Cortex uses an Ontological Corpus Graph to structure web-scale corpora, creating a refined 24.14B-token corpus and a new benchmark validated on eight LLMs.
hub
The State and Fate of Linguistic Diversity and Inclusion in the NLP World
44 Pith papers cite this work, alongside 289 external citations. Polarity classification is still indexing.
44
Pith papers citing it
289
external citations · Crossref
hub tools
citation-role summary
background 4
citation-polarity summary
roles
background 4representative citing papers
OpenBibleTTS supplies speech data and alignments for 37 underrepresented languages and shows that no single TTS system leads on all metrics, with Gemini-TTS highest in listener ratings but monolingual EveryVoice models strongest on intelligibility for several African languages.
Evaluation across 1.1 million instances shows sycophancy rates spike in low-resource languages, remain topic-agnostic, and correlate with tokenizer fertility.
RL with chrF reward trains LLMs to better utilize in-context linguistic knowledge for zero-shot translation of unseen languages, outperforming ICL and SFT.
MIDI is a new multilingual idiom dataset with sentence and conversational contexts; benchmarking reveals worse performance in low-resource languages and on literal vs. figurative uses.
Repetition rate mismatch between small-scale proxies and target budgets is the main reason data mixture experiments do not scale; a subsampling procedure that equalizes repetition rates recovers optimal mixtures from 1/16-scale experiments.
TriMix dynamically fuses logits from three model sources to outperform baselines and Proxy Tuning on eight low-resource languages across four model families.
Lesioning a shared core in multilingual LLMs drops whole-brain fMRI encoding correlation by 60.32%, while language-specific lesions selectively weaken predictions only for the matched native language.
LLMs default to responses more similar to opinions from the USA and some European and South American countries; prompting for a country shifts alignment but can introduce stereotypes, while translation does not reliably match language speakers.
A new multilingual corpus for Cham, Khmer, and Tay-Nung, plus a script-aware ELECTRA-style recipe, achieves the best topic-classification accuracy among the tested encoders — and shows why lexical-overlap retrieval can mislead low-resource benchmarks.
Dialectal robustness and generation are dissociated in LLMs: benchmarks are driven by pretraining and SFT while alignment reshapes generation invisibly to benchmarks, and the method maximizing dialectal reward is least preferred by human evaluators.
PluraMath extends PolyMath with human-validated math problems in 18 mid-to-extreme low-resource languages and benchmarks 27 reasoning LLMs, finding a persistent high- vs low-resource performance gap.
Benign multilingual fine-tuning causes language-specific safety drifts with adversarial compliance rates rising up to four-fold, decoupled from capability gains.
Analysis of 11 LLMs on 21 disputed inventions across 12 languages and 75,896 responses finds query language systematically shifts credit toward lower-status claimants in their associated language while Anglophone figures remain stable.
Optimal interpolation of query embeddings from parallel translations outperforms the best monolingual query in 88/105 cases on mMARCO, showing English-driven asymmetry and negative correlation with typological distance.
MUDIDI introduces a two-stage LLM pipeline for multilingual dictionary digitization, releases a human-annotated dataset from 30 dictionaries, and shows LLMs outperforming OCR and VLMs on character recognition, markup, and entry segmentation.
Using a 1PL IRT model on real cultural questions across 13 locales, the study identifies a local-language knowledge-access advantage masked by lower proficiency in raw accuracy.
Manual audit shows web-scraped Lombard corpora are largely noisy and biased toward Western varieties over Eastern ones.
CRAFT is a Pareto-front prompt optimizer that allocates scarce LLM validation calls to candidates near the current front using accuracy- and cost-oriented generators plus NSGA-II retention.
Geometric analysis shows LLM judges use collapsed score ranges and nearly orthogonal axes to humans on subjective rubrics across Indic languages, with inter-LLM agreement not equaling human alignment except on factual tasks.
Luar is a reinforcement learning method enabling reasoning language models to decide when to invoke English translation for improved multilingual reasoning.
