from English. This paper investigates the relationship between first language background and English ASR performance. Through empirical analysis, we observe that the correlation between speakers' L1 distance and ASR error rates yields a systematic effect on English Speech, with its strength varying across datasets and models. This association is statistically significant in a follow-up analysis accounting for dataset-level variation in Tweedie mixed-effects models ($p<0.001$ across evaluated models). In addition, analysis of the latent space reveals a L1-based spatial segregation across deeper acoustic layers in the majority of evaluated architectures

The vocal music of each language carries a distinctive sonic identity, even without instrumental accompaniment. We ask whether these differences are measurable and traceable to specific phonemes. To tackle this question, we introduce phoneme-conditional analysis, which isolates the acoustic effect of typologically distinctive phonemes by comparing marker syllables against matched non-marker controls within the same song, holding singer, melody, and genre constant. Across nine typologically diverse languages and thousands of songs, we measure effects along five acoustic dimensions. Song-level profiles built from these effects identify the language of an unaccompanied vocal at 85.5% balanced accuracy in a nine-way classification with folds grouped by artist; whether the separability arises by accumulation of the phoneme-local effects themselves is left open. Our findings suggest that phonological structure leaves systematic and measurable traces in how each language is sung.

Recent multilingual vision--language encoders cover hundreds of languages in a single model, yet on two state-of-the-art instances retrieval on low-resource languages (LRL; e.g. Swahili) trails high-resource ones (HRL; e.g. English) by $30^+$,pp. We ask where in the trained encoder this gap is located. Prior modality-gap and cross-lingual subspace work suggests a linear language direction at the output crowds out alignment-relevant geometry. We falsify this: LEACE drives the linear language classifier from $>99\%$ to near chance and iterated INLP to $37$--$50\%$ while LRL retrieval moves within $\pm 1.5$,pp and all tier means within $2.2$,pp, tracking random controls. The linear bias is a \emph{symptom}, not the cause. Instead, the alignment-causal factor lies along the encoder's forward path: the EOS (end-of-sequence) hidden state's per-language trajectory diverges with depth. Substituting the EOS with its parallel English value three blocks before the projector lifts Swahili from $22.1\%$ to $69.1\%$ on one encoder (and reproduces on the other); three controls rule out pooled-position tautology and English specificity. A front-layer trunk that pulls each language's projection toward the parallel-content centroid corroborates the diagnosis at training time, recovering $+9.6$ / $+17.1$,pp on LRL XM3600 retrieval (1{,}000-image subset), with consistent gains across three further benchmarks while preserving HRL performance.

Standard language proficiency tests rely on linguistic tasks such as vocabulary, grammar and reading comprehension quizzes. An alternative, cognitively motivated approach, introduced in Berzak et al. (2018), proposed instead to predict language proficiency from behavioral traces of eye movements in reading. In this work, we validate and extend this approach from single sentences to more naturalistic reading of contextualized passages in English as a second language, new proficiency measures, prediction models, and reading in an information seeking regime. We find that the approach is effective in all these evaluations. We further address two key open questions on eye movement based proficiency testing: (1) potential scoring biases that reflect the proximity of the reader's native language to English, which may undermine validity, and (2) its reliability. We find that eye movement based proficiency scores are indeed biased towards L1s that are linguistically closer to English. We propose a score debiasing method which effectively remedies this issue. The reliability analyses suggest that eye movement proficiency scores are more reliable than standard language proficiency scores. Overall, our results strengthen and broaden the empirical foundations for future eye movement based language assessment technologies.

Pixel-based language models (LMs) replace traditional tokenizers by processing rendered images of text, making cross-lingual transfer heavily dependent on the visual and structural properties of writing systems. However, the dynamics of adapting these models to low-resource languages with complex morphology and written in unique scripts are not yet explored. Using Tibetan as a case study, we analyze how continued pre-training of pixel-based LMs is influenced by data scale, initial script exposure, and cross-lingual transfer from languages written in other Brahmic scripts. We introduce four rendering-level metrics to quantify visual script similarity. We evaluate downstream performance across three tasks. Our results show that higher orthographic proximity enhances semantic transfer, even under severe data constraints. Additionally, we find a performance asymmetry based on the pre-training starting point: while multilingual pre-training PIXEL-M4 has stronger initial performance, its capacity for subsequent adaptation seems to be constrained, whereas adapting a monolingual model PIXEL with mixed scripts yields more gains on sentence-level tasks. Our metrics and case study offer empirical observations that could help inform data selection and script adaptation choices when working with pixel-based models in similar low-resource settings.

