# Internationalization (i18n)

Internationalization (i18n) is the practice of designing and developing systems, content, and applications so they can be adapted for users from different cultures, regions, or languages.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## OVHcloud AI Summit 2026: 55 launches to take AI into production

DevFeed: [OVHcloud AI Summit 2026: 55 launches to take AI into production](<https://devfeed.tech/articles/ovhcloud-ai-summit-2026-55-launches-to-take-ai-into-production-144298.md>)

Original publisher: [Read original article](<https://blog.ovhcloud.com/en/posts/ai-summit-2026-announcements/>)

Author: OVHcloud Team

Published: 2026-11-19T06:30:00Z

Content type: article

Language: en

Sources: [OVHcloud Blog](<https://devfeed.tech/sources/ovhcloud-blog.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [Amazon Bedrock AgentCore](<https://devfeed.tech/topics/amazon-bedrock-agentcore.md>), [model-deployment](<https://devfeed.tech/topics/model-deployment.md>), [The Three Main Cloud Service Models](<https://devfeed.tech/topics/the-three-main-cloud-service-models.md>), [Amazon Bedrock Knowledge Bases](<https://devfeed.tech/topics/amazon-bedrock-knowledge-bases.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>), [OpenAPI Specification](<https://devfeed.tech/topics/openapi.md>)

Tags: [17](<https://devfeed.tech/tags/17.md>), [2026](<https://devfeed.tech/tags/2026.md>), [accessible](<https://devfeed.tech/tags/accessible.md>), [act](<https://devfeed.tech/tags/act.md>), [actions](<https://devfeed.tech/tags/actions.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [agentic-platform](<https://devfeed.tech/tags/agentic-platform.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-act](<https://devfeed.tech/tags/ai-act.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [gpus](<https://devfeed.tech/tags/gpus.md>), [launches](<https://devfeed.tech/tags/launches.md>), [ovhcloud](<https://devfeed.tech/tags/ovhcloud.md>), [product-news](<https://devfeed.tech/tags/product-news.md>), [production](<https://devfeed.tech/tags/production.md>), [summit](<https://devfeed.tech/tags/summit.md>), [workspace](<https://devfeed.tech/tags/workspace.md>)

### AI overview

OVHcloud announced 55 launches at its 19 November 2026 AI Summit: 38 across cloud foundations and 17 spanning the AI stack. The excerpt emphasizes private networking and security features intended to support production workloads, alongside GPUs, orchestration, inference routing, agents, search, and applications. Olala 1.0 mini 7A1B is an open model with 7 billion parameters and 1 billion active per request; OVHcloud says it can run on an 8 GB GPU or a standard CPU and plans to release it under Apache 2.0. OVHai Workspace combines productivity, website creation, and company-data search; its free alpha launched in France with 5 GB per user, European data hosting, and a commitment not to use customer data for model training. These details describe the supplied excerpt, not necessarily all 55 launches.

### Source excerpt

Olala, OVHai Workspace, GPUs, and AI agents: discover the 55 launches from the OVHcloud AI Summit 2026, and the ones you can try right now.

## Improving Cross-Lingual Token Representations by Adding a Pinch of SALT

DevFeed: [Improving Cross-Lingual Token Representations by Adding a Pinch of SALT](<https://devfeed.tech/articles/improving-cross-lingual-token-representations-by-adding-a-pinch-of-salt-142559.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2609.09953>)

Author: Guillem Ram\\'irez

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Representation learning](<https://devfeed.tech/topics/representation-learning.md>), [Enhancing AI model performance](<https://devfeed.tech/topics/enhancing-ai-model-performance.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>)

Tags: [alignment](<https://devfeed.tech/tags/alignment.md>), [classification](<https://devfeed.tech/tags/classification.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [detection](<https://devfeed.tech/tags/detection.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [hallucination-detection](<https://devfeed.tech/tags/hallucination-detection.md>), [languages](<https://devfeed.tech/tags/languages.md>), [learning](<https://devfeed.tech/tags/learning.md>), [lightweight](<https://devfeed.tech/tags/lightweight.md>), [mining](<https://devfeed.tech/tags/mining.md>), [performance](<https://devfeed.tech/tags/performance.md>), [post-training](<https://devfeed.tech/tags/post-training.md>), [retrieval](<https://devfeed.tech/tags/retrieval.md>), [salt](<https://devfeed.tech/tags/salt.md>), [sequence](<https://devfeed.tech/tags/sequence.md>), [signal](<https://devfeed.tech/tags/signal.md>), [span](<https://devfeed.tech/tags/span.md>), [supervision](<https://devfeed.tech/tags/supervision.md>), [tasks](<https://devfeed.tech/tags/tasks.md>), [token](<https://devfeed.tech/tags/token.md>), [translation](<https://devfeed.tech/tags/translation.md>)

### AI overview

SALT improves token representations in cross-lingual sentence encoders by post-training them with span-level supervision. Across five multilingual token-level benchmarks, it achieves the best overall results on four, outperforming alternative fine-tuning strategies and competitive encoders. The abstract also reports gains on cross-lingual retrieval and classification, suggesting the method can benefit sentence-level performance too.

### Source excerpt

arXiv:2609.09953v2 Announce Type: replace-cross Abstract: Cross-lingual sentence encoders enable scalable transfer across hundreds of languages, powering applications such as translation mining and zero-shot learning in low-resource settings. Although trained for sentence-level alignment, they are increasingly also applied to token-level tasks such as hallucination detection and sequence tagging, exposing a mismatch between training and usage. We propose SALT, a lightweight post-training method that improves token representations by injecting span-level supervision into existing sentence encoders. Across five multilingual token-level benchmarks, SALT achieves the best overall results on four of them, outperforming alternative fine-tuning strategies and competitive encoders. It also improves sentence-level performance on cross-lingual retrieval and classification tasks. These results demonstrate that span-level supervision is an effective signal for improving both token and sentence representations.

## LusoClin: A Synthetic Clinical Note Dataset for Medical Information Retrieval in European Portuguese

DevFeed: [LusoClin: A Synthetic Clinical Note Dataset for Medical Information Retrieval in European Portuguese](<https://devfeed.tech/articles/lusoclin-a-synthetic-clinical-note-dataset-for-medical-information-retrieval-in-european-portuguese-142657.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23454>)

Author: Carla Teixeira Lopes, Daniel F\\'elix

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [information retrieval](<https://devfeed.tech/topics/information-retrieval.md>), [AI-driven information retrieval](<https://devfeed.tech/topics/ai-driven-information-retrieval.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [lora](<https://devfeed.tech/topics/lora.md>), [llama](<https://devfeed.tech/topics/llama.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [managing large-scale text data in AI](<https://devfeed.tech/topics/managing-large-scale-text-data-in-ai.md>), [Open Source Models & Datasets](<https://devfeed.tech/topics/open-source-models-datasets.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>)

Tags: [bert](<https://devfeed.tech/tags/bert.md>), [clinical-search](<https://devfeed.tech/tags/clinical-search.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [document-retrieval](<https://devfeed.tech/tags/document-retrieval.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [information-retrieval](<https://devfeed.tech/tags/information-retrieval.md>), [languages](<https://devfeed.tech/tags/languages.md>), [llama](<https://devfeed.tech/tags/llama.md>), [low-resource-languages](<https://devfeed.tech/tags/low-resource-languages.md>), [medical](<https://devfeed.tech/tags/medical.md>), [note](<https://devfeed.tech/tags/note.md>), [resource](<https://devfeed.tech/tags/resource.md>), [synthetic](<https://devfeed.tech/tags/synthetic.md>)

### AI overview

LusoClin is a publicly released dataset of synthetic clinical notes in European Portuguese, intended to support privacy-preserving research on clinical information retrieval in a language with limited access to real clinical text. It was created by preprocessing real clinical data, fine-tuning Llama 3.2 (3B) with Quantized Low-Rank Adaptation, and generating notes from prompts; a BERT-based model automatically annotates clinical entities for search and indexing. Evaluations using automatic metrics and human review indicate that the dataset captures relevant clinical language and may support document retrieval and domain-specific model adaptation. However, annotators can often distinguish its notes from real ones, and some medical inconsistencies remain.

### Source excerpt

arXiv:2611.23454v1 Announce Type: new Abstract: Access to large-scale clinical text collections is crucial for advancing medical information retrieval research, including clinical note search, patient cohort identification, and domain-adapted retrieval models. However, real clinical notes are rarely available due to privacy regulations, which particularly limits progress for low-resource languages such as European Portuguese. We present LusoClin, a publicly released dataset of synthetic clinical notes in European Portuguese, created to support privacy-preserving research on clinical text retrieval. LusoClin was generated by pre-processing real clinical data, fine-tuning the open-source Llama 3.2 (3B) instruction-tuned model using Quantized Low-Rank Adaptation, and applying prompt-based clinical note generation. To facilitate semantic indexing and clinical search, the notes are automatically annotated with clinical entities using a BERT-based model. We evaluate LusoClin in terms of utility, fidelity, diversity, and privacy using automatic metrics and human evaluation. Results indicate that the dataset captures clinically relevant language patterns and can support downstream IR applications such as clinical document retrieval and domain-specific model adaptation, although annotators can often distinguish synthetic from real notes and some medical inconsistencies persist.