Introduces the MEA benchmark for multi-target cross-lingual summarization across 24 languages and demonstrates that activation steering from English summarization representations improves performance.
LLMs default to U.S. frameworks for English prompts and China frameworks for Chinese prompts on jurisdiction-underspecified legal-administrative queries, with the pattern holding across all seven tested models.
A Bayesian framework decomposes mLLM variance, showing language features explain 79-92% of language identity variance and that model identity vs. benchmark-model interactions dominate differently for understanding versus reasoning tasks.
citing papers explorer
-
CORTEX: High-Quality Cross-Domain Organization of Web-Scale Corpora through Ontological Corpus Graph
Cortex uses an Ontological Corpus Graph to structure web-scale corpora, creating a refined 24.14B-token corpus and a new benchmark validated on eight LLMs.
-
OpenBibleTTS: Large-Scale Speech Resources and TTS Models for Low-Resource Languages
OpenBibleTTS supplies speech data and alignments for 37 underrepresented languages and shows that no single TTS system leads on all metrics, with Gemini-TTS highest in listener ratings but monolingual EveryVoice models strongest on intelligibility for several African languages.
-
Sycophancy as a Multilingual Alignment Failure: How Safety Degrades Across Languages, Topics, and Models
Evaluation across 1.1 million instances shows sycophancy rates spike in low-resource languages, remain topic-agnostic, and correlate with tokenizer fertility.
-
Reinforcement Learning Elicits Contextual Learning of Unseen Language Translation
RL with chrF reward trains LLMs to better utilize in-context linguistic knowledge for zero-shot translation of unseen languages, outperforming ICL and SFT.
-
Multilingual Idioms in Sentences and Conversations Across High-, Medium-, and Low-Resource Languages
MIDI is a new multilingual idiom dataset with sentence and conversational contexts; benchmarking reveals worse performance in low-resource languages and on literal vs. figurative uses.
-
Repetition Mismatch: Why Data Mixture Experiments Don't Scale and How to Fix Them
Repetition rate mismatch between small-scale proxies and target budgets is the main reason data mixture experiments do not scale; a subsampling procedure that equalizes repetition rates recovers optimal mixtures from 1/16-scale experiments.
-
Efficient Low-Resource Language Adaptation via Multi-Source Dynamic Logit Fusion
TriMix dynamically fuses logits from three model sources to outperform baselines and Proxy Tuning on eight low-resource languages across four model families.
-
Computational Lesions in Multilingual Language Models Separate Shared and Language-specific Brain Alignment
Lesioning a shared core in multilingual LLMs drops whole-brain fMRI encoding correlation by 60.32%, while language-specific lesions selectively weaken predictions only for the matched native language.
-
Towards Measuring the Representation of Subjective Global Opinions in Language Models
LLMs default to responses more similar to opinions from the USA and some European and South American countries; prompting for a country shifts alignment but can introduce stereotypes, while translation does not reliably match language speakers.
-
Echoes Across Vietnam's Highlands, Delta, and Coast: A Multilingual Corpus for Cham, Khmer, and Tay-Nung
A new multilingual corpus for Cham, Khmer, and Tay-Nung, plus a script-aware ELECTRA-style recipe, achieves the best topic-classification accuracy among the tested encoders — and shows why lexical-overlap retrieval can mislead low-resource benchmarks.
-
DiaLLM: An Investigation into the Robustness-Generation Gap in English Dialect Adaptation
Dialectal robustness and generation are dissociated in LLMs: benchmarks are driven by pretraining and SFT while alignment reshapes generation invisibly to benchmarks, and the method maximizing dialectal reward is least preferred by human evaluators.
-
PluraMath: Extending Mathematical Reasoning Evaluation Beyond High-Resource Languages
PluraMath extends PolyMath with human-validated math problems in 18 mid-to-extreme low-resource languages and benchmarks 27 reasoning LLMs, finding a persistent high- vs low-resource performance gap.