Solvability detection is one of the most challenging aspects of mathematical reasoning for Large Language Models (LLMs). While prior work has studied this capability extensively, these analyses have been limited to English. Consequently, it remains unclear whether multilingual failures arise from differences in internal Solvability Belief or from language-dependent failures to express it. To address this gap, we introduce the first multilingual benchmark of paired solvable and unsolvable mathematical problems, extending ReliableMath to French and Greek. Using this, we train multilingual probes predicting Solvability Belief and analyze the solvability detection capabilities of state-of-the-art LLMs behaviorally, representationally, and in terms of faithfulness. We find that Solvability Belief is encoded as a largely universal, language-agnostic feature, and that higher-resource languages such as English, despite achieving stronger mathematical reasoning performance, exhibit lower solvability-detection faithfulness.

Large language models exhibit substantial performance variation across languages, even when solving semantically equivalent tasks. Existing analyses often treat this phenomenon as an observational disparity caused by differences in pretraining data, tokenization, or benchmark coverage. We study a complementary hypothesis: high-resource languages (HRLs) may more reliably elicit latent computations useful for task-specific (i.e. mathematical) reasoning, while lower-resource languages (LRLs) may under-activate those computations despite expressing the same task. To test this hypothesis, we introduce a mechanistic intervention framework for identifying and transferring task-relevant sparse latent features across languages. Using sparse autoencoders over residual-stream activations, we isolate features enriched in successful HRL task-specific reasoning while filtering out source-language and generic-generation features. We then construct steering directions from these features and inject them during LRL inference. The resulting interventions test whether the selected features are functionally involved in the observed reasoning gap: suppressing them should impair source-language reasoning, while activating them should partially recover target-language reasoning beyond random and non-task controls. Our framework reframes some cross-lingual reasoning gaps as failures of mechanism elicitation rather than capability absence, and offers a causally testable route to feature-mediated transfer without translation, fine-tuning, or changing the user-facing language.

Cross-lingual aspect-based sentiment analysis (ABSA) transfers knowledge from a source language with annotated data to a target language, enabling fine-grained sentiment analysis without annotated target-language data. While monolingual ABSA has seen significant progress, cross-lingual ABSA remains underexplored, especially for complex tasks involving multiple sentiment elements like target-aspect-sentiment detection (TASD). In this paper, we propose a novel SeqLab framework that enhances cross-lingual ABSA using a sequence-to-sequence model with an auxiliary sequence-labelling task performed by the encoder, enhancing aspect term recognition and sentiment predictions. Additionally, we incorporate aspect-code switching (ACS), a translation-based technique that swaps aspect terms between source and translated sentences, generating additional training data to enhance the model's cross-lingual understanding. We evaluate our approach across eleven languages, three domains, and two backbone models, surpassing previous state-of-the-art results for the commonly studied E2E-ABSA task. Unlike most prior work that relies solely on English as the source language, we systematically assess different source-target language pairs and extend our evaluation to the more challenging, yet underexplored TASD task in cross-lingual settings. Finally, we provide a detailed error analysis highlighting key challenges and limitations.

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

We present Arkios, a 1.04B-parameter dense transformer pretrained from scratch on 150B tokens of bilingual English-Nepali text, using a custom single-file C/CUDA training stack and a Devanagari-aware byte-level BPE tokenizer built for this project. On ARC-Easy and ARC-Challenge, Arkios exceeds three comparably sized open models (Pythia-1.4B, TinyLlama-1.1B, OLMo-1B) despite an order of magnitude fewer training tokens, likely aided by a match between our educational-web-text pretraining data and ARC's grade-school-science format rather than a general capability advantage. We report full evaluation results under standard protocols, including a correction to an earlier partial-sample estimate, and findings specific to evaluating small models in a low-resource language: the standard multiple-choice-letter prompt format used by common evaluation harnesses places this model at chance on Nepali reading comprehension, and simultaneously at chance on English in the same format, which would lead a naive benchmark run to conclude the model has no Nepali ability when in fact it does. Concretely, both languages score at chance in the letter-choice format (0.240 Nepali, 0.236 English, against a chance baseline of 0.250), while scoring the answer text directly reveals genuine, English-favoring comprehension (0.306 Nepali, 0.387 English). We describe a manifest-conditioned tool-use contract introduced during instruction tuning, where tool calls are permitted only when a tool manifest is declared in context and suppressed otherwise, and report where that contract holds and where it does not. We release both the base and instruction-tuned model weights under Apache-2.0. The training code and a small privately-sourced portion of the Nepali pretraining corpus are not released; everything needed to reproduce the reported numbers from the released weights is included here.