## Cross-cultural Adaptation of Recipes for Low-Resource Languages: A Benchmark and Multilingual Culinary Lexicon for Hausa, Yor\\\`ub\\'a, and Igbo

DevFeed: [Cross-cultural Adaptation of Recipes for Low-Resource Languages: A Benchmark and Multilingual Culinary Lexicon for Hausa, Yor\\\`ub\\'a, and Igbo](<https://devfeed.tech/articles/cross-cultural-adaptation-of-recipes-for-low-resource-languages-a-benchmark-and-multilingual-culinary-lexicon-for-hausa-yor-ub-a-and-igbo-140069.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.22708>)

Author: Nuhu Ibrahim, Maryam Abdulazeez, Fatsuma Jauro, Salisu Aliyu, Robert Stevens, Riza Batista-Navarro

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Localization (l10n)](<https://devfeed.tech/topics/localization.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>)

Tags: [benchmark](<https://devfeed.tech/tags/benchmark.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [few-shot](<https://devfeed.tech/tags/few-shot.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [improvements](<https://devfeed.tech/tags/improvements.md>), [large-language-models-llms](<https://devfeed.tech/tags/large-language-models-llms.md>), [low-resource-languages](<https://devfeed.tech/tags/low-resource-languages.md>), [measurements](<https://devfeed.tech/tags/measurements.md>), [tools](<https://devfeed.tech/tags/tools.md>), [translation](<https://devfeed.tech/tags/translation.md>), [zero-shot](<https://devfeed.tech/tags/zero-shot.md>)

### AI overview

The study introduces ReciLoReL, a benchmark of 600 expert-translated, culturally adapted recipes in Yorùbá, Hausa, and Igbo, and ReciLex, a lexicon of about 15,000 translated culinary entities in ten categories. It evaluates English-to-language translation by multilingual LLMs in zero-shot and few-shot settings, including the effect of adding ReciLex without large target-language recipe corpora for training or fine-tuning. Across the selected models and languages, lexicon augmentation raises average BLEU from 13.57 to 14.51 and chrF from 37.89 to 38.74, and lowers TER from 77.14 to 75.12. Human ratings also improve by 0.11-0.36 points across five cultural-adaptation criteria on a five-point scale.

### Source excerpt

arXiv:2611.22708v1 Announce Type: cross Abstract: Recent multilingual large language models (LLMs) have improved machine translation, yet culturally adaptive translation for low-resource languages (LRLs) remains underexplored. This is particularly challenging for recipes, where ingredients, cooking tools, measurements, and procedures may require adaptation to the target cultural context. In this work, we introduce ReciLoReL, a benchmark of 600 expert-translated and culturally adapted recipes across Yor\`ub\'a, Hausa, and Igbo (three of the most prominent languages in Nigeria), and ReciLex, a multilingual culinary lexicon containing approximately 15,000 translated entities across ten recipe entity categories, including ingredients, tools, chef actions, state of tools, and durations. We evaluate state-of-the-art multilingual LLMs for English-to-LRL recipe translation under zero-shot and few-shot settings and investigate whether augmenting LLMs with ReciLex improves culturally adaptive recipe translation without requiring large target-language recipe corpora for training or fine-tuning. Beyond conventional reference-based translation evaluation using BLEU, chrF, and TER, we introduce five criteria for evaluating cultural adaptation, adapted from established cross-cultural adaptation methods. Across the selected LLMs and languages, ReciLex improves average BLEU from 13.57 to 14.51 and chrF from 37.89 to 38.74, while reducing TER from 77.14 to 75.12. Human evaluation further shows improvements ranging from 0.11 to 0.36 points across the five cultural adaptation evaluation criteria on a five-point scale. Overall, our work provides new resources, evaluation criteria, and empirical evaluation for studying culturally adaptive recipe translation in low-resource settings.

## Evaluating the Robustness of Fact-Checking LLMs on Multilingual Evidence

DevFeed: [Evaluating the Robustness of Fact-Checking LLMs on Multilingual Evidence](<https://devfeed.tech/articles/evaluating-the-robustness-of-fact-checking-llms-on-multilingual-evidence-137307.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.18623>)

Author: Radhika Shah

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Ground truth / benchmark quality](<https://devfeed.tech/topics/ground-truth-benchmark-quality.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>)

Tags: [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [corruption](<https://devfeed.tech/tags/corruption.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [fact-checking](<https://devfeed.tech/tags/fact-checking.md>), [llms](<https://devfeed.tech/tags/llms.md>), [robustness](<https://devfeed.tech/tags/robustness.md>)

### AI overview

The evaluation finds that fact-checking LLM verdict accuracy is highly vulnerable to corrupted evidence and weaker-resource languages. Using EuroVerdict claims in English, German, Greek, and Romanian, the authors created gold evidence sets and tested four corruption strategies at two severity levels. A misleading edit reduced Llama 3.1 8B's English accuracy from 80.00% to 16.84%, with losses widening as language resource levels fell. Translating Greek and Romanian claims into English recovered much, but not all, of the low-resource gap. Rubric-based LLM judge scores declined more slowly than verdict accuracy, suggesting they understate degradation. The authors caution against deploying these systems on noisy multilingual evidence and report releasing datasets and evaluation code.

### Source excerpt

arXiv:2611.18623v1 Announce Type: cross Abstract: Automated fact-checking benchmarks typically pair claims with clean, curated evidence in high-resource languages, but real-world evidence is often incomplete, misleading, or written in a language where models are weaker. We evaluate the robustness of fact-checking LLMs for different languages. Building on EuroVerdict, we construct gold evidence sets for 190 claims in each of English, German, Greek, and Romanian, then apply four corruption strategies at two severity levels. We test LLMs on a set of claims, providing them with evidence from the sets above. A rubric-based LLM-as-a-judge scores reasoning quality. Verdict accuracy proves highly sensitive to both factors: Misleading edit cut Llama 3.1 8B's English accuracy from 80.00% to 16.84%, and losses widen as language resource level decreases. Machine-translating Greek and Romanian claims into English recovers much, but not all, of the low-resource gap. Judge scores decline more slowly than accuracy, suggesting that automated reasoning evaluation underestimates degradation. These results caution against deploying LLM fact-checkers on noisy multilingual evidence, and we release our evidence datasets and evaluation code to support future work.

## Calibration and Conformal Prediction for Medical Question Answering in Arabic and English: Evidence from Two Non-Parallel Benchmarks

DevFeed: [Calibration and Conformal Prediction for Medical Question Answering in Arabic and English: Evidence from Two Non-Parallel Benchmarks](<https://devfeed.tech/articles/calibration-and-conformal-prediction-for-medical-question-answering-in-arabic-and-english-evidence-from-two-non-parallel-benchmarks-133151.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.16821>)

Author: Mariya (Stanley College of Engineering,Technology for Women, Hyderabad, India)

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [qwen](<https://devfeed.tech/topics/qwen.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>)

Tags: [arabic](<https://devfeed.tech/tags/arabic.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [conformal-prediction](<https://devfeed.tech/tags/conformal-prediction.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [english](<https://devfeed.tech/tags/english.md>), [medical](<https://devfeed.tech/tags/medical.md>), [question](<https://devfeed.tech/tags/question.md>)

### AI overview

On these two non-parallel medical-question benchmarks, both models' raw option probabilities were overconfident, while temperature scaling reduced expected calibration error by 82%-93%. Split conformal prediction met nominal coverage across all 24 tested settings, but at 90% coverage its answer sets averaged 63%-89% of the options, with larger sets for lower-accuracy results. Qwen2.5-7B-Instruct was more accurate than Jais-adapted-7B-chat in both benchmarks; confidence-based abstention also reduced Qwen's English error from 0.439 to 0.098 when answering only the most confident 20%. Jais strongly favored option C; scoring Latin letters only reduced that bias in Arabic, with little change to accuracy, calibration error or discrimination. Since the benchmarks differ in subject, level and number of options, the results do not isolate a language effect.

### Source excerpt

arXiv:2611.16821v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly tested on medical exam questions, but accuracy alone does not show when an answer can be trusted. I study the calibration and uncertainty of two open-weight models, Qwen2.5-7B-Instruct and Jais-adapted-7B-chat (both run in 4-bit precision), on 2,816 English multiple-choice questions from MedMCQA and 4,927 Arabic multiple-choice questions from MedAraBench. Accuracy was 0.561 (English) and 0.430 (Arabic) for Qwen, and 0.369 and 0.308 for Jais. Raw option probabilities were overconfident in all four settings (expected calibration error 0.242 to 0.486), and a single temperature parameter removed 82% to 93% of that error. Split conformal prediction reached its nominal coverage in all 24 settings I tested (within 0.006), but the sets were large when accuracy was low: at 90% coverage, temperature-scaled sets contained on average 63% (Qwen, English), 78% (Qwen, Arabic), 86% (Jais, English) and 89% (Jais, Arabic) of the options. The margin between the top two option probabilities separated correct from incorrect answers best for Qwen on English (AUROC 0.734) and worst for Jais on Arabic (0.579). Answering only the 20% most confident questions lowered Qwen's English error from 0.439 to 0.098. Jais showed a strong prior for option C (49% of English and 60% of Arabic predictions). Scoring Latin letters only cut the Arabic share to 29% but left accuracy, raw ECE and AUROC nearly unchanged. Because the two benchmarks differ in subject, level and number of options, I do not attribute differences between languages to language itself. Code is available at https://github.com/Mariyaaleem/arabic-english-medqa-uncertainty.