-
The Heterogeneous Safety Impacts of Benign Multilingual Fine-Tuning
Benign multilingual fine-tuning causes language-specific safety drifts with adversarial compliance rates rising up to four-fold, decoupled from capability gains.
-
Same question, different history: language, national identity, and credit in large language models
Analysis of 11 LLMs on 21 disputed inventions across 12 languages and 75,896 responses finds query language systematically shifts credit toward lower-status claimants in their associated language while Anglophone figures remain stable.
-
When Does Mixing Help? Analyzing Query Embedding Interpolation in Multilingual Dense Retrieval
Optimal interpolation of query embeddings from parallel translations outperforms the best monolingual query in 88/105 cases on mMARCO, showing English-driven asymmetry and negative correlation with typological distance.
-
MUDIDI: A Two-Stage Framework for Multilingual Dictionary Digitization with Language Models
MUDIDI introduces a two-stage LLM pipeline for multilingual dictionary digitization, releases a human-annotated dataset from 30 dictionaries, and shows LLMs outperforming OCR and VLMs on character recognition, markup, and entry segmentation.
-
The Masked Advantage: Uncovering Local-Language Access to Cultural Knowledge in LLMs
Using a 1PL IRT model on real cultural questions across 13 locales, the study identifies a local-language knowledge-access advantage masked by lower proficiency in raw accuracy.
-
"Chi nas dal soch el sent de legn" -- Auditing Text Corpora for Lombard
Manual audit shows web-scraped Lombard corpora are largely noisy and biased toward Western varieties over Eastern ones.
-
CRAFT: Cost-aware Refinement And Front-aware Tuning of Prompts
CRAFT is a Pareto-front prompt optimizer that allocates scarce LLM validation calls to candidates near the current front using accuracy- and cost-oriented generators plus NSGA-II retention.
-
The Geometry of LLM-as-Judge: Why Inter-LLM Consensus Is Not Human Alignment
Geometric analysis shows LLM judges use collapsed score ranges and nearly orthogonal axes to humans on subjective rubrics across Indic languages, with inter-LLM agreement not equaling human alignment except on factual tasks.
-
Learning When to Translate for Multilingual Reasoning
Luar is a reinforcement learning method enabling reasoning language models to decide when to invoke English translation for improved multilingual reasoning.
-
Understanding LLM Behavior in Multi-Target Cross-Lingual Summarization
Introduces the MEA benchmark for multi-target cross-lingual summarization across 24 languages and demonstrates that activation steering from English summarization representations improves performance.
-
Which Institutional Frameworks Do Chatbots Assume? Auditing Jurisdictional Defaults in Multilingual LLMs
LLMs default to U.S. frameworks for English prompts and China frameworks for Chinese prompts on jurisdiction-underspecified legal-administrative queries, with the pattern holding across all seven tested models.
-
DEPART: DEcomposing PARiTy across Multilingual LLMs
A Bayesian framework decomposes mLLM variance, showing language features explain 79-92% of language identity variance and that model identity vs. benchmark-model interactions dominate differently for understanding versus reasoning tasks.
-
Beyond Catalogue Counts: the Dataset Visibility Asymmetry in Low-Resource Multilingual NLP
Catalogue records show 141 languages as data-poor, but citation mining reveals 609 datasets across 53 languages, exposing a visibility gap in multilingual NLP resources.
-
Scaling Laws for Mixture Pretraining Under Data Constraints
Empirical study shows mixture pretraining tolerates higher target data repetition than single-source training, with a new repetition-aware scaling law enabling principled mixture selection based on data size, compute, and model scale.
-
Lost in the Tower of Babel: The Adverse Effects of Incidental Multilingualism in LLMs
Frontier LLMs' self-declared language support is unstable and over-optimistic, verified behavior is task-dependent, and language mismatch alone degrades collaborative agent performance.