When a generative search interface answers a commercial question, which market's products it names is decided before the model reasons about the products. We report a controlled probe of 234 runs against the logged-out ChatGPT web interface and the OpenAI API, collected on 29 and 30 August 2026 across four exit countries and six query languages, with six identical runs per cell. Three results. First, the top recommendation is unstable: it changed across six identical runs on four of six prompts, and that rate was identical in the browser interface and in the API with web search both enabled and disabled, so instability is a property of the system and not of the surface. Second, query language, and not location, decides whether local suppliers appear at all. Where the query language matched the country, a global brand won 1 of 24 runs; asked in English on the same connections, local brands took 0 of 6 runs in Estonia and Turkiye. Third, language and location are separable and act on different things: holding the query language fixed and moving only the exit IP moves the market whose brands are named while the answer stays in the query language. We show this on two unrelated pairs, Turkish asked from Berlin and Russian asked from Tallinn, and in both the answer names the resident country's suppliers. A minority language occupies a middle tier: Russian asked from Estonia names an Estonian supplier in 4 of 6 runs and a global one in all six, where Estonian names a local supplier in every run and English names none. A negative control in a second category, coded with the same instrument, shows no language effect at all, and disconfirms our own expectation: that category does have domestic suppliers and none was named in any language, which points the explanation at whether a category is nationally regulated rather than at whether it is nationally supplied.

Language is humanity's most consequential technology, yet for over a billion speakers across India's twenty-two constitutionally recognised languages, its digital layer remains structurally incomplete. Named Entity Recognition (NER), the foundational step in transforming raw text into machine-interpretable knowledge, has been studied exhaustively for English but remains largely unsolved across most Indic languages. This paper presents a rigorous comparative study of generative and encoder-based neural architectures for NER on all eleven languages of the Naamapadam benchmark. We evaluate five classic model families spanning sequence-to-sequence transformers and multilingual encoders; four decoder-only large language models (LLMs) fine-tuned with LoRA and 4-bit NF4 quantisation; and nine generative models in zero-to-5-shot inference. Under strict CoNLL span-level evaluation, encoder-based models (mBERT and XLM-R, both F1=0.675 on Hindi) substantially outperform every generative architecture in ten of eleven languages, with gaps of 7.5-40 percentage points against the strongest competitor (Gemma-2-2B: avg F1=0.427). The best few-shot result reaches only 28% of the encoder baseline. We identify three language clusters--encoder-dominant, partial-coverage, and failure-zone; and provide actionable deployment guidelines grounded in transfer learning and low-resource NLP principles.

Large language models often answer complex reasoning questions without revealing intermediate steps, raising whether they reason latently or complete patterns. We propose the Hidden CoT Detection Score (HCDS), a comparative behavioral and mechanistic signal measuring whether neutral-prompt behavior aligns more closely with explicit CoT or explicit no- CoT. Here, hidden CoT operationally denotes this neutral-prompt CoT-like alignment; HCDS does not directly observe or prove an unexposed reasoning trace. On GSM8K, HCDS is significantly positive for both Qwen3-4B variants (Thinking $+1.87$, $p = 1.2 \times 10^{-7}$; Instruct $+1.41$, $p = 1.9 \times 10^{-4}$), replicates across a different inference stack and quantization within $0.08$ ($+1.80$ and $+1.45$), and is not significantly positive in seven of eight length-adjusted calibration-control cells. The unadjusted score produces large positive scores on single-step arithmetic and numeric factual lookup. The variants also respond differently to no-CoT instructions: Instruct complies from the prompt alone, whereas Thinking continues reasoning and requires intervention. These findings show stronger, less prompt-conditional CoT-like behavior in the reasoning-tuned model, consistent with but not proof of latent reasoning. HCDS thus investigates latent reasoning without relying on models' self-reported traces.

Safety alignment of large language models (LLMs) degrades across languages, yet the internal mechanism driving this asymmetry remains poorly understood. Our work, therefore, presents a systematic mechanistic analysis of multilingual safety using sparse autoencoder (SAE) features, sparse interpretable directions in the residual stream associated with harmful and harmless model behavior across three instruction-tuned LLMs, eight languages, and all model layers. We observe that safety-relevant features are architecture-dependent in terms of where they are located and how they are distributed across layers. Additionally, they are geometrically entangled with language identity and exhibit cross-lingual sharing patterns, i.e., languages share safety features to varying degrees across model depths and architectures. This safety-language entanglement has direct consequences such that ablating safety features impacts not only harmful response rates but also target language, with the degree of intervention predicted by the relationship between safety and language features. Our findings qualify the language-universality of safety alignment as architecture-dependent and offer a mechanistic account of multilingual safety interventions.