## How We Built a Self-Correcting Localization System for OpenRouter

DevFeed: [How We Built a Self-Correcting Localization System for OpenRouter](<https://devfeed.tech/articles/how-we-built-a-self-correcting-localization-system-for-openrouter-130215.md>)

Original publisher: [Read original article](<https://openrouter.ai/blog/announcements/openrouter-in-8-languages/>)

Author: Robert Sun

Published: 2026-11-12T00:00:00Z

Content type: article

Language: en

Sources: [OpenRouter](<https://devfeed.tech/sources/openrouter.md>)

Topics: [Localization (l10n)](<https://devfeed.tech/topics/localization.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>), [human review](<https://devfeed.tech/topics/human-review.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [api](<https://devfeed.tech/tags/api.md>), [code](<https://devfeed.tech/tags/code.md>), [code-checks](<https://devfeed.tech/tags/code-checks.md>), [costs](<https://devfeed.tech/tags/costs.md>), [languages](<https://devfeed.tech/tags/languages.md>), [localization](<https://devfeed.tech/tags/localization.md>), [openrouter](<https://devfeed.tech/tags/openrouter.md>), [pipeline](<https://devfeed.tech/tags/pipeline.md>), [reports](<https://devfeed.tech/tags/reports.md>), [system](<https://devfeed.tech/tags/system.md>), [translation](<https://devfeed.tech/tags/translation.md>)

### AI overview

OpenRouter's localization pipeline keeps translations current by fingerprinting source strings and processing changed entries daily. It sends batches of up to 40 to GPT-6.1 Sol with style notes, glossary rules, and UI context; code checks placeholders, markup, whitespace, and terms, then a second call verifies meaning. Rejections return with reasons for retries; strings that still fail remain in English and are flagged. Daily pull requests receive AI review and a human merge. In a 454-case comparison, the full pipeline reduced authored check failures from 147 to 4 and outside judges counted 109 critical errors versus 203 for a bare prompt, though overall preference favored the pipeline narrowly (139 to 117), and back-translation scores were treated as a tie. The cheaper GPT-6 Luna model had lower first-try accuracy (91.7% versus 99.2%) at about one-third the cost on the 254-case comparison. Reader reports and reviewer-found defects become glossary rules, style notes, overrides, or regression cases. A full translation and verification of today's 16,918 strings into seven languages is estimated at about $460; daily costs vary with changed strings.

### Source excerpt

OpenRouter now speaks 8 languages. Here is the translation pipeline behind it, what it costs, and how reader reports become fixes.

## The Indic Token Tax in 2026: Measuring Tokenizer. Disparity and Its Cost for the Languages of India

DevFeed: [The Indic Token Tax in 2026: Measuring Tokenizer. Disparity and Its Cost for the Languages of India](<https://devfeed.tech/articles/the-indic-token-tax-in-2026-measuring-tokenizer-disparity-and-its-cost-for-the-languages-of-india-128248.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.13058>)

Author: Johin Johny Arimbur, Ankur Ashit Ghosh, Shervin Sam

Published: 2026-11-11T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Language models](<https://devfeed.tech/topics/language-models.md>), [context window](<https://devfeed.tech/topics/context-window.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [india](<https://devfeed.tech/tags/india.md>), [indic](<https://devfeed.tech/tags/indic.md>), [languages](<https://devfeed.tech/tags/languages.md>), [measuring](<https://devfeed.tech/tags/measuring.md>), [tax](<https://devfeed.tech/tags/tax.md>), [token](<https://devfeed.tech/tags/token.md>)

### AI overview

Across 25 Indian languages, tokenizers often require substantially more tokens than English for equivalent text, increasing potential API cost and reducing the text that fits in a context window. The study compares nine public tokenizers using aligned FLORES-200 translations and Hinglish-English social media pairs. For GPT-4o, token counts range from 1.57 times English for Hindi to 5.01 times for Odia and 13.7 times for Santali in Ol Chiki; GPT-4 reaches up to 12.8 times. Sarvam-30B/105B is the only third-party tokenizer reported below 2.6 times across every language, with a median of 1.53. An illustrative GPT-4o support-bot estimate puts 100,000 monthly chats at INR 56,640 in English versus INR 102,829 in Marathi, based on stated assumptions and November 5, 2026 prices, not production measurements. On a Hindi and Marathi morphology set, no tokenizer aligns more than 35% of word-internal boundaries with morpheme boundaries (boundary F1 no higher than 0.48).

### Source excerpt

arXiv:2611.13058v1 Announce Type: cross Abstract: Commercial language model APIs bill per token, so a tokenizer that needs more tokens for the same content in one language makes that language more expensive, slower, and able to fit less text in a fixed context window. We measure this disparity for nine public tokenizers, including those of GPT-4, GPT-4o, Qwen2.5, DeepSeek-V3, Mistral-Nemo and four Sarvam models, on all 25 languages of India available in FLORES-200 (2,009 professionally translated sentences aligned with English) and on 13,732 human-translated Hinglish-English social media pairs from PHINC. On GPT-4o's tokenizer the same content needs between 1.57 times (Hindi) and 5.01 times (Odia) the tokens of English, and 13.7 times for Santali in Ol Chiki script; GPT-4's tokenizer needs up to 12.8 times. Sarvam-30B/105B is the only third-party tokenizer that stays below 2.6 times on every language (median 1.53). In an illustrative scenario using official list prices accessed on 5 November 2026 and stated assumptions, a support bot handling 100,000 chats a month on GPT-4o would cost about INR 56,640 in English and INR 102,829 in Marathi; these are estimates, not measured production costs. Against an openly licensed Hindi and Marathi morphological segmentation set, no tokenizer places more than 35% of its word-internal boundaries on morpheme boundaries (boundary F1 at most 0.48). In a companion paper we present a tokenizer that reduces this tax. Code, pinned artifacts and raw token counts are available at https://github.com/HavonAI/indic-token-tax.

## Candidate Statement for the 2026 W3C TAG Election

DevFeed: [Candidate Statement for the 2026 W3C TAG Election](<https://devfeed.tech/articles/candidate-statement-for-the-2026-w3c-tag-election-129507.md>)

Original publisher: [Read original article](<https://lea.verou.me/blog/2026/tag/>)

Author: Lea Verou

Published: 2026-11-11T00:00:00Z

Content type: opinion

Language: en

Sources: [Lea Verou's blog](<https://devfeed.tech/sources/lea-verou-s-blog.md>)

Topics: [W3C](<https://devfeed.tech/topics/w3c.md>), [Web platform](<https://devfeed.tech/topics/web-platform.md>), [human computer interaction](<https://devfeed.tech/topics/human-computer-interaction.md>), [Web Accessibility (a11y)](<https://devfeed.tech/topics/web-accessibility.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>), [inclusive design](<https://devfeed.tech/topics/inclusive-design.md>), [model architecture](<https://devfeed.tech/topics/model-architecture.md>), [engineering-leadership](<https://devfeed.tech/topics/engineering-leadership.md>), [Model Routing](<https://devfeed.tech/topics/model-routing.md>)

Tags: [2020](<https://devfeed.tech/tags/2020.md>), [2021](<https://devfeed.tech/tags/2021.md>), [2022](<https://devfeed.tech/tags/2022.md>), [2025](<https://devfeed.tech/tags/2025.md>), [2026](<https://devfeed.tech/tags/2026.md>), [academic](<https://devfeed.tech/tags/academic.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [award](<https://devfeed.tech/tags/award.md>), [blog](<https://devfeed.tech/tags/blog.md>), [building](<https://devfeed.tech/tags/building.md>), [candidate](<https://devfeed.tech/tags/candidate.md>), [design](<https://devfeed.tech/tags/design.md>), [design-principles](<https://devfeed.tech/tags/design-principles.md>), [election](<https://devfeed.tech/tags/election.md>), [entrepreneurship](<https://devfeed.tech/tags/entrepreneurship.md>), [foundation](<https://devfeed.tech/tags/foundation.md>), [google](<https://devfeed.tech/tags/google.md>), [human-computer-interaction](<https://devfeed.tech/tags/human-computer-interaction.md>), [impact](<https://devfeed.tech/tags/impact.md>), [personal](<https://devfeed.tech/tags/personal.md>), [statement](<https://devfeed.tech/tags/statement.md>), [w3c](<https://devfeed.tech/tags/w3c.md>), [web-standards](<https://devfeed.tech/tags/web-standards.md>)

### AI overview

Lea seeks to return to the W3C Technical Architecture Group (TAG), arguing it should shape Web architecture proactively, not mainly review features after their design is set or they ship. She proposes identifying cross-group gaps, sketching coherent solutions, and bringing relevant working groups into the work without taking over their responsibilities. In her prior two terms (2021-2025), she co-led Web Platform Design Principles, started the Web Platform Gaps initiative, liaised with TC39, and took part in hundreds of design reviews. She says her work across CSS, TC39, WHATWG, web development, and open source gives her broad platform experience. She also urges the TAG to assess AI-written code and agent use of the Web through evidence and architectural principles, while upholding user agency, privacy, accessibility, and interoperability. Google, Microsoft, and the OpenJS Foundation have pooled resources to fund her participation if elected; she says they do not direct or veto her positions.