-
COMPASS: COntinual Multilingual PEFT with Adaptive Semantic Sampling
COMPASS uses semantic clustering on multilingual embeddings to select auxiliary data for PEFT adapters, outperforming linguistic-similarity baselines on multilingual benchmarks while supporting continual adaptation.
-
SLoW: Select Low-frequency Words! Automatic Dictionary Selection for Translation on Large Language Models
SLoW selects low-frequency word dictionaries to boost LLM translation quality and efficiency across 100 languages from FLORES.
-
How Good is Your Wikipedia? Auditing Data Quality for Low-resource and Multilingual NLP
The study filters non-English Wikipedia, reveals quality problems, proposes a 4-level ranking, and shows filtered data matches or beats raw data in language modeling with largest gains for lower-quality editions.
-
Ethical and social risks of harm from Language Models
The authors provide a detailed taxonomy of 21 risks associated with language models, covering discrimination, information leaks, misinformation, malicious applications, interaction harms, and societal impacts like job loss and environmental costs.
-
Challenges and Recommendations for LLMs-as-a-Judge in Multilingual Settings and Low-Resource Languages
Meta-analysis of 33 ACL papers shows inconsistent LLM-as-a-Judge results, overtrust, and single-model reliance in multilingual/low-resource settings, with recommendations for better practice.
-
LANG: Reinforcement Learning for Multilingual Reasoning with Language-Adaptive Hint Guidance
LANG combines language-adaptive hint guidance, progressive decay, and difficulty-tailored learning horizons in RL to boost non-English reasoning performance while preserving language consistency.
-
High-Volume Plaintiff-Side Counsel and Single-Appearance Eviction Cases in Philadelphia
High-volume plaintiff-side counsel in Philadelphia eviction cases scales up filing volume and procedural steps but does not produce a broad premium on adverse tenant outcomes such as default or judgment.
-
Which Are the Low-Resource Languages of the Semantic Web?
A multi-level categorization from language distributions in DBpedia, BabelNet, and Wikidata defines low-resource languages for Semantic Web knowledge graphs.
-
Marco-MoE: Open Multilingual Mixture-of-Expert Language Models with Efficient Upcycling
Marco-MoE delivers open multilingual MoE models with 5% activation sparsity that outperform similarly sized dense models on English and multilingual benchmarks through efficient upcycling.
-
Adam's Law: Textual Frequency Law on Large Language Models
For meaning-matched paraphrases, higher estimated sentence frequency improves LLM prompting and fine-tuning across math, translation, commonsense, and tool-calling tasks.
-
How do datasets, developers, and models affect biases in a low-resourced language?: The Case of the Bengali Language
Bengali sentiment analysis models exhibit persistent identity-based biases across datasets and developer backgrounds despite similar semantic content.
-
Bridging the Usability Gap: Lessons from Interpreting Studies for Machine Interpreting Design
Machine interpreting should shift from fidelity metrics to three design priorities—agency, grounding, and experience—drawn from interpreting studies to close the usability gap with human-mediated communication.
-
Model-Based Quality Assessment for Massively Multilingual Parallel Data
Large-scale benchmarks of multilingual embeddings and QE models show no universal performer; direction-aware routing and calibration recommended for parallel data assessment.
-
In Data or Invisible: Toward a Better Digital Representation of Low-Resource Languages with Knowledge Graphs
A research plan to analyze language distribution in LOD knowledge graphs and explore cross-lingual transfer plus analogical reasoning to improve coverage for low-resource languages.
-
CAT-Translate: Building Compact Open-Source Models for Japanese-English Translation
Compact 0.8B-7B models for bidirectional Japanese-English translation outperform large multilingual models on real-world domain benchmarks.
-
Multilingual Vision-Language Models, A Survey
The survey identifies a key tension in multilingual vision-language models between language neutrality via contrastive learning and cultural awareness via diverse data, with most benchmarks relying on translation-based evaluation.
- Lessons from the Trenches on Reproducible Evaluation of Language Models