The rapid proliferation of LLMs has further heightened the need to develop dependable AI-generated text detection, especially beyond English. Nevertheless, current benchmarks pay little attention to Indic languages and test detectors in idealized settings that do not represent the real world. We present a generalized benchmark for AI-generated text detection in Hindi, Telugu, and Tamil, which we call IndicDetect, designed to assess the robustness of detectors under realistic distribution shifts. IndicDetect comprises highly curated human-written texts matched with LLM-generated counterparts across various domains and generators, and systematically evaluates detectors in the presence of domain shift, generator shift, and adversarial perturbation. Using a single and repeatable evaluation scheme, we evaluate a wide range of statistical and neural detectors. We find substantial robustness failures: supervised neural detectors perform well in-distribution, while training-free methods degrade considerably under unseen generators and adversarial attacks. The severity of these failures varies across languages, with Hindi exhibiting the largest overall degradation under adversarial perturbations. These results highlight that the primary weakness of existing detectors in Indic settings lies in their robustness, not in their peak accuracy. IndicDetect provides standard data splits, an evaluation protocol, and baselines to establish a robust, language-aware foundation for AI-generated text detection in Indic scripts.

Reinforcement learning from human feedback (RLHF) has become the dominant paradigm for aligning large language models (LLMs) with human preferences. However, traditional RLHF relies on scalar reward signals that lack interpretability and fail to capture the multifaceted nature of response quality. Rubric-guided reinforcement learning addresses these limitations by introducing structured, interpretable evaluation criteria, or rubrics, as the backbone of reward design, feedback generation, and policy optimization. In this survey, we introduce a Bayesian framework that defines constitutions as prior distributions $P(R)$ over evaluation criteria and rubrics as conditional instantiations $R_x \sim P(R|x)$. Under this unified view, we present a taxonomy of rubric-guided RL along the prior-posterior axis, covering constitutional AI, instance-specific rubrics, process-level supervision, self-evolving rubrics, and their agentic and multimodal extensions. Furthermore, as rubrics are natural-language artifacts, we present a linguistic analysis of how granularity trade-offs, semantic drift, and linguistic reward hacking impact alignment reliability, identifying key open problems for future research.

A concept can carry different associations across languages, while modern language models learn English alongside many other languages during pretraining. Yet comparisons among existing models cannot easily isolate how any one language changes the way these models represent English concepts because their training corpora, compute, architectures, and random seeds all differ. We study this question through a controlled experiment with 40 matched 310M-parameter decoder-only models that share an architecture, tokenizer, training recipe, and English data source. Each bilingual condition adds one of eight languages, while four experimental comparisons separately account for English exposure, total training, and English-document overlap. We align each model pair using 3,000 common English words, then measure where 1,000 held-out English concepts fall along 50 fixed semantic contrasts, such as red versus white. Across 32 experimental comparisons, English concept positions differ more between bilingual and English-only conditions than between English-only runs with different random seeds. These differences are larger in contextual states than in token embeddings and peak in middle layers. The language learned alongside English can therefore change how a model represents English concepts even when its English input representations are explicitly aligned.

Multilingual language models transfer knowledge across languages through shared subword vocabulary, a mechanism that breaks down when related languages use different writing systems. Prior work addresses this via script equalization (romanization or IPA transcription), but direct comparisons are rare; the focus has been on encoder-only models, with most work adapting existing pretrained models. We systematically compare different input representations in autoregressive multilingual pretraining, comparing orthographic text, IPA, and romanization in a controlled setup across three scales (467M, 709M, and 1.03B) on eight languages in four typologically motivated pairs. Across a wide range of downstream tasks on seen and unseen languages, romanized pretraining yields the strongest cross-lingual transfer, and the advantage over text widens with scale. IPA improves over text in most settings but trails romanization. Surprisingly, finetuning a text-pretrained model on romanized data hurts performance on languages already covered by the base model, only marginally helping when the model lacks script coverage. Our results indicate that for multilingual models spanning typologically diverse scripts, to obtain maximum benefits, romanization should be treated as a core design choice applied at pretraining rather than a post hoc fix.

Crosslingual evaluation of language models that enables fair comparisons remains a fundamental challenge in multilingual NLP. Existing studies adopt a variety of downstream tasks and intrinsic metrics with different theoretical justifications, yet there has been little empirical investigation into whether these approaches yield meaningful crosslingual conclusions. We systematically examine crosslingual evaluation approaches using controlled monolingual language models trained on parallel data with varying tokenizer vocabulary sizes and model sizes, and further validate our findings on multilingual LLMs. We further discuss challenges in achieving comparable downstream evaluation across languages. Our results show that several widely used normalized metrics introduce crosslinguistic biases rooted in tokenization, encoding, and orthographic differences. In contrast, sentence-level negative log-likelihood computed over semantically equivalent sequences provides more meaningful and consistent crosslingual comparisons.