### Source excerpt

Group photo of the W3C TAG in July 2022 with me to the left of Sir Tim Berners-Lee (inventor of the Web and TAG chair emeritus). Photo credit: Dan Appelquist I had the great honor of being elected to the TAG by the W3C membership in 2020 and again in 2022, serving two terms from 2021 to 2025. After a two-year break, this year I'm running to return, thanks to Google, Microsoft, and OpenJS Foundation pooling resources to make it possible. 💙 The W3C Technical Architecture Group (TAG) is the Working Group that ensures that Web Platform technologies are usable and follow consistent design principles, whether they are created inside or outside W3C. It advocates for the needs of everyone who uses the Web and everyone who works on the Web. If you work for a company that is a W3C Member, please consider encouraging your AC rep to vote for me! (Ask me if you don't know who that is) My candidate statement follows. I'm Lea, and I'm running to return to the TAG, where I served two terms from 2021 to 2025. It was some of the most fulfilling work of my career. The only reason I did not run in 2024 is that the job takes hundreds of hours a year, which as an independent I could not sustain pro bono. This year, in recognition of my service, Google, Microsoft, and the OpenJS Foundation have pooled resources to fully fund my participation, so I can finally commit the time the job deserves while remaining independent. My work spans the entire web platform: 14 years in the CSS Working Group, active participation in TC39 and WHATWG, two decades of building for the web, and open source running on millions of websites. In 2025 the OpenJS Foundation recognized my overall impact on the web platform with its Pathfinder for Standards Award. That breadth is exactly what the TAG requires. With a PhD in Human-Computer Interaction from MIT and a 20+ year career that includes entrepreneurship and product leadership, I combine academic rigor and big-picture perspective with pragmatism, an eye for har

## Otter AI can now follow conversations across 30+ languages

DevFeed: [Otter AI can now follow conversations across 30+ languages](<https://devfeed.tech/articles/otter-ai-can-now-follow-conversations-across-30-languages-125009.md>)

Original publisher: [Read original article](<https://techcrunch.com/2026/11/09/otter-ai-can-now-follow-conversations-across-30-languages/>)

Author: Lauren Forristal

Published: 2026-11-10T07:00:00Z

Content type: news

Language: en

Sources: [TechCrunch | AI News & Artificial Intelligence](<https://devfeed.tech/sources/techcrunch-ai-news-artificial-intelligence.md>)

Topics: [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>), [Conversational AI](<https://devfeed.tech/topics/conversational-ai.md>)

Tags: [added](<https://devfeed.tech/tags/added.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-tools](<https://devfeed.tech/tags/ai-tools.md>), [announcement](<https://devfeed.tech/tags/announcement.md>), [automatically](<https://devfeed.tech/tags/automatically.md>), [conversations](<https://devfeed.tech/tags/conversations.md>), [feature](<https://devfeed.tech/tags/feature.md>), [global](<https://devfeed.tech/tags/global.md>), [languages](<https://devfeed.tech/tags/languages.md>), [meetings](<https://devfeed.tech/tags/meetings.md>), [models](<https://devfeed.tech/tags/models.md>), [otter](<https://devfeed.tech/tags/otter.md>), [otter-ai](<https://devfeed.tech/tags/otter-ai.md>), [support](<https://devfeed.tech/tags/support.md>), [switch](<https://devfeed.tech/tags/switch.md>), [transcription](<https://devfeed.tech/tags/transcription.md>)

### AI overview

Otter.ai is expanding to 31 supported languages and plans a feature that automatically detects when speakers switch languages during a meeting, continuing transcription without users changing settings. Thirteen of the 27 added languages are available now; 14 are coming soon. The new languages use Otter's own speech models, trained on publicly available datasets and reviewed by native-speaking volunteers for transcript accuracy. The article describes the language-switching feature as upcoming.

### Source excerpt

Otter AI just added support for 27 languages and revealed an upcoming feature that can follow along if speakers switch between languages.

## The Narrative Gap: Can LLMs Help Us Navigate Diverse Narratives Across Languages?

DevFeed: [The Narrative Gap: Can LLMs Help Us Navigate Diverse Narratives Across Languages?](<https://devfeed.tech/articles/the-narrative-gap-can-llms-help-us-navigate-diverse-narratives-across-languages-126305.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.12750>)

Author: Nikhil Sharma, Kelly Marchisio, Kenton Murray, Ziang Xiao

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>)

Tags: [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-hc](<https://devfeed.tech/tags/cs-hc.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [diversity](<https://devfeed.tech/tags/diversity.md>), [languages](<https://devfeed.tech/tags/languages.md>), [large-language-models-llms](<https://devfeed.tech/tags/large-language-models-llms.md>), [llms](<https://devfeed.tech/tags/llms.md>)

### AI overview

The authors introduce NarrativeBench to assess whether language models help people navigate competing accounts across languages. It contains 1,333 news articles in 23 publication languages covering 23 contested topics, plus 460 Reddit-grounded queries; 14 LLMs were evaluated across 24 query languages. The paper defines six capabilities: including and excluding narratives, identifying missing narratives, presenting diverse narratives, retaining coverage under confirmatory queries, and maintaining parity across languages. The evaluation finds that current models do not reliably provide these capabilities, with gaps in competing perspectives even when diversity is requested and access varying by query framing and language. The authors say these disparities can risk echo chambers and linguistic filter bubbles; the benchmark is intended to support systems that give users more agency over the narratives they encounter.

### Source excerpt

arXiv:2611.12750v1 Announce Type: cross Abstract: Large language models (LLMs) increasingly mediate how people seek and consume information, including accounts that differ across languages and regions, especially on contested topics. Supporting diverse information seeking requires models to help users navigate competing narratives across languages and information streams. We define six capabilities necessary for diverse multilingual information seeking: three foundational capabilities for narrative inclusion, narrative exclusion, and missing narrative identification; and three higher-level capabilities for presenting diverse narratives, preserving narrative coverage under confirmatory queries, and maintaining information parity across languages. To guide the development of those information seeking capabilities, we introduce \textsc{NarrativeBench}, a benchmark comprising 1,333 news articles in 23 publication languages across 23 contested topics and 460 Reddit-grounded queries. Evaluating 14 LLMs across 24 query languages, we find that current models lack the capabilities needed to reliably navigate diverse narratives, limiting users' control over the perspectives they receive and leaving substantial gaps in exposure to competing accounts even when diversity is explicitly requested. Information access also differs across query framings and languages, risking echo chambers, linguistic filter bubbles, and information disparities that weaken the common ground needed for democratic discourse. \textsc{NarrativeBench} provides a foundation for developing information systems that give users across languages and regions agency to navigate diverse narratives.

## What Travels Across Languages? Reading, Transfer and Erasure of Cross-Lingual Concept Subspaces in Bangla Idioms

DevFeed: [What Travels Across Languages? Reading, Transfer and Erasure of Cross-Lingual Concept Subspaces in Bangla Idioms](<https://devfeed.tech/articles/what-travels-across-languages-reading-transfer-and-erasure-of-cross-lingual-concept-subspaces-in-bangla-idioms-125373.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11386>)

Author: Syed Rifat Raiyan, Yousuf Ibne Kamal Niloy

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>)

Tags: [analysis](<https://devfeed.tech/tags/analysis.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [class](<https://devfeed.tech/tags/class.md>), [classifiers](<https://devfeed.tech/tags/classifiers.md>), [controls](<https://devfeed.tech/tags/controls.md>), [covariance](<https://devfeed.tech/tags/covariance.md>), [cross](<https://devfeed.tech/tags/cross.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [discovery](<https://devfeed.tech/tags/discovery.md>), [english](<https://devfeed.tech/tags/english.md>), [estimator](<https://devfeed.tech/tags/estimator.md>), [filter](<https://devfeed.tech/tags/filter.md>), [gaussian](<https://devfeed.tech/tags/gaussian.md>), [information](<https://devfeed.tech/tags/information.md>), [input](<https://devfeed.tech/tags/input.md>), [languages](<https://devfeed.tech/tags/languages.md>), [layers](<https://devfeed.tech/tags/layers.md>), [logistic-regression](<https://devfeed.tech/tags/logistic-regression.md>), [map](<https://devfeed.tech/tags/map.md>), [meaning](<https://devfeed.tech/tags/meaning.md>), [reading](<https://devfeed.tech/tags/reading.md>), [stat-ml](<https://devfeed.tech/tags/stat-ml.md>)

### AI overview

Across six frozen encoders, Bangla idiom meaning can be detected across languages, but successful reading does not guarantee that an English-derived direction can erase the concept in Bangla. Class-mean directions beat three null baselines in all 12 post hoc model-label tests, while a prospective re-split supported idiom-alone sharing in 10 of 12 tests. In the original confirmatory logistic-regression family, both planned nulls were rejected in only 4 of 12 tests, with English directions gaining 3.0-7.3 accuracy points over the majority-class floor. Transfer varied with input mapping and regularization: English-trained class-mean readouts beat zero-shot nulls in 8 of 12 discovery tests, and explicit glosses showed more coarse-class sharing than idiom counterparts. Erasure was asymmetric: English-fitted LEACE removed under one percentage point of Bangla figurative-class accuracy, versus 4-11 points for Bangla-fitted LEACE; INLP showed a similar pattern. The authors argue that reading, transfer, erasure and overlap should be evaluated separately.

### Source excerpt

arXiv:2611.11386v1 Announce Type: cross Abstract: Cross-lingual concept analysis requires distinguishing the information that a subspace retains from the classifiers and erasers that transfer through it. We study these operations in six frozen encoders using 10,359 Bangla (Bengali) idiom senses, eight input conditions, and figurative-class and sentiment labels. Class-mean directions provide the most consistent evidence of shared coarse meaning: in post hoc analyses, they outperform random-subspace, permuted-label and covariance-matched nulls in all 12 model-label tests under balanced accuracy. A prospective internal re-split supports idiom-alone sharing in 10 of 12 tests, spanning five encoders. The original confirmatory logistic-regression family rejects both planned nulls in 4 of 12 tests, where English directions gain only a few accuracy points above the majority-class floor (3.0-7.3). Classifier transfer depends on the input map and regularization: English-trained class-mean readouts beat both zero-shot nulls in 8 of 12 discovery tests, and stronger source-selected penalties improve transfer through logistic and LEACE directions. These readouts use Bangla-validation-selected layers and unlabeled target preprocessing. Input controls show greater coarse-class sharing for explicit glosses than for idiom counterparts. Reading, however, does not imply effective erasure: English-fitted LEACE removes less than one percentage point of Bangla figurative-class accuracy, compared with 4-11 points for Bangla-fitted LEACE; INLP shows a similar asymmetry. Pattern-filter analysis, covariance-weighted input maps and a binary Gaussian overlap bound explain how these operations can diverge. The analogous multiclass ordering is an empirical diagnostic, observed in 11 of 12 cells. The results support an estimator-sensitive account of cross-lingual concept sharing and motivate evaluating reading, classifier transfer, erasure and overlap separately.

## Small Tweaks, Great Output: What Merchants Can Do to Maximize Pay By Bank Conversion

DevFeed: [Small Tweaks, Great Output: What Merchants Can Do to Maximize Pay By Bank Conversion](<https://devfeed.tech/articles/small-tweaks-great-output-what-merchants-can-do-to-maximize-pay-by-bank-conversion-119763.md>)

Original publisher: [Read original article](<https://hackernoon.com/small-tweaks-great-output-what-merchants-can-do-to-maximize-pay-by-bank-conversion?source=rss>)

Author: Noda

Published: 2026-11-05T22:01:20Z

Content type: tutorial

Language: en

Sources: [HackerNoon](<https://devfeed.tech/sources/hackernoon.md>)

Topics: [User interface design](<https://devfeed.tech/topics/ui-design.md>), [inclusive design](<https://devfeed.tech/topics/inclusive-design.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>)

Tags: [accessibility](<https://devfeed.tech/tags/accessibility.md>), [banks](<https://devfeed.tech/tags/banks.md>), [checkout](<https://devfeed.tech/tags/checkout.md>), [checkout-conversion](<https://devfeed.tech/tags/checkout-conversion.md>), [conversion](<https://devfeed.tech/tags/conversion.md>), [good-company](<https://devfeed.tech/tags/good-company.md>), [open-banking](<https://devfeed.tech/tags/open-banking.md>), [pay-by-bank](<https://devfeed.tech/tags/pay-by-bank.md>), [pay-by-bank-explained](<https://devfeed.tech/tags/pay-by-bank-explained.md>), [payment-processing](<https://devfeed.tech/tags/payment-processing.md>), [practical](<https://devfeed.tech/tags/practical.md>), [ux](<https://devfeed.tech/tags/ux.md>), [ux-design](<https://devfeed.tech/tags/ux-design.md>), [what-is-pay-by-bank](<https://devfeed.tech/tags/what-is-pay-by-bank.md>)

### AI overview

The article says Pay by Bank uptake can improve when checkout makes the option clear, trustworthy and easy to find. It recommends labeling it "Pay by Bank," showing familiar bank logos so users can recognize supported institutions, and placing it visibly--first where awareness is high, or second where it is still growing. Avoid competing primary buttons and keep individual bank choices to a short preview (the article cites five outperforming nine), with the rest behind "show more." Match button colors and typography to the site and meet WCAG readability guidance. These are presented as practical, low-investment changes; the article also notes that routing solutions such as Noda's BIN-to-Bank can add Pay by Bank through a card form without checkout UI changes.

### Source excerpt

Practical checkout UX tips from Noda's product designer. Learn 5 simple, tested checkout tweaks merchants can make to turn open banking into the obvious choice.

## New LibreOffice "Get involved" flyers for handing out (and translations!)

DevFeed: [New LibreOffice "Get involved" flyers for handing out (and translations!)](<https://devfeed.tech/articles/new-libreoffice-get-involved-flyers-for-handing-out-and-translations-119143.md>)

Original publisher: [Read original article](<https://blog.documentfoundation.org/blog/2026/11/05/new-libreoffice-get-involved-flyers-for-handing-out-and-translations/>)

Author: Mike Saunders

Published: 2026-11-05T07:15:29Z

Content type: release

Language: en

Sources: [TDF Community](<https://devfeed.tech/sources/tdf-community-blog.md>)

Topics: [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>), [Publishing](<https://devfeed.tech/topics/publishing.md>)

Tags: [community](<https://devfeed.tech/tags/community.md>), [events](<https://devfeed.tech/tags/events.md>), [get-involved](<https://devfeed.tech/tags/get-involved.md>), [libreoffice](<https://devfeed.tech/tags/libreoffice.md>), [marketing](<https://devfeed.tech/tags/marketing.md>), [translations](<https://devfeed.tech/tags/translations.md>)

### AI overview

LibreOffice has updated its "Get involved" flyer for events where attendees already know the software and may be interested in joining its community. It highlights ways to contribute beyond programming. Volunteers planning to hand out flyers can contact the marketing team through its mailing list or Telegram group to arrange printed copies. To translate the flyer, download the English original, edit it in an SVG editor such as Inkscape, then upload the translation to the wiki and email the team a link or attach the file.

### Source excerpt

LibreOffice's marketing team helps spread the word about the software (and the community that makes it). It produces flyers that can be translated, printed and handed out at events - and now we have an updated design of the "Get involved" flyer. This flyer ideal for distributing at events where

## How to manage multi-region websites and localization at scale

DevFeed: [How to manage multi-region websites and localization at scale](<https://devfeed.tech/articles/how-to-manage-multi-region-websites-and-localization-at-scale-127689.md>)

Original publisher: [Read original article](<https://www.webflow.com/blog/how-to-manage-a-multi-region-website>)

Author: Webflow Team

Published: 2026-11-04T00:00:00Z

Content type: tutorial

Language: en

Sources: [Webflow Blog](<https://devfeed.tech/sources/webflow-blog.md>)

Topics: [Localization (l10n)](<https://devfeed.tech/topics/localization.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>), [webflow](<https://devfeed.tech/topics/webflow.md>)

Tags: [adapting](<https://devfeed.tech/tags/adapting.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [audit](<https://devfeed.tech/tags/audit.md>), [automatic](<https://devfeed.tech/tags/automatic.md>), [automatically](<https://devfeed.tech/tags/automatically.md>), [avoid](<https://devfeed.tech/tags/avoid.md>), [backups](<https://devfeed.tech/tags/backups.md>), [balance](<https://devfeed.tech/tags/balance.md>), [behind-the-scenes](<https://devfeed.tech/tags/behind-the-scenes.md>), [branding](<https://devfeed.tech/tags/branding.md>), [content](<https://devfeed.tech/tags/content.md>), [development](<https://devfeed.tech/tags/development.md>), [how-to](<https://devfeed.tech/tags/how-to.md>), [languages](<https://devfeed.tech/tags/languages.md>), [localization](<https://devfeed.tech/tags/localization.md>), [manage](<https://devfeed.tech/tags/manage.md>), [multi-region](<https://devfeed.tech/tags/multi-region.md>), [region](<https://devfeed.tech/tags/region.md>), [scale](<https://devfeed.tech/tags/scale.md>), [websites](<https://devfeed.tech/tags/websites.md>)

### AI overview

Managing regional websites at scale requires adapting content and workflows beyond literal translation while keeping the brand voice consistent. The excerpt recommends auditing what must stay consistent versus vary by market, adapting copy, design, imagery, currency, and metadata, then having regional reviewers check pages before publication. For international SEO, it advises using crawlable, consistently structured regional URLs, correctly reciprocal hreflang tags, localized keywords and metadata, and avoiding redirects based only on IP address or browser language. A shared CMS can support reuse with regional variation, role-based permissions and approvals, and integrated localization or translation management. The excerpt presents Webflow Localization as one example of a platform that handles content, design, and SEO settings for multiple locales in one project.

### Source excerpt

Explore how to manage multi-region websites at scale by translating languages, localizing content, and keeping branding consistent for regional audiences.

## A Token Is Not a Unit of Difficulty: Segmentation Controls Where Adaptive Computation Goes Inside a Word

DevFeed: [A Token Is Not a Unit of Difficulty: Segmentation Controls Where Adaptive Computation Goes Inside a Word](<https://devfeed.tech/articles/a-token-is-not-a-unit-of-difficulty-segmentation-controls-where-adaptive-computation-goes-inside-a-word-115115.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.00983>)

Author: Dipto Sumit, Mahima Mahzabin Sandria, Farig Sadeque, Md. Golam Rabiul Alam

Published: 2026-11-03T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [AI Chat](<https://devfeed.tech/topics/ai-chat.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>), [reproducibility](<https://devfeed.tech/topics/reproducibility.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [analysis](<https://devfeed.tech/tags/analysis.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [bloom](<https://devfeed.tech/tags/bloom.md>), [budget](<https://devfeed.tech/tags/budget.md>), [confidence](<https://devfeed.tech/tags/confidence.md>), [continuation](<https://devfeed.tech/tags/continuation.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [english](<https://devfeed.tech/tags/english.md>), [exit](<https://devfeed.tech/tags/exit.md>), [gap](<https://devfeed.tech/tags/gap.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [languages](<https://devfeed.tech/tags/languages.md>), [layers](<https://devfeed.tech/tags/layers.md>), [models](<https://devfeed.tech/tags/models.md>), [offline](<https://devfeed.tech/tags/offline.md>), [split](<https://devfeed.tech/tags/split.md>), [token](<https://devfeed.tech/tags/token.md>), [tokenizers](<https://devfeed.tech/tags/tokenizers.md>), [transformer](<https://devfeed.tech/tags/transformer.md>), [word](<https://devfeed.tech/tags/word.md>)

### AI overview

The analysis finds that tokenizer segmentation changes where confidence-based early-exit policies spend computation within words. In Qwen3, continuation pieces receive less computation than first pieces in English, German, and Turkish, but more in Bangla, across both spiking-conversion timesteps and Transformer layers. Changing segmentation with BPE-dropout shifts this gap by up to eight exits; training small models with two tokenizers shifts Bangla's gap by six, while BLOOM and mGPT, which split Bangla less finely, show the English pattern. Characters per token do not predict the cross-language gap, whereas within-word surprisal explains 75% of its variance. A learned stopping head improves on confidence stopping by 7-8 points, but improves by less than one point over the best equal-cost mix of fixed exits and sometimes performs worse. These are offline analyses; the excerpt reports no measured speedups.

### Source excerpt

arXiv:2611.00983v1 Announce Type: cross Abstract: Early-exit language models decide for each token how much computation to spend, usually by stopping once the prediction is confident. Yet a token is not a fixed unit: tokenizers cut words into different numbers of pieces in different languages. We ask where inside a word such a policy spends its budget. In Qwen3, along two adaptive axes (the timesteps of a spiking conversion and the layers of the original Transformer), confidence stopping gives the continuation pieces of a word less computation than its first piece in English, German, and Turkish, but more in Bangla. Segmentation drives this. Re-segmenting the same parallel passages with BPE-dropout shifts the gap by up to eight exits on both axes, small models trained under two tokenizers shift Bangla's gap by six exits, and BLOOM and mGPT, which split Bangla less finely, give it the English pattern. Characters per token do not predict the gap across languages; within-word surprisal explains 75% of its variance. A learned stopping head beats confidence stopping by 7 to 8 points, but gains less than one point over the best mix of fixed exits at equal cost, and sometimes loses. All results come from offline analysis, not measured speedups.

## IndicDBBench - Benchmarking Multilingual Text-to-SQL Capabilities in Indian Languages

DevFeed: [IndicDBBench - Benchmarking Multilingual Text-to-SQL Capabilities in Indian Languages](<https://devfeed.tech/articles/indicdbbench-benchmarking-multilingual-text-to-sql-capabilities-in-indian-languages-113629.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2604.13686>)

Author: Aviral Dawar, Roshan Karanth, Vikram Goyal, Rakhee Kulshrestha, Sumanta Pasari, Dhruv Kumar

Published: 2026-11-03T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Databases](<https://devfeed.tech/sources/arxiv-databases.md>)

Topics: [text2sql](<https://devfeed.tech/topics/text2sql.md>), [AI for SQL query generation](<https://devfeed.tech/topics/ai-for-sql-query-generation.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [SQL](<https://devfeed.tech/topics/sql.md>), [sql-database](<https://devfeed.tech/topics/sql-database.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>), [JOIN](<https://devfeed.tech/topics/join.md>)

Tags: [benchmarking](<https://devfeed.tech/tags/benchmarking.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-db](<https://devfeed.tech/tags/cs-db.md>), [database](<https://devfeed.tech/tags/database.md>), [databases](<https://devfeed.tech/tags/databases.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [deepseek](<https://devfeed.tech/tags/deepseek.md>), [deepseek-v3](<https://devfeed.tech/tags/deepseek-v3.md>), [english](<https://devfeed.tech/tags/english.md>), [gap](<https://devfeed.tech/tags/gap.md>), [government](<https://devfeed.tech/tags/government.md>), [grounding](<https://devfeed.tech/tags/grounding.md>), [idp](<https://devfeed.tech/tags/idp.md>), [indian](<https://devfeed.tech/tags/indian.md>), [indic](<https://devfeed.tech/tags/indic.md>), [join](<https://devfeed.tech/tags/join.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [languages](<https://devfeed.tech/tags/languages.md>), [large-language-models-llms](<https://devfeed.tech/tags/large-language-models-llms.md>), [models](<https://devfeed.tech/tags/models.md>), [sql](<https://devfeed.tech/tags/sql.md>)

### AI overview

IndicDBBench evaluates text-to-SQL across English and six Indic languages using 15,612 tasks over 20 databases and 237 tables derived from NDAP and IDP data. Its three-agent pipeline constructs relational schemas from denormalized government data, with up to six-deep joins; tasks use real database values and require multi-table reasoning. Tests of DeepSeek V3.2, MiniMax M2.7, Llama 3.3, and Qwen3 found execution accuracy 9 points lower in Indic languages than in English. For the two models tested with extra grounding, supplying gold tables and literal values improved accuracy similarly across languages, leaving the gap in place. The supplied abstract says the dataset and code are publicly available.

### Source excerpt

arXiv:2604.13686v2 Announce Type: replace-cross Abstract: While Large Language Models (LLMs) have significantly advanced Text-to-SQL performance, existing benchmarks largely focus on Western contexts leaving a gap in real-world, non-Western applications. We introduce IndicDBBench, a comprehensive multilingual Text-to-SQL benchmark designed to evaluate cross-lingual semantic parsing across diverse Indic language families. The underlying schemas are derived from real-world open-data platforms (NDAP and IDP), ensuring realistic structural complexity. IndicDBBench comprises 20 databases spanning 237 tables. To construct complex relational structures from denormalized government data, we employ a three-agent pipeline (Architect, Auditor, Refiner), achieving high relational density (11.85 tables per database) and join depths of up to six. We generate 15,612 tasks spanning seven languages: English and six Indic languages, ensuring that queries are grounded in real database values and require multi-table reasoning. We evaluate state-of-the-art models, including DeepSeek V3.2, MiniMax M2.7, Llama 3.3, and Qwen3, across seven linguistic variants. Our results reveal a 9.0-point drop in execution accuracy from English to Indic languages, exposing an "Indic Gap" that, for the two models tested, persists even when gold tables and literal values are supplied, since this grounding raises accuracy in every language by similar amounts. The dataset and code are publicly available at https://anonymous.4open.science/r/multilingualText2Sql-Indic--DDCC/

## PlagiaCorpus: A Bilingual, Adversarial, and Cross-Lingual Dataset for Forensic Plagiarism Detection

DevFeed: [PlagiaCorpus: A Bilingual, Adversarial, and Cross-Lingual Dataset for Forensic Plagiarism Detection](<https://devfeed.tech/articles/plagiacorpus-a-bilingual-adversarial-and-cross-lingual-dataset-for-forensic-plagiarism-detection-115404.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.02940>)

Author: Adel Ammar, Serry Sibaee, Wadii Boulila

Published: 2026-11-03T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [managing large-scale text data in AI](<https://devfeed.tech/topics/managing-large-scale-text-data-in-ai.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>), [Phoronix Test Suite](<https://devfeed.tech/topics/phoronix-test-suite.md>)

Tags: [adversarial](<https://devfeed.tech/tags/adversarial.md>), [arabic](<https://devfeed.tech/tags/arabic.md>), [benchmarking](<https://devfeed.tech/tags/benchmarking.md>), [book](<https://devfeed.tech/tags/book.md>), [cc-by-sa](<https://devfeed.tech/tags/cc-by-sa.md>), [collections](<https://devfeed.tech/tags/collections.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-cy](<https://devfeed.tech/tags/cs-cy.md>), [cs-dl](<https://devfeed.tech/tags/cs-dl.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [detection](<https://devfeed.tech/tags/detection.md>), [english](<https://devfeed.tech/tags/english.md>), [forensic](<https://devfeed.tech/tags/forensic.md>), [formats](<https://devfeed.tech/tags/formats.md>), [json](<https://devfeed.tech/tags/json.md>), [language](<https://devfeed.tech/tags/language.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [large-language-models-llms](<https://devfeed.tech/tags/large-language-models-llms.md>), [license](<https://devfeed.tech/tags/license.md>), [llm](<https://devfeed.tech/tags/llm.md>), [parquet](<https://devfeed.tech/tags/parquet.md>), [plagiarism](<https://devfeed.tech/tags/plagiarism.md>)

### AI overview

PlagiaCorpus provides 198,643 Arabic-English records for benchmarking plagiarism and text-reuse detection, including LLM paraphrasing and cross-language reuse. Its 34,885 source passages come from Wikipedia and openly licensed books; 163,758 queries include training and test sets across copying, character-level obfuscation, same-language LLM paraphrasing, English-Arabic translation with structural paraphrasing, and non-reused text. Positive queries link to source passages, and source-grouped splitting keeps each passage's queries within one split. Curation included language identification, removal of LLM generation artefacts, and exact and cross-lingual deduplication. Reported checks include class-specific length statistics, split validation, and inter-annotator agreement on a manually rated sample. The dataset is available in JSON and Parquet under CC BY-SA 4.0.

### Source excerpt

arXiv:2611.02940v1 Announce Type: cross Abstract: PlagiaCorpus is a bilingual Arabic-English dataset for benchmarking plagiarism and text-reuse detection, including reuse produced by large language models (LLMs) and reuse across languages. It contains 198,643 text records: a source corpus of 34,885 passages drawn from English and Arabic Wikipedia and two openly licensed book collections, and 163,758 queries (147,381 for training and 16,377 for testing). The queries cover five reuse classes of increasing difficulty, namely verbatim copying, character-level obfuscation, same-language paraphrasing by two different LLMs, and English--Arabic translation with structural paraphrasing, together with a Safe class of non-reused passages. Every positive query is linked to its source passage, and the train/test split is grouped by source passage so that no passage contributes queries to both splits. The release was curated through language identification, removal of LLM generation artefacts, and exact and cross-lingual de-duplication. Technical validation reports per-class length statistics, automatic checks of the train/test split, and inter-annotator agreement on a manually rated sample. The dataset is distributed in JSON and Parquet formats under a CC BY-SA 4.0 license.

## From using AI to operating with AI: What changes in product management when the bottlenecks disappear

DevFeed: [From using AI to operating with AI: What changes in product management when the bottlenecks disappear](<https://devfeed.tech/articles/from-using-ai-to-operating-with-ai-what-changes-in-product-management-when-the-bottlenecks-disappear-106921.md>)

Original publisher: [Read original article](<https://building.nu.com/from-using-ai-to-operating-with-ai-what-changes-in-product-management-when-the-bottlenecks-disappear/>)

Author: Nubank Editorial

Published: 2026-10-29T16:30:40Z

Content type: article

Language: en

Sources: [Building Nubank](<https://devfeed.tech/sources/building-nubank.md>), [Nubank](<https://devfeed.tech/sources/nubank.md>)

Topics: [Product Management](<https://devfeed.tech/topics/product-management.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [human review](<https://devfeed.tech/topics/human-review.md>), [Tool](<https://devfeed.tech/topics/tool.md>), [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>), [product analytics](<https://devfeed.tech/topics/product-analytics.md>), [Pull Request](<https://devfeed.tech/topics/pull-request.md>), [Spec Driven Development](<https://devfeed.tech/topics/spec-driven-development.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>), [Google Calendar](<https://devfeed.tech/topics/google-calendar.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [ai-coding-assistant](<https://devfeed.tech/tags/ai-coding-assistant.md>), [analysis](<https://devfeed.tech/tags/analysis.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [bottlenecks](<https://devfeed.tech/tags/bottlenecks.md>), [chatgpt](<https://devfeed.tech/tags/chatgpt.md>), [claude-cowork](<https://devfeed.tech/tags/claude-cowork.md>), [product](<https://devfeed.tech/tags/product.md>), [product-management](<https://devfeed.tech/tags/product-management.md>)

### AI overview

Cinthia Scarlatti's examples show how product managers can use AI to reduce execution delays while keeping problem definition, validation and decisions in human hands. For a Mexico-specific customer support correction, she gave Cursor a Jira-style description of the problem, desired behavior and constraints; it proposed options, implemented the selected change and opened a pull request for engineering approval. Reported lead time fell from over a week to under a day. In another case, Cursor used Glean documentation to configure Databricks access, found suitable datasets and analyzed example chats in minutes, helping identify a training issue rather than a technical bug. Scarlatti stresses checking data sources, filters and results before acting. Her Claude Cowork Daily Digest gathers project questions, Slack threads, notes and meeting transcripts into daily summaries, updated documentation and a draft status post; she estimates it saves about an hour per day. The approach depends on clear requirements, relevant company context and review, and she presents the examples as adaptable rather than a universal recipe.

### Source excerpt

How a Lead Product Manager at Nubank turned problem definition into her main tool for working with AI agents, and why judgment stays with the PM The post From using AI to operating with AI: What changes in product management when the bottlenecks disappear appeared first on Building Nubank.

## NVIDIA Nemotron 3.5 Transcribe and Granary v2: Intelligent Multilingual Transcription You Can Prompt and Adapt

DevFeed: [NVIDIA Nemotron 3.5 Transcribe and Granary v2: Intelligent Multilingual Transcription You Can Prompt and Adapt](<https://devfeed.tech/articles/nvidia-nemotron-3-5-transcribe-and-granary-v2-intelligent-multilingual-transcription-you-can-prompt-and-adapt-105893.md>)

Original publisher: [Read original article](<https://huggingface.co/blog/nvidia/nemotron-transcribe-granary>)

Author: Maryam Motamedi; Piotr Żelasko; Kunal Dhawan; Nikolay Karpov; Nune Tadevosyan; Meline Mkrtchyan; Nithin Rao Koluguri; Francesco

Published: 2026-10-28T14:58:17Z

Content type: release

Language: en

Sources: [Hugging Face](<https://devfeed.tech/sources/hugging-face-blog.md>)

Topics: [Nemotron](<https://devfeed.tech/topics/nemotron.md>), [voice ai](<https://devfeed.tech/topics/voice-ai.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [Prompt Engineering](<https://devfeed.tech/topics/prompt-engineering.md>), [Enhancing AI model performance](<https://devfeed.tech/topics/enhancing-ai-model-performance.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [applications](<https://devfeed.tech/tags/applications.md>), [audio](<https://devfeed.tech/tags/audio.md>), [data](<https://devfeed.tech/tags/data.md>), [data-pipeline](<https://devfeed.tech/tags/data-pipeline.md>), [domain](<https://devfeed.tech/tags/domain.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [format](<https://devfeed.tech/tags/format.md>), [granary-v2](<https://devfeed.tech/tags/granary-v2.md>), [leaderboard](<https://devfeed.tech/tags/leaderboard.md>), [model](<https://devfeed.tech/tags/model.md>), [nemotron](<https://devfeed.tech/tags/nemotron.md>), [nemotron-3-5](<https://devfeed.tech/tags/nemotron-3-5.md>), [nemotron-3-5-transcribe](<https://devfeed.tech/tags/nemotron-3-5-transcribe.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-nemotron](<https://devfeed.tech/tags/nvidia-nemotron.md>), [open](<https://devfeed.tech/tags/open.md>), [pipeline](<https://devfeed.tech/tags/pipeline.md>), [prompt](<https://devfeed.tech/tags/prompt.md>), [transcription](<https://devfeed.tech/tags/transcription.md>), [v2](<https://devfeed.tech/tags/v2.md>)

### AI overview

NVIDIA Nemotron 3.5 Transcribe combines multilingual transcription, speaker attribution, translation, and audio Q&A in one promptable open-weight checkpoint. It supports transcription in 40 languages, translation between English and 31 languages, up to eight speakers, and recordings up to an hour; language support varies by task, and audio Q&A is trained in English. It is designed for recorded audio, with vLLM serving available and quantized local deployment described as forthcoming. Granary v2 provides multilingual manifests and a reusable curation pipeline, while NeMo Speech tools support fine-tuning. In an Indian medical ASR case study, fine-tuning reduced held-out WER from 22.44% to 12.17% versus the best-prompt baseline, with a slight LibriSpeech test-clean WER increase from 1.54% to 1.69%; a medical specialist still led on exact brand-name matches. The source recommends evaluating domain entities and general-speech retention.

### Source excerpt

# NVIDIA Nemotron 3.5 Transcribe and Granary v2: Intelligent Multilingual Transcription You Can Prompt and Adapt ## One open model for multilingual transcription, speaker attribution, translation, and audio Q&A, with top leaderboard results, plus the open Granary v2 data pipeline and fine-tuning recipes to adapt it. A transcript can capture every spoken word and still leave users or applications wondering who said what, how to translate it, or how to format domain-specific jargon.

## It costs a lot more to use AI models in some languages

DevFeed: [It costs a lot more to use AI models in some languages](<https://devfeed.tech/articles/it-costs-a-lot-more-to-use-ai-models-in-some-languages-105563.md>)

Original publisher: [Read original article](<https://www.computerworld.com/article/4241844/ai-models-are-much-more-expensive-for-less-common-languages.html>)

Author: Viktor Eriksson

Published: 2026-10-28T13:52:00Z

Content type: article

Language: en

Sources: [Computerworld](<https://devfeed.tech/sources/computerworld.md>)

Topics: [AI Models](<https://devfeed.tech/topics/ai-models.md>), [context window](<https://devfeed.tech/topics/context-window.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-models](<https://devfeed.tech/tags/ai-models.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [artificial-intelligence-generative-ai-productivity-software](<https://devfeed.tech/tags/artificial-intelligence-generative-ai-productivity-software.md>), [costs](<https://devfeed.tech/tags/costs.md>), [edition-us](<https://devfeed.tech/tags/edition-us.md>), [generative-ai](<https://devfeed.tech/tags/generative-ai.md>), [inequality](<https://devfeed.tech/tags/inequality.md>), [language-en](<https://devfeed.tech/tags/language-en.md>), [languages](<https://devfeed.tech/tags/languages.md>), [productivity](<https://devfeed.tech/tags/productivity.md>), [productivity-software](<https://devfeed.tech/tags/productivity-software.md>), [storytype-news-brief](<https://devfeed.tech/tags/storytype-news-brief.md>), [tokens](<https://devfeed.tech/tags/tokens.md>)

### AI overview

A Coalition for Language Equity in AI Research study finds that tokenizers can make the same prompt cost more to process in languages less represented in digitized data. Among 20 languages with the largest web presence, some used 1.5 to 5 times as many tokens as English; across 200 languages, one model showed a gap of up to 17 times. Because providers commonly charge by token, the extra tokens can raise costs, lengthen response times, and reduce how much fits in a model's context window. The researchers call this a "tokenizer tax" that can affect large populations, including Punjabi, Bengali, and Urdu speakers. Burmese, Tibetan, Khmer, and Lao are cited as languages whose writing systems are less common in training data; Latin-alphabet languages often have an advantage.

### Source excerpt

A new study from the Coalition for Language Equity in AI Research shows that AI models require significantly more tokens to process certain languages than others. These differences mean that users of underrepresented languages risk paying more for the same services, reports Rest of World. An LLM's tokenizer breaks text down into blocks, or tokens, but it does so much more optimally for languages with abundant digitized data, the researchers explain. They found that among the 20 languages with the largest web presence, less common languages used from 1.5 to 5 times as many tokens as the same prompt in English. When the study was expanded to 200 languages, the difference increased to as much as 17 times in one of the models examined. Since most LLM providers charge by tokens used, that translates to much higher costs. In addition, more tokens mean longer response times and a smaller amount of information that's able to fit into the AI models' context windows. The researchers describe this as a hidden "tokenizer tax" and warn that it exacerbates digital inequality. This disparity is not limited to languages with few speakers, the researchers note. "If we sort by speakers instead of web presence, the tax lands on enormous populations: Punjabi (~200 million speakers), Bengali (~266 million), Urdu (~288 million), Burmese (~36 million), Amharic (~36 million). This means hundreds of millions of people billed at multiples of the English rate for the same intelligence," the study says. Languages that use writing systems not as commonly found in the AI models' training data are particularly affected. Examples include Burmese, Tibetan, Khmer, and Lao. Languages that use the Latin alphabet, on the other hand, often have an advantage, even when spoken by relatively few people. Related: CIOs grapple with subpar global genAI models

## Agaram 2.5B: The First Tamil-First Language Model Trained from Scratch with Tool Calling. A Playbook for Under-Represented Languages

DevFeed: [Agaram 2.5B: The First Tamil-First Language Model Trained from Scratch with Tool Calling. A Playbook for Under-Represented Languages](<https://devfeed.tech/articles/agaram-2-5b-the-first-tamil-first-language-model-trained-from-scratch-with-tool-calling-a-playbook-for-under-represented-languages-103022.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.18420>)

Author: Ramchand Kumaresan, Madhusudhanan G, Kaushik Kezendirane

Published: 2026-10-28T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Large Language Model](<https://devfeed.tech/topics/llm.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>), [post-training](<https://devfeed.tech/topics/post-training.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [apache-2-0](<https://devfeed.tech/tags/apache-2-0.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [future](<https://devfeed.tech/tags/future.md>), [gguf](<https://devfeed.tech/tags/gguf.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [languages](<https://devfeed.tech/tags/languages.md>), [llm](<https://devfeed.tech/tags/llm.md>), [open-problems](<https://devfeed.tech/tags/open-problems.md>), [parameter](<https://devfeed.tech/tags/parameter.md>), [tool-calling](<https://devfeed.tech/tags/tool-calling.md>), [tools](<https://devfeed.tech/tags/tools.md>), [training](<https://devfeed.tech/tags/training.md>), [translation](<https://devfeed.tech/tags/translation.md>)

### AI overview

Agaram 2.5B is a Tamil-first model trained from scratch to handle Tamil script, Tanglish, English, and tool calls. Its 2.53B-parameter base was trained on 50 billion tokens, 58% Tamil; its tokenizer uses 1.55 tokens per Tamil word, compared with 9.44 for Qwen2.5. After a four-day public preview, the team rebuilt supervised fine-tuning using 222,662 records. On reported evaluations, Agaram scored above Sarvam-1 in all six measured translation directions and above Gemma 4 E4B-it and the authors' fine-tuned Gemma variant in both FLORES translation directions. It selected the expected tool on all 25 prompts in a set used for checkpoint selection and represented in training, but produced exact arguments on 16. An LLM judge preferred it to the preview model on 100 sealed human-written prompts. The authors identify factual knowledge and reliable refusal of harmful requests as open problems. Weights and GGUF files are released under Apache-2.0.

### Source excerpt

arXiv:2610.18420v2 Announce Type: replace-cross Abstract: We present Agaram 2.5B, a Tamil-first language model trained from scratch that converses in Tamil script, romanized Tanglish and English, and calls tools. Our first model, Agaram 1.5B, guided the new base, tokenizer and fine-tuning recipe: we pretrained a new 2.53B-parameter base on 50 billion tokens, 58% of them Tamil, with a tokenizer that needs 1.55 tokens per Tamil word where Qwen2.5's needs 9.44. A four-day public preview of our first 2.5B chat model showed where it fell short with real users, and we rebuilt supervised fine-tuning around what we saw, on 222,662 records. The released model scores above Sarvam-1 in all six translation directions we measured (FLORES chrF 56.9 into Tamil and 55.4 into English), chooses the expected tool on all 25 prompts of a tool set that was used during checkpoint selection and whose schemas appear in training, with exact arguments on 16, and was preferred over the preview model by an LLM judge on 100 sealed human-written prompts (net preference +30 points, 95% interval +14.5 to +45.0). On FLORES it also scores above Google's Gemma 4 E4B-it and our fine-tuned variant of it in both directions, with 2.53B parameters against E4B's 4.5B effective parameters. Factual knowledge and reliable refusal of harmful requests remain open problems, which we set out as future work. The base and chat weights and GGUF files are released under Apache-2.0, and this report gives a full account of the data, training and evaluation as a playbook for building language models for under-represented languages.

## Who Gets Flagged? The Pluralistic Evaluation Gap in AI Content Watermarking

DevFeed: [Who Gets Flagged? The Pluralistic Evaluation Gap in AI Content Watermarking](<https://devfeed.tech/articles/who-gets-flagged-the-pluralistic-evaluation-gap-in-ai-content-watermarking-103321.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2604.13776>)

Author: Alexander Nemecek, Osama Zafar, Yuqiao Xu, Wenbiao Li, Erman Ayday

Published: 2026-10-28T04:00:00Z

Content type: opinion

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [synthid](<https://devfeed.tech/topics/synthid.md>), [ai-governance](<https://devfeed.tech/topics/ai-governance.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>), [Localization (l10n)](<https://devfeed.tech/topics/localization.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-content](<https://devfeed.tech/tags/ai-content.md>), [ai-models](<https://devfeed.tech/tags/ai-models.md>), [alignment](<https://devfeed.tech/tags/alignment.md>), [audio](<https://devfeed.tech/tags/audio.md>), [auditing](<https://devfeed.tech/tags/auditing.md>), [authentication](<https://devfeed.tech/tags/authentication.md>), [benchmarking](<https://devfeed.tech/tags/benchmarking.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [bias](<https://devfeed.tech/tags/bias.md>), [content](<https://devfeed.tech/tags/content.md>), [coverage](<https://devfeed.tech/tags/coverage.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-cy](<https://devfeed.tech/tags/cs-cy.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [detection](<https://devfeed.tech/tags/detection.md>), [disaggregation](<https://devfeed.tech/tags/disaggregation.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [fairness](<https://devfeed.tech/tags/fairness.md>), [frameworks](<https://devfeed.tech/tags/frameworks.md>), [governance](<https://devfeed.tech/tags/governance.md>), [watermarking](<https://devfeed.tech/tags/watermarking.md>)

### AI overview

The authors report that, with one exception, major AI watermarking benchmarks across text, image, and audio do not measure performance across languages, cultural content types, or population groups. Because watermark signal strength, detectability, and robustness depend on content properties that vary across these groups, they propose evaluating cross-lingual detection parity, culturally diverse content coverage, and demographic breakdowns of detection metrics. They argue fairness evaluation should precede deployment and that bias audits applied to AI models should also cover watermark verification; the governance frameworks they discuss mandate watermarking without requiring such evaluation.

### Source excerpt

arXiv:2604.13776v3 Announce Type: replace-cross Abstract: Watermarking is becoming the default mechanism for AI content authentication, with governance policies and frameworks referencing it as infrastructure for content provenance. Yet across text, image, and audio modalities, watermark signal strength, detectability, and robustness depend on statistical properties of the content itself, properties that vary systematically across languages, cultural visual traditions, and demographic groups. We examine how this content dependence creates modality-specific pathways to bias. Reviewing the major watermarking benchmarks across modalities, we find that, with one exception, none report performance across languages, cultural content types, or population groups. To address this, we propose three concrete evaluation dimensions for pluralistic watermark benchmarking: cross-lingual detection parity, culturally diverse content coverage, and demographic disaggregation of detection metrics. We argue that watermarking is part of the pluralistic alignment pipeline and should be held to the same evaluation standards. We connect this to governance frameworks currently mandating watermarking deployment without requiring fairness evaluation. Our position is that evaluation must precede deployment, and that the same bias auditing requirements applied to AI models should extend to the verification layer.

## The Score Is Not the Structure: Brain Alignment and Cross-Lingual Transfer

DevFeed: [The Score Is Not the Structure: Brain Alignment and Cross-Lingual Transfer](<https://devfeed.tech/articles/the-score-is-not-the-structure-brain-alignment-and-cross-lingual-transfer-103018.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.03827>)

Author: Saman Rahbar

Published: 2026-10-28T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [ai alignment](<https://devfeed.tech/topics/ai-alignment.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>)

Tags: [alignment](<https://devfeed.tech/tags/alignment.md>), [brain](<https://devfeed.tech/tags/brain.md>), [cross](<https://devfeed.tech/tags/cross.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [language](<https://devfeed.tech/tags/language.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [q-bio-nc](<https://devfeed.tech/tags/q-bio-nc.md>), [score](<https://devfeed.tech/tags/score.md>), [structure](<https://devfeed.tech/tags/structure.md>)

### AI overview

The reported similarity scores remain high even when the correspondence they are meant to measure is removed, so the scores alone do not establish shared structure. In cross-lingual grammaticality transfer across 17 languages, transfer declines with language distance (r = -0.66), but probe accuracy declines too and is at chance in four languages. Excluding those languages halves explained variance and narrows the distance range, limiting conclusions about the cause. Treating 272 language pairs as independent yields p = 0.0006, while analysis with languages as the unit finds no significant steering effect (p = 0.155). In brain alignment, training raises model-fMRI similarity from 0.10 to 0.34 (reliability ceiling 0.54), yet targets with brain correspondence destroyed reach 0.31. The excerpt reports no distance-graded steering effect or syntactic benefit from alignment, and recommends checking how much a score survives without its intended correspondence.

### Source excerpt

arXiv:2610.03827v4 Announce Type: replace-cross Abstract: Similarity scores are often offered as evidence that a model shares structure with the brain or across languages. We ask what such a score reads when that structure is removed, or when the instrument measuring it does not work, in two settings. Across seventeen languages, a grammaticality probe transfers worse between more distant languages (r = -0.66). But the probe's own accuracy falls along the same axis and is at chance in four languages, four of the five most distant. Dropping them halves the explained variance, and because it also narrows the range of distances, the design cannot say how much of the gradient is the instrument. Counting the 272 language pairs as independent gives p = 0.0006 for a steering effect that is null when the seventeen languages are the unit (p = 0.155). In brain alignment, a training objective raises a language model's similarity to fMRI responses from 0.10 to 0.34 (reliability ceiling 0.54), yet targets with the brain correspondence destroyed still reach 0.31. A shuffled target scores about k/n against the real one, for rank k and n sentences, a baseline that can be computed before any model is trained. Both interventions work, yet we detect no distance-graded steering effect and no syntactic benefit from alignment. Before reading a correspondence score, measure how much of it survives without the correspondence.

[Next page](<https://devfeed.tech/topics/i18n.md?cursor=WyIyMDI2LTEwLTI4VDA0OjAwOjAwKzAwOjAwIiwgImNjMTE1OTM3LWJhZGEtNDIyNC05MDhmLTkwOGVkYjYyYTY5MCJd